https://kemlg.upc.edu Intelligent System Project Intelligent Data Science and Artificial Intelligence Research Center Intelligent System Project (Part I - INTRODUCTION) Miquel Sànchez-Marrè Intelligent Data Science and Artificial Intelligence Research Centre (IDEAI-UPC) Knowledge Engineering and Machine Learning Group (KEMLG-UPC) Computer Science Dept. Universitat Politècnica de Catalunya · BarcelonaTech [email protected]http://www.cs.upc.edu/~miquel Course 2021/2022
Transcript
httpskemlgupcedu
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Intelligent System Project
(Part I - INTRODUCTION)Miquel Sagravenchez-Marregrave
Intelligent Data Science and Artificial Intelligence Research Centre (IDEAI-UPC)Knowledge Engineering and Machine Learning Group (KEMLG-UPC)
Computer Science DeptUniversitat Politegravecnica de Catalunya middot BarcelonaTech
miquelcsupceduhttpwwwcsupcedu~miquel
Course 20212022
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PART 1 ndash INTRODUCTION
2copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
3
Contents (1)
Introduction Description of the aims of the course Description of the team works Information about the IS project timeline Deliverables of the IS project Examples of past ISP projects
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
4
Contents (2)
Problem Analysis Problem Feature Analysis InformationData Analysis Viability Analysis Economic Analysis Environmental and Sustainability Analysis
Definition of the Intelligent System project issues Definition of main goals of the IS project Definition of sub-goals Task Analysis
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
5
Contents (3)
Development of an Intelligent System Project DataInformation Extraction Data Mining amp Knowledge Acquisition Process KnowledgeOntological Analysis Planning and selection of
IntelligentStatisticalMathematical MethodsTechniques Construction of Models and implementation of Techniques Module Integration Validation of ModelsTechniques Comparison of
Techniques Proposed Solution
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
6
Contents (4)
Intelligent System Project Milestones MS1 Definition of the System MS2 Midterm Reporting MS3 Final System amp Final Software MS4 Public Defense of the Project
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
7
Contents (5)
Intelligent Methods and Models Review of main Intelligent Methods available
Software tools Review of main software tools available
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
INTRODUCTION
8copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Aims of the course
Development of an Intelligent System Project Design and construction of an Intelligent System to
solve a non-trivial problem Integration and application of different know ledge
acquired in previous Master courses for the solving of complex problems using Artificial Intelligence techniques
Writing and communication abilit ies of your technical and research work and achievements about Intelligent Systems both to a general audience and to a specialized audience
Consolidation of team working abilit ies 9copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Description of the Teamwork
Assign a role to each member Select a team leader Maintain a Project Library including
Documents Project Schedule (Ganttrsquos chart) Project Time sheet
Plan periodical teamwork discussions among members of the team
Put high efforts in the Knowledge Acquisition process Data-driven techniques Model-driven techniques
10copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Intelligent System Project Timeline (1)
Activity Week Team Formation W1 Searching for a non-trivial problem W2 Definition of the Project W3 MS1-D1 Definition of the Project Document W4
MS1-D1 is due on October 7th 2021 MS2-D2 Midterm Document W8-W9
MS2-D2 is due on November 18th 2021 MS3-D3 Final Document W16
MS3-D3 is due on January 19th 2022 MS3-D4 Software Delivery W16
MS3-D4 is due on January 19th 2022 MS4-D5 Public Presentation and Defence of the Project W16
MS4-D5 on January 20th 2022
11copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (1)
Definition of the Project Document (MS1-D1) Identification of a topic and a concrete problem Relationship to other similar works Specification of the goals of the project Outlining of a strategy to achieve the goals Preliminary Environmental amp Sustainability Analysis Preliminary Economic Analysis
12copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (2)
Midterm Document (MS2-D2) Specifications of the Problem to be solved Requirement Analysis of the System Initial Architecture of the System Proposed Solution Design
Initial Task Analysis Methods and alternatives for each task
Final Project Scheduling (Gantt diagram) Final Task Assignment
The Project Time Sheet Conclusions and Future Work
Final Software Delivery (MS3-D4) Final Software
Code Executable object
15copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (5)
Presentation slides for the Defense of the project (MS4-D5)
16copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects
17copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (1)20132014 Analysing and interpreting tweets related to weather talking
about the past or about the future making some sentimentmoodanalysis
Detecting talks about topics of interest related to some businessin the Linkedin Network (WhoTalk)
20142015 A prediction system for bike and spot availabilities (Bicing
predictor)20152016 A recommendation engine for movies (BAGmovies) An Image Search Engine based on Artistic Style PCC - Parrot Communication with children - An intelligent
interaction system for children with difficult emotion expressing skills through a flying drone with camera
18copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (2)
20162017 PartnerTIP Online dating system based on vkcom 20172018 Robust Euro Notes Classification (Adversarial Attack-Defense) Meeting the right people A Classification System for fictional stories20182019 Answering Machine (Question Answering) Finding Lost Pets A web platform that works as an animal
profiling service to help people find their lost pets as well as cataloguing stray animals
Cookit Track ingredients purchased and suggest recipes based on the list of items
19copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (3) Deduplication engine a software based solution to this problem
using a Support Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
20192020 Market Price Suggestion Multi-Modal Emotion Classifier Automatic Piano fingering through data-driven knowledge Gun Detection on images via MRCNN Sharesio (Secure and convenient photo sharing) An Intelligent System for Stock Market Prediction 20202021 Face Interpolator Midi to Tab Automatic generation of guitar tablatures Money Tracker
20copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Analysing and interpreting tweets related to weather talking aboutthe past or about the future making some sentimentmood analysis
21
Our objective is to create a system capable to analyze tweet data related to weather By analyzing we really mean that the system should be able to understand some of the information in the tweet
This information is clustered into three tasks
bull The first main task we want to be able to do is to detect when the information of a tweet happened is the tweet talking about the past the present or is it making a prediction
bull The second task is to understand if the tweet is positive negative or neutral This task is usually referred as sentiment or mood analysis
bull The third and last task is to infer what kind of weather is a tweet refer-ring to Is the tweet talking about the rain wind etc
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PART 1 ndash INTRODUCTION
2copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
3
Contents (1)
Introduction Description of the aims of the course Description of the team works Information about the IS project timeline Deliverables of the IS project Examples of past ISP projects
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
4
Contents (2)
Problem Analysis Problem Feature Analysis InformationData Analysis Viability Analysis Economic Analysis Environmental and Sustainability Analysis
Definition of the Intelligent System project issues Definition of main goals of the IS project Definition of sub-goals Task Analysis
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
5
Contents (3)
Development of an Intelligent System Project DataInformation Extraction Data Mining amp Knowledge Acquisition Process KnowledgeOntological Analysis Planning and selection of
IntelligentStatisticalMathematical MethodsTechniques Construction of Models and implementation of Techniques Module Integration Validation of ModelsTechniques Comparison of
Techniques Proposed Solution
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
6
Contents (4)
Intelligent System Project Milestones MS1 Definition of the System MS2 Midterm Reporting MS3 Final System amp Final Software MS4 Public Defense of the Project
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
7
Contents (5)
Intelligent Methods and Models Review of main Intelligent Methods available
Software tools Review of main software tools available
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
INTRODUCTION
8copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Aims of the course
Development of an Intelligent System Project Design and construction of an Intelligent System to
solve a non-trivial problem Integration and application of different know ledge
acquired in previous Master courses for the solving of complex problems using Artificial Intelligence techniques
Writing and communication abilit ies of your technical and research work and achievements about Intelligent Systems both to a general audience and to a specialized audience
Consolidation of team working abilit ies 9copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Description of the Teamwork
Assign a role to each member Select a team leader Maintain a Project Library including
Documents Project Schedule (Ganttrsquos chart) Project Time sheet
Plan periodical teamwork discussions among members of the team
Put high efforts in the Knowledge Acquisition process Data-driven techniques Model-driven techniques
10copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Intelligent System Project Timeline (1)
Activity Week Team Formation W1 Searching for a non-trivial problem W2 Definition of the Project W3 MS1-D1 Definition of the Project Document W4
MS1-D1 is due on October 7th 2021 MS2-D2 Midterm Document W8-W9
MS2-D2 is due on November 18th 2021 MS3-D3 Final Document W16
MS3-D3 is due on January 19th 2022 MS3-D4 Software Delivery W16
MS3-D4 is due on January 19th 2022 MS4-D5 Public Presentation and Defence of the Project W16
MS4-D5 on January 20th 2022
11copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (1)
Definition of the Project Document (MS1-D1) Identification of a topic and a concrete problem Relationship to other similar works Specification of the goals of the project Outlining of a strategy to achieve the goals Preliminary Environmental amp Sustainability Analysis Preliminary Economic Analysis
12copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (2)
Midterm Document (MS2-D2) Specifications of the Problem to be solved Requirement Analysis of the System Initial Architecture of the System Proposed Solution Design
Initial Task Analysis Methods and alternatives for each task
Final Project Scheduling (Gantt diagram) Final Task Assignment
The Project Time Sheet Conclusions and Future Work
Final Software Delivery (MS3-D4) Final Software
Code Executable object
15copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (5)
Presentation slides for the Defense of the project (MS4-D5)
16copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects
17copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (1)20132014 Analysing and interpreting tweets related to weather talking
about the past or about the future making some sentimentmoodanalysis
Detecting talks about topics of interest related to some businessin the Linkedin Network (WhoTalk)
20142015 A prediction system for bike and spot availabilities (Bicing
predictor)20152016 A recommendation engine for movies (BAGmovies) An Image Search Engine based on Artistic Style PCC - Parrot Communication with children - An intelligent
interaction system for children with difficult emotion expressing skills through a flying drone with camera
18copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (2)
20162017 PartnerTIP Online dating system based on vkcom 20172018 Robust Euro Notes Classification (Adversarial Attack-Defense) Meeting the right people A Classification System for fictional stories20182019 Answering Machine (Question Answering) Finding Lost Pets A web platform that works as an animal
profiling service to help people find their lost pets as well as cataloguing stray animals
Cookit Track ingredients purchased and suggest recipes based on the list of items
19copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (3) Deduplication engine a software based solution to this problem
using a Support Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
20192020 Market Price Suggestion Multi-Modal Emotion Classifier Automatic Piano fingering through data-driven knowledge Gun Detection on images via MRCNN Sharesio (Secure and convenient photo sharing) An Intelligent System for Stock Market Prediction 20202021 Face Interpolator Midi to Tab Automatic generation of guitar tablatures Money Tracker
20copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Analysing and interpreting tweets related to weather talking aboutthe past or about the future making some sentimentmood analysis
21
Our objective is to create a system capable to analyze tweet data related to weather By analyzing we really mean that the system should be able to understand some of the information in the tweet
This information is clustered into three tasks
bull The first main task we want to be able to do is to detect when the information of a tweet happened is the tweet talking about the past the present or is it making a prediction
bull The second task is to understand if the tweet is positive negative or neutral This task is usually referred as sentiment or mood analysis
bull The third and last task is to infer what kind of weather is a tweet refer-ring to Is the tweet talking about the rain wind etc
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
3
Contents (1)
Introduction Description of the aims of the course Description of the team works Information about the IS project timeline Deliverables of the IS project Examples of past ISP projects
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
4
Contents (2)
Problem Analysis Problem Feature Analysis InformationData Analysis Viability Analysis Economic Analysis Environmental and Sustainability Analysis
Definition of the Intelligent System project issues Definition of main goals of the IS project Definition of sub-goals Task Analysis
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
5
Contents (3)
Development of an Intelligent System Project DataInformation Extraction Data Mining amp Knowledge Acquisition Process KnowledgeOntological Analysis Planning and selection of
IntelligentStatisticalMathematical MethodsTechniques Construction of Models and implementation of Techniques Module Integration Validation of ModelsTechniques Comparison of
Techniques Proposed Solution
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
6
Contents (4)
Intelligent System Project Milestones MS1 Definition of the System MS2 Midterm Reporting MS3 Final System amp Final Software MS4 Public Defense of the Project
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
7
Contents (5)
Intelligent Methods and Models Review of main Intelligent Methods available
Software tools Review of main software tools available
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
INTRODUCTION
8copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Aims of the course
Development of an Intelligent System Project Design and construction of an Intelligent System to
solve a non-trivial problem Integration and application of different know ledge
acquired in previous Master courses for the solving of complex problems using Artificial Intelligence techniques
Writing and communication abilit ies of your technical and research work and achievements about Intelligent Systems both to a general audience and to a specialized audience
Consolidation of team working abilit ies 9copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Description of the Teamwork
Assign a role to each member Select a team leader Maintain a Project Library including
Documents Project Schedule (Ganttrsquos chart) Project Time sheet
Plan periodical teamwork discussions among members of the team
Put high efforts in the Knowledge Acquisition process Data-driven techniques Model-driven techniques
10copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Intelligent System Project Timeline (1)
Activity Week Team Formation W1 Searching for a non-trivial problem W2 Definition of the Project W3 MS1-D1 Definition of the Project Document W4
MS1-D1 is due on October 7th 2021 MS2-D2 Midterm Document W8-W9
MS2-D2 is due on November 18th 2021 MS3-D3 Final Document W16
MS3-D3 is due on January 19th 2022 MS3-D4 Software Delivery W16
MS3-D4 is due on January 19th 2022 MS4-D5 Public Presentation and Defence of the Project W16
MS4-D5 on January 20th 2022
11copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (1)
Definition of the Project Document (MS1-D1) Identification of a topic and a concrete problem Relationship to other similar works Specification of the goals of the project Outlining of a strategy to achieve the goals Preliminary Environmental amp Sustainability Analysis Preliminary Economic Analysis
12copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (2)
Midterm Document (MS2-D2) Specifications of the Problem to be solved Requirement Analysis of the System Initial Architecture of the System Proposed Solution Design
Initial Task Analysis Methods and alternatives for each task
Final Project Scheduling (Gantt diagram) Final Task Assignment
The Project Time Sheet Conclusions and Future Work
Final Software Delivery (MS3-D4) Final Software
Code Executable object
15copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (5)
Presentation slides for the Defense of the project (MS4-D5)
16copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects
17copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (1)20132014 Analysing and interpreting tweets related to weather talking
about the past or about the future making some sentimentmoodanalysis
Detecting talks about topics of interest related to some businessin the Linkedin Network (WhoTalk)
20142015 A prediction system for bike and spot availabilities (Bicing
predictor)20152016 A recommendation engine for movies (BAGmovies) An Image Search Engine based on Artistic Style PCC - Parrot Communication with children - An intelligent
interaction system for children with difficult emotion expressing skills through a flying drone with camera
18copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (2)
20162017 PartnerTIP Online dating system based on vkcom 20172018 Robust Euro Notes Classification (Adversarial Attack-Defense) Meeting the right people A Classification System for fictional stories20182019 Answering Machine (Question Answering) Finding Lost Pets A web platform that works as an animal
profiling service to help people find their lost pets as well as cataloguing stray animals
Cookit Track ingredients purchased and suggest recipes based on the list of items
19copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (3) Deduplication engine a software based solution to this problem
using a Support Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
20192020 Market Price Suggestion Multi-Modal Emotion Classifier Automatic Piano fingering through data-driven knowledge Gun Detection on images via MRCNN Sharesio (Secure and convenient photo sharing) An Intelligent System for Stock Market Prediction 20202021 Face Interpolator Midi to Tab Automatic generation of guitar tablatures Money Tracker
20copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Analysing and interpreting tweets related to weather talking aboutthe past or about the future making some sentimentmood analysis
21
Our objective is to create a system capable to analyze tweet data related to weather By analyzing we really mean that the system should be able to understand some of the information in the tweet
This information is clustered into three tasks
bull The first main task we want to be able to do is to detect when the information of a tweet happened is the tweet talking about the past the present or is it making a prediction
bull The second task is to understand if the tweet is positive negative or neutral This task is usually referred as sentiment or mood analysis
bull The third and last task is to infer what kind of weather is a tweet refer-ring to Is the tweet talking about the rain wind etc
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
4
Contents (2)
Problem Analysis Problem Feature Analysis InformationData Analysis Viability Analysis Economic Analysis Environmental and Sustainability Analysis
Definition of the Intelligent System project issues Definition of main goals of the IS project Definition of sub-goals Task Analysis
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
5
Contents (3)
Development of an Intelligent System Project DataInformation Extraction Data Mining amp Knowledge Acquisition Process KnowledgeOntological Analysis Planning and selection of
IntelligentStatisticalMathematical MethodsTechniques Construction of Models and implementation of Techniques Module Integration Validation of ModelsTechniques Comparison of
Techniques Proposed Solution
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
6
Contents (4)
Intelligent System Project Milestones MS1 Definition of the System MS2 Midterm Reporting MS3 Final System amp Final Software MS4 Public Defense of the Project
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
7
Contents (5)
Intelligent Methods and Models Review of main Intelligent Methods available
Software tools Review of main software tools available
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
INTRODUCTION
8copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Aims of the course
Development of an Intelligent System Project Design and construction of an Intelligent System to
solve a non-trivial problem Integration and application of different know ledge
acquired in previous Master courses for the solving of complex problems using Artificial Intelligence techniques
Writing and communication abilit ies of your technical and research work and achievements about Intelligent Systems both to a general audience and to a specialized audience
Consolidation of team working abilit ies 9copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Description of the Teamwork
Assign a role to each member Select a team leader Maintain a Project Library including
Documents Project Schedule (Ganttrsquos chart) Project Time sheet
Plan periodical teamwork discussions among members of the team
Put high efforts in the Knowledge Acquisition process Data-driven techniques Model-driven techniques
10copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Intelligent System Project Timeline (1)
Activity Week Team Formation W1 Searching for a non-trivial problem W2 Definition of the Project W3 MS1-D1 Definition of the Project Document W4
MS1-D1 is due on October 7th 2021 MS2-D2 Midterm Document W8-W9
MS2-D2 is due on November 18th 2021 MS3-D3 Final Document W16
MS3-D3 is due on January 19th 2022 MS3-D4 Software Delivery W16
MS3-D4 is due on January 19th 2022 MS4-D5 Public Presentation and Defence of the Project W16
MS4-D5 on January 20th 2022
11copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (1)
Definition of the Project Document (MS1-D1) Identification of a topic and a concrete problem Relationship to other similar works Specification of the goals of the project Outlining of a strategy to achieve the goals Preliminary Environmental amp Sustainability Analysis Preliminary Economic Analysis
12copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (2)
Midterm Document (MS2-D2) Specifications of the Problem to be solved Requirement Analysis of the System Initial Architecture of the System Proposed Solution Design
Initial Task Analysis Methods and alternatives for each task
Final Project Scheduling (Gantt diagram) Final Task Assignment
The Project Time Sheet Conclusions and Future Work
Final Software Delivery (MS3-D4) Final Software
Code Executable object
15copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (5)
Presentation slides for the Defense of the project (MS4-D5)
16copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects
17copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (1)20132014 Analysing and interpreting tweets related to weather talking
about the past or about the future making some sentimentmoodanalysis
Detecting talks about topics of interest related to some businessin the Linkedin Network (WhoTalk)
20142015 A prediction system for bike and spot availabilities (Bicing
predictor)20152016 A recommendation engine for movies (BAGmovies) An Image Search Engine based on Artistic Style PCC - Parrot Communication with children - An intelligent
interaction system for children with difficult emotion expressing skills through a flying drone with camera
18copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (2)
20162017 PartnerTIP Online dating system based on vkcom 20172018 Robust Euro Notes Classification (Adversarial Attack-Defense) Meeting the right people A Classification System for fictional stories20182019 Answering Machine (Question Answering) Finding Lost Pets A web platform that works as an animal
profiling service to help people find their lost pets as well as cataloguing stray animals
Cookit Track ingredients purchased and suggest recipes based on the list of items
19copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (3) Deduplication engine a software based solution to this problem
using a Support Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
20192020 Market Price Suggestion Multi-Modal Emotion Classifier Automatic Piano fingering through data-driven knowledge Gun Detection on images via MRCNN Sharesio (Secure and convenient photo sharing) An Intelligent System for Stock Market Prediction 20202021 Face Interpolator Midi to Tab Automatic generation of guitar tablatures Money Tracker
20copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Analysing and interpreting tweets related to weather talking aboutthe past or about the future making some sentimentmood analysis
21
Our objective is to create a system capable to analyze tweet data related to weather By analyzing we really mean that the system should be able to understand some of the information in the tweet
This information is clustered into three tasks
bull The first main task we want to be able to do is to detect when the information of a tweet happened is the tweet talking about the past the present or is it making a prediction
bull The second task is to understand if the tweet is positive negative or neutral This task is usually referred as sentiment or mood analysis
bull The third and last task is to infer what kind of weather is a tweet refer-ring to Is the tweet talking about the rain wind etc
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
5
Contents (3)
Development of an Intelligent System Project DataInformation Extraction Data Mining amp Knowledge Acquisition Process KnowledgeOntological Analysis Planning and selection of
IntelligentStatisticalMathematical MethodsTechniques Construction of Models and implementation of Techniques Module Integration Validation of ModelsTechniques Comparison of
Techniques Proposed Solution
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
6
Contents (4)
Intelligent System Project Milestones MS1 Definition of the System MS2 Midterm Reporting MS3 Final System amp Final Software MS4 Public Defense of the Project
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
7
Contents (5)
Intelligent Methods and Models Review of main Intelligent Methods available
Software tools Review of main software tools available
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
INTRODUCTION
8copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Aims of the course
Development of an Intelligent System Project Design and construction of an Intelligent System to
solve a non-trivial problem Integration and application of different know ledge
acquired in previous Master courses for the solving of complex problems using Artificial Intelligence techniques
Writing and communication abilit ies of your technical and research work and achievements about Intelligent Systems both to a general audience and to a specialized audience
Consolidation of team working abilit ies 9copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Description of the Teamwork
Assign a role to each member Select a team leader Maintain a Project Library including
Documents Project Schedule (Ganttrsquos chart) Project Time sheet
Plan periodical teamwork discussions among members of the team
Put high efforts in the Knowledge Acquisition process Data-driven techniques Model-driven techniques
10copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Intelligent System Project Timeline (1)
Activity Week Team Formation W1 Searching for a non-trivial problem W2 Definition of the Project W3 MS1-D1 Definition of the Project Document W4
MS1-D1 is due on October 7th 2021 MS2-D2 Midterm Document W8-W9
MS2-D2 is due on November 18th 2021 MS3-D3 Final Document W16
MS3-D3 is due on January 19th 2022 MS3-D4 Software Delivery W16
MS3-D4 is due on January 19th 2022 MS4-D5 Public Presentation and Defence of the Project W16
MS4-D5 on January 20th 2022
11copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (1)
Definition of the Project Document (MS1-D1) Identification of a topic and a concrete problem Relationship to other similar works Specification of the goals of the project Outlining of a strategy to achieve the goals Preliminary Environmental amp Sustainability Analysis Preliminary Economic Analysis
12copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (2)
Midterm Document (MS2-D2) Specifications of the Problem to be solved Requirement Analysis of the System Initial Architecture of the System Proposed Solution Design
Initial Task Analysis Methods and alternatives for each task
Final Project Scheduling (Gantt diagram) Final Task Assignment
The Project Time Sheet Conclusions and Future Work
Final Software Delivery (MS3-D4) Final Software
Code Executable object
15copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (5)
Presentation slides for the Defense of the project (MS4-D5)
16copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects
17copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (1)20132014 Analysing and interpreting tweets related to weather talking
about the past or about the future making some sentimentmoodanalysis
Detecting talks about topics of interest related to some businessin the Linkedin Network (WhoTalk)
20142015 A prediction system for bike and spot availabilities (Bicing
predictor)20152016 A recommendation engine for movies (BAGmovies) An Image Search Engine based on Artistic Style PCC - Parrot Communication with children - An intelligent
interaction system for children with difficult emotion expressing skills through a flying drone with camera
18copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (2)
20162017 PartnerTIP Online dating system based on vkcom 20172018 Robust Euro Notes Classification (Adversarial Attack-Defense) Meeting the right people A Classification System for fictional stories20182019 Answering Machine (Question Answering) Finding Lost Pets A web platform that works as an animal
profiling service to help people find their lost pets as well as cataloguing stray animals
Cookit Track ingredients purchased and suggest recipes based on the list of items
19copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (3) Deduplication engine a software based solution to this problem
using a Support Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
20192020 Market Price Suggestion Multi-Modal Emotion Classifier Automatic Piano fingering through data-driven knowledge Gun Detection on images via MRCNN Sharesio (Secure and convenient photo sharing) An Intelligent System for Stock Market Prediction 20202021 Face Interpolator Midi to Tab Automatic generation of guitar tablatures Money Tracker
20copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Analysing and interpreting tweets related to weather talking aboutthe past or about the future making some sentimentmood analysis
21
Our objective is to create a system capable to analyze tweet data related to weather By analyzing we really mean that the system should be able to understand some of the information in the tweet
This information is clustered into three tasks
bull The first main task we want to be able to do is to detect when the information of a tweet happened is the tweet talking about the past the present or is it making a prediction
bull The second task is to understand if the tweet is positive negative or neutral This task is usually referred as sentiment or mood analysis
bull The third and last task is to infer what kind of weather is a tweet refer-ring to Is the tweet talking about the rain wind etc
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
6
Contents (4)
Intelligent System Project Milestones MS1 Definition of the System MS2 Midterm Reporting MS3 Final System amp Final Software MS4 Public Defense of the Project
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
7
Contents (5)
Intelligent Methods and Models Review of main Intelligent Methods available
Software tools Review of main software tools available
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
INTRODUCTION
8copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Aims of the course
Development of an Intelligent System Project Design and construction of an Intelligent System to
solve a non-trivial problem Integration and application of different know ledge
acquired in previous Master courses for the solving of complex problems using Artificial Intelligence techniques
Writing and communication abilit ies of your technical and research work and achievements about Intelligent Systems both to a general audience and to a specialized audience
Consolidation of team working abilit ies 9copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Description of the Teamwork
Assign a role to each member Select a team leader Maintain a Project Library including
Documents Project Schedule (Ganttrsquos chart) Project Time sheet
Plan periodical teamwork discussions among members of the team
Put high efforts in the Knowledge Acquisition process Data-driven techniques Model-driven techniques
10copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Intelligent System Project Timeline (1)
Activity Week Team Formation W1 Searching for a non-trivial problem W2 Definition of the Project W3 MS1-D1 Definition of the Project Document W4
MS1-D1 is due on October 7th 2021 MS2-D2 Midterm Document W8-W9
MS2-D2 is due on November 18th 2021 MS3-D3 Final Document W16
MS3-D3 is due on January 19th 2022 MS3-D4 Software Delivery W16
MS3-D4 is due on January 19th 2022 MS4-D5 Public Presentation and Defence of the Project W16
MS4-D5 on January 20th 2022
11copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (1)
Definition of the Project Document (MS1-D1) Identification of a topic and a concrete problem Relationship to other similar works Specification of the goals of the project Outlining of a strategy to achieve the goals Preliminary Environmental amp Sustainability Analysis Preliminary Economic Analysis
12copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (2)
Midterm Document (MS2-D2) Specifications of the Problem to be solved Requirement Analysis of the System Initial Architecture of the System Proposed Solution Design
Initial Task Analysis Methods and alternatives for each task
Final Project Scheduling (Gantt diagram) Final Task Assignment
The Project Time Sheet Conclusions and Future Work
Final Software Delivery (MS3-D4) Final Software
Code Executable object
15copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (5)
Presentation slides for the Defense of the project (MS4-D5)
16copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects
17copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (1)20132014 Analysing and interpreting tweets related to weather talking
about the past or about the future making some sentimentmoodanalysis
Detecting talks about topics of interest related to some businessin the Linkedin Network (WhoTalk)
20142015 A prediction system for bike and spot availabilities (Bicing
predictor)20152016 A recommendation engine for movies (BAGmovies) An Image Search Engine based on Artistic Style PCC - Parrot Communication with children - An intelligent
interaction system for children with difficult emotion expressing skills through a flying drone with camera
18copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (2)
20162017 PartnerTIP Online dating system based on vkcom 20172018 Robust Euro Notes Classification (Adversarial Attack-Defense) Meeting the right people A Classification System for fictional stories20182019 Answering Machine (Question Answering) Finding Lost Pets A web platform that works as an animal
profiling service to help people find their lost pets as well as cataloguing stray animals
Cookit Track ingredients purchased and suggest recipes based on the list of items
19copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (3) Deduplication engine a software based solution to this problem
using a Support Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
20192020 Market Price Suggestion Multi-Modal Emotion Classifier Automatic Piano fingering through data-driven knowledge Gun Detection on images via MRCNN Sharesio (Secure and convenient photo sharing) An Intelligent System for Stock Market Prediction 20202021 Face Interpolator Midi to Tab Automatic generation of guitar tablatures Money Tracker
20copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Analysing and interpreting tweets related to weather talking aboutthe past or about the future making some sentimentmood analysis
21
Our objective is to create a system capable to analyze tweet data related to weather By analyzing we really mean that the system should be able to understand some of the information in the tweet
This information is clustered into three tasks
bull The first main task we want to be able to do is to detect when the information of a tweet happened is the tweet talking about the past the present or is it making a prediction
bull The second task is to understand if the tweet is positive negative or neutral This task is usually referred as sentiment or mood analysis
bull The third and last task is to infer what kind of weather is a tweet refer-ring to Is the tweet talking about the rain wind etc
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
7
Contents (5)
Intelligent Methods and Models Review of main Intelligent Methods available
Software tools Review of main software tools available
copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
INTRODUCTION
8copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Aims of the course
Development of an Intelligent System Project Design and construction of an Intelligent System to
solve a non-trivial problem Integration and application of different know ledge
acquired in previous Master courses for the solving of complex problems using Artificial Intelligence techniques
Writing and communication abilit ies of your technical and research work and achievements about Intelligent Systems both to a general audience and to a specialized audience
Consolidation of team working abilit ies 9copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Description of the Teamwork
Assign a role to each member Select a team leader Maintain a Project Library including
Documents Project Schedule (Ganttrsquos chart) Project Time sheet
Plan periodical teamwork discussions among members of the team
Put high efforts in the Knowledge Acquisition process Data-driven techniques Model-driven techniques
10copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Intelligent System Project Timeline (1)
Activity Week Team Formation W1 Searching for a non-trivial problem W2 Definition of the Project W3 MS1-D1 Definition of the Project Document W4
MS1-D1 is due on October 7th 2021 MS2-D2 Midterm Document W8-W9
MS2-D2 is due on November 18th 2021 MS3-D3 Final Document W16
MS3-D3 is due on January 19th 2022 MS3-D4 Software Delivery W16
MS3-D4 is due on January 19th 2022 MS4-D5 Public Presentation and Defence of the Project W16
MS4-D5 on January 20th 2022
11copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (1)
Definition of the Project Document (MS1-D1) Identification of a topic and a concrete problem Relationship to other similar works Specification of the goals of the project Outlining of a strategy to achieve the goals Preliminary Environmental amp Sustainability Analysis Preliminary Economic Analysis
12copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (2)
Midterm Document (MS2-D2) Specifications of the Problem to be solved Requirement Analysis of the System Initial Architecture of the System Proposed Solution Design
Initial Task Analysis Methods and alternatives for each task
Final Project Scheduling (Gantt diagram) Final Task Assignment
The Project Time Sheet Conclusions and Future Work
Final Software Delivery (MS3-D4) Final Software
Code Executable object
15copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (5)
Presentation slides for the Defense of the project (MS4-D5)
16copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects
17copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (1)20132014 Analysing and interpreting tweets related to weather talking
about the past or about the future making some sentimentmoodanalysis
Detecting talks about topics of interest related to some businessin the Linkedin Network (WhoTalk)
20142015 A prediction system for bike and spot availabilities (Bicing
predictor)20152016 A recommendation engine for movies (BAGmovies) An Image Search Engine based on Artistic Style PCC - Parrot Communication with children - An intelligent
interaction system for children with difficult emotion expressing skills through a flying drone with camera
18copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (2)
20162017 PartnerTIP Online dating system based on vkcom 20172018 Robust Euro Notes Classification (Adversarial Attack-Defense) Meeting the right people A Classification System for fictional stories20182019 Answering Machine (Question Answering) Finding Lost Pets A web platform that works as an animal
profiling service to help people find their lost pets as well as cataloguing stray animals
Cookit Track ingredients purchased and suggest recipes based on the list of items
19copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (3) Deduplication engine a software based solution to this problem
using a Support Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
20192020 Market Price Suggestion Multi-Modal Emotion Classifier Automatic Piano fingering through data-driven knowledge Gun Detection on images via MRCNN Sharesio (Secure and convenient photo sharing) An Intelligent System for Stock Market Prediction 20202021 Face Interpolator Midi to Tab Automatic generation of guitar tablatures Money Tracker
20copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Analysing and interpreting tweets related to weather talking aboutthe past or about the future making some sentimentmood analysis
21
Our objective is to create a system capable to analyze tweet data related to weather By analyzing we really mean that the system should be able to understand some of the information in the tweet
This information is clustered into three tasks
bull The first main task we want to be able to do is to detect when the information of a tweet happened is the tweet talking about the past the present or is it making a prediction
bull The second task is to understand if the tweet is positive negative or neutral This task is usually referred as sentiment or mood analysis
bull The third and last task is to infer what kind of weather is a tweet refer-ring to Is the tweet talking about the rain wind etc
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
INTRODUCTION
8copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Aims of the course
Development of an Intelligent System Project Design and construction of an Intelligent System to
solve a non-trivial problem Integration and application of different know ledge
acquired in previous Master courses for the solving of complex problems using Artificial Intelligence techniques
Writing and communication abilit ies of your technical and research work and achievements about Intelligent Systems both to a general audience and to a specialized audience
Consolidation of team working abilit ies 9copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Description of the Teamwork
Assign a role to each member Select a team leader Maintain a Project Library including
Documents Project Schedule (Ganttrsquos chart) Project Time sheet
Plan periodical teamwork discussions among members of the team
Put high efforts in the Knowledge Acquisition process Data-driven techniques Model-driven techniques
10copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Intelligent System Project Timeline (1)
Activity Week Team Formation W1 Searching for a non-trivial problem W2 Definition of the Project W3 MS1-D1 Definition of the Project Document W4
MS1-D1 is due on October 7th 2021 MS2-D2 Midterm Document W8-W9
MS2-D2 is due on November 18th 2021 MS3-D3 Final Document W16
MS3-D3 is due on January 19th 2022 MS3-D4 Software Delivery W16
MS3-D4 is due on January 19th 2022 MS4-D5 Public Presentation and Defence of the Project W16
MS4-D5 on January 20th 2022
11copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (1)
Definition of the Project Document (MS1-D1) Identification of a topic and a concrete problem Relationship to other similar works Specification of the goals of the project Outlining of a strategy to achieve the goals Preliminary Environmental amp Sustainability Analysis Preliminary Economic Analysis
12copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (2)
Midterm Document (MS2-D2) Specifications of the Problem to be solved Requirement Analysis of the System Initial Architecture of the System Proposed Solution Design
Initial Task Analysis Methods and alternatives for each task
Final Project Scheduling (Gantt diagram) Final Task Assignment
The Project Time Sheet Conclusions and Future Work
Final Software Delivery (MS3-D4) Final Software
Code Executable object
15copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (5)
Presentation slides for the Defense of the project (MS4-D5)
16copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects
17copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (1)20132014 Analysing and interpreting tweets related to weather talking
about the past or about the future making some sentimentmoodanalysis
Detecting talks about topics of interest related to some businessin the Linkedin Network (WhoTalk)
20142015 A prediction system for bike and spot availabilities (Bicing
predictor)20152016 A recommendation engine for movies (BAGmovies) An Image Search Engine based on Artistic Style PCC - Parrot Communication with children - An intelligent
interaction system for children with difficult emotion expressing skills through a flying drone with camera
18copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (2)
20162017 PartnerTIP Online dating system based on vkcom 20172018 Robust Euro Notes Classification (Adversarial Attack-Defense) Meeting the right people A Classification System for fictional stories20182019 Answering Machine (Question Answering) Finding Lost Pets A web platform that works as an animal
profiling service to help people find their lost pets as well as cataloguing stray animals
Cookit Track ingredients purchased and suggest recipes based on the list of items
19copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (3) Deduplication engine a software based solution to this problem
using a Support Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
20192020 Market Price Suggestion Multi-Modal Emotion Classifier Automatic Piano fingering through data-driven knowledge Gun Detection on images via MRCNN Sharesio (Secure and convenient photo sharing) An Intelligent System for Stock Market Prediction 20202021 Face Interpolator Midi to Tab Automatic generation of guitar tablatures Money Tracker
20copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Analysing and interpreting tweets related to weather talking aboutthe past or about the future making some sentimentmood analysis
21
Our objective is to create a system capable to analyze tweet data related to weather By analyzing we really mean that the system should be able to understand some of the information in the tweet
This information is clustered into three tasks
bull The first main task we want to be able to do is to detect when the information of a tweet happened is the tweet talking about the past the present or is it making a prediction
bull The second task is to understand if the tweet is positive negative or neutral This task is usually referred as sentiment or mood analysis
bull The third and last task is to infer what kind of weather is a tweet refer-ring to Is the tweet talking about the rain wind etc
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Aims of the course
Development of an Intelligent System Project Design and construction of an Intelligent System to
solve a non-trivial problem Integration and application of different know ledge
acquired in previous Master courses for the solving of complex problems using Artificial Intelligence techniques
Writing and communication abilit ies of your technical and research work and achievements about Intelligent Systems both to a general audience and to a specialized audience
Consolidation of team working abilit ies 9copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Description of the Teamwork
Assign a role to each member Select a team leader Maintain a Project Library including
Documents Project Schedule (Ganttrsquos chart) Project Time sheet
Plan periodical teamwork discussions among members of the team
Put high efforts in the Knowledge Acquisition process Data-driven techniques Model-driven techniques
10copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Intelligent System Project Timeline (1)
Activity Week Team Formation W1 Searching for a non-trivial problem W2 Definition of the Project W3 MS1-D1 Definition of the Project Document W4
MS1-D1 is due on October 7th 2021 MS2-D2 Midterm Document W8-W9
MS2-D2 is due on November 18th 2021 MS3-D3 Final Document W16
MS3-D3 is due on January 19th 2022 MS3-D4 Software Delivery W16
MS3-D4 is due on January 19th 2022 MS4-D5 Public Presentation and Defence of the Project W16
MS4-D5 on January 20th 2022
11copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (1)
Definition of the Project Document (MS1-D1) Identification of a topic and a concrete problem Relationship to other similar works Specification of the goals of the project Outlining of a strategy to achieve the goals Preliminary Environmental amp Sustainability Analysis Preliminary Economic Analysis
12copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (2)
Midterm Document (MS2-D2) Specifications of the Problem to be solved Requirement Analysis of the System Initial Architecture of the System Proposed Solution Design
Initial Task Analysis Methods and alternatives for each task
Final Project Scheduling (Gantt diagram) Final Task Assignment
The Project Time Sheet Conclusions and Future Work
Final Software Delivery (MS3-D4) Final Software
Code Executable object
15copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (5)
Presentation slides for the Defense of the project (MS4-D5)
16copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects
17copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (1)20132014 Analysing and interpreting tweets related to weather talking
about the past or about the future making some sentimentmoodanalysis
Detecting talks about topics of interest related to some businessin the Linkedin Network (WhoTalk)
20142015 A prediction system for bike and spot availabilities (Bicing
predictor)20152016 A recommendation engine for movies (BAGmovies) An Image Search Engine based on Artistic Style PCC - Parrot Communication with children - An intelligent
interaction system for children with difficult emotion expressing skills through a flying drone with camera
18copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (2)
20162017 PartnerTIP Online dating system based on vkcom 20172018 Robust Euro Notes Classification (Adversarial Attack-Defense) Meeting the right people A Classification System for fictional stories20182019 Answering Machine (Question Answering) Finding Lost Pets A web platform that works as an animal
profiling service to help people find their lost pets as well as cataloguing stray animals
Cookit Track ingredients purchased and suggest recipes based on the list of items
19copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (3) Deduplication engine a software based solution to this problem
using a Support Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
20192020 Market Price Suggestion Multi-Modal Emotion Classifier Automatic Piano fingering through data-driven knowledge Gun Detection on images via MRCNN Sharesio (Secure and convenient photo sharing) An Intelligent System for Stock Market Prediction 20202021 Face Interpolator Midi to Tab Automatic generation of guitar tablatures Money Tracker
20copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Analysing and interpreting tweets related to weather talking aboutthe past or about the future making some sentimentmood analysis
21
Our objective is to create a system capable to analyze tweet data related to weather By analyzing we really mean that the system should be able to understand some of the information in the tweet
This information is clustered into three tasks
bull The first main task we want to be able to do is to detect when the information of a tweet happened is the tweet talking about the past the present or is it making a prediction
bull The second task is to understand if the tweet is positive negative or neutral This task is usually referred as sentiment or mood analysis
bull The third and last task is to infer what kind of weather is a tweet refer-ring to Is the tweet talking about the rain wind etc
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Description of the Teamwork
Assign a role to each member Select a team leader Maintain a Project Library including
Documents Project Schedule (Ganttrsquos chart) Project Time sheet
Plan periodical teamwork discussions among members of the team
Put high efforts in the Knowledge Acquisition process Data-driven techniques Model-driven techniques
10copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Intelligent System Project Timeline (1)
Activity Week Team Formation W1 Searching for a non-trivial problem W2 Definition of the Project W3 MS1-D1 Definition of the Project Document W4
MS1-D1 is due on October 7th 2021 MS2-D2 Midterm Document W8-W9
MS2-D2 is due on November 18th 2021 MS3-D3 Final Document W16
MS3-D3 is due on January 19th 2022 MS3-D4 Software Delivery W16
MS3-D4 is due on January 19th 2022 MS4-D5 Public Presentation and Defence of the Project W16
MS4-D5 on January 20th 2022
11copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (1)
Definition of the Project Document (MS1-D1) Identification of a topic and a concrete problem Relationship to other similar works Specification of the goals of the project Outlining of a strategy to achieve the goals Preliminary Environmental amp Sustainability Analysis Preliminary Economic Analysis
12copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (2)
Midterm Document (MS2-D2) Specifications of the Problem to be solved Requirement Analysis of the System Initial Architecture of the System Proposed Solution Design
Initial Task Analysis Methods and alternatives for each task
Final Project Scheduling (Gantt diagram) Final Task Assignment
The Project Time Sheet Conclusions and Future Work
Final Software Delivery (MS3-D4) Final Software
Code Executable object
15copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (5)
Presentation slides for the Defense of the project (MS4-D5)
16copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects
17copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (1)20132014 Analysing and interpreting tweets related to weather talking
about the past or about the future making some sentimentmoodanalysis
Detecting talks about topics of interest related to some businessin the Linkedin Network (WhoTalk)
20142015 A prediction system for bike and spot availabilities (Bicing
predictor)20152016 A recommendation engine for movies (BAGmovies) An Image Search Engine based on Artistic Style PCC - Parrot Communication with children - An intelligent
interaction system for children with difficult emotion expressing skills through a flying drone with camera
18copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (2)
20162017 PartnerTIP Online dating system based on vkcom 20172018 Robust Euro Notes Classification (Adversarial Attack-Defense) Meeting the right people A Classification System for fictional stories20182019 Answering Machine (Question Answering) Finding Lost Pets A web platform that works as an animal
profiling service to help people find their lost pets as well as cataloguing stray animals
Cookit Track ingredients purchased and suggest recipes based on the list of items
19copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (3) Deduplication engine a software based solution to this problem
using a Support Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
20192020 Market Price Suggestion Multi-Modal Emotion Classifier Automatic Piano fingering through data-driven knowledge Gun Detection on images via MRCNN Sharesio (Secure and convenient photo sharing) An Intelligent System for Stock Market Prediction 20202021 Face Interpolator Midi to Tab Automatic generation of guitar tablatures Money Tracker
20copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Analysing and interpreting tweets related to weather talking aboutthe past or about the future making some sentimentmood analysis
21
Our objective is to create a system capable to analyze tweet data related to weather By analyzing we really mean that the system should be able to understand some of the information in the tweet
This information is clustered into three tasks
bull The first main task we want to be able to do is to detect when the information of a tweet happened is the tweet talking about the past the present or is it making a prediction
bull The second task is to understand if the tweet is positive negative or neutral This task is usually referred as sentiment or mood analysis
bull The third and last task is to infer what kind of weather is a tweet refer-ring to Is the tweet talking about the rain wind etc
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Intelligent System Project Timeline (1)
Activity Week Team Formation W1 Searching for a non-trivial problem W2 Definition of the Project W3 MS1-D1 Definition of the Project Document W4
MS1-D1 is due on October 7th 2021 MS2-D2 Midterm Document W8-W9
MS2-D2 is due on November 18th 2021 MS3-D3 Final Document W16
MS3-D3 is due on January 19th 2022 MS3-D4 Software Delivery W16
MS3-D4 is due on January 19th 2022 MS4-D5 Public Presentation and Defence of the Project W16
MS4-D5 on January 20th 2022
11copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (1)
Definition of the Project Document (MS1-D1) Identification of a topic and a concrete problem Relationship to other similar works Specification of the goals of the project Outlining of a strategy to achieve the goals Preliminary Environmental amp Sustainability Analysis Preliminary Economic Analysis
12copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (2)
Midterm Document (MS2-D2) Specifications of the Problem to be solved Requirement Analysis of the System Initial Architecture of the System Proposed Solution Design
Initial Task Analysis Methods and alternatives for each task
Final Project Scheduling (Gantt diagram) Final Task Assignment
The Project Time Sheet Conclusions and Future Work
Final Software Delivery (MS3-D4) Final Software
Code Executable object
15copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (5)
Presentation slides for the Defense of the project (MS4-D5)
16copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects
17copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (1)20132014 Analysing and interpreting tweets related to weather talking
about the past or about the future making some sentimentmoodanalysis
Detecting talks about topics of interest related to some businessin the Linkedin Network (WhoTalk)
20142015 A prediction system for bike and spot availabilities (Bicing
predictor)20152016 A recommendation engine for movies (BAGmovies) An Image Search Engine based on Artistic Style PCC - Parrot Communication with children - An intelligent
interaction system for children with difficult emotion expressing skills through a flying drone with camera
18copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (2)
20162017 PartnerTIP Online dating system based on vkcom 20172018 Robust Euro Notes Classification (Adversarial Attack-Defense) Meeting the right people A Classification System for fictional stories20182019 Answering Machine (Question Answering) Finding Lost Pets A web platform that works as an animal
profiling service to help people find their lost pets as well as cataloguing stray animals
Cookit Track ingredients purchased and suggest recipes based on the list of items
19copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (3) Deduplication engine a software based solution to this problem
using a Support Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
20192020 Market Price Suggestion Multi-Modal Emotion Classifier Automatic Piano fingering through data-driven knowledge Gun Detection on images via MRCNN Sharesio (Secure and convenient photo sharing) An Intelligent System for Stock Market Prediction 20202021 Face Interpolator Midi to Tab Automatic generation of guitar tablatures Money Tracker
20copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Analysing and interpreting tweets related to weather talking aboutthe past or about the future making some sentimentmood analysis
21
Our objective is to create a system capable to analyze tweet data related to weather By analyzing we really mean that the system should be able to understand some of the information in the tweet
This information is clustered into three tasks
bull The first main task we want to be able to do is to detect when the information of a tweet happened is the tweet talking about the past the present or is it making a prediction
bull The second task is to understand if the tweet is positive negative or neutral This task is usually referred as sentiment or mood analysis
bull The third and last task is to infer what kind of weather is a tweet refer-ring to Is the tweet talking about the rain wind etc
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (1)
Definition of the Project Document (MS1-D1) Identification of a topic and a concrete problem Relationship to other similar works Specification of the goals of the project Outlining of a strategy to achieve the goals Preliminary Environmental amp Sustainability Analysis Preliminary Economic Analysis
12copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (2)
Midterm Document (MS2-D2) Specifications of the Problem to be solved Requirement Analysis of the System Initial Architecture of the System Proposed Solution Design
Initial Task Analysis Methods and alternatives for each task
Final Project Scheduling (Gantt diagram) Final Task Assignment
The Project Time Sheet Conclusions and Future Work
Final Software Delivery (MS3-D4) Final Software
Code Executable object
15copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (5)
Presentation slides for the Defense of the project (MS4-D5)
16copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects
17copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (1)20132014 Analysing and interpreting tweets related to weather talking
about the past or about the future making some sentimentmoodanalysis
Detecting talks about topics of interest related to some businessin the Linkedin Network (WhoTalk)
20142015 A prediction system for bike and spot availabilities (Bicing
predictor)20152016 A recommendation engine for movies (BAGmovies) An Image Search Engine based on Artistic Style PCC - Parrot Communication with children - An intelligent
interaction system for children with difficult emotion expressing skills through a flying drone with camera
18copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (2)
20162017 PartnerTIP Online dating system based on vkcom 20172018 Robust Euro Notes Classification (Adversarial Attack-Defense) Meeting the right people A Classification System for fictional stories20182019 Answering Machine (Question Answering) Finding Lost Pets A web platform that works as an animal
profiling service to help people find their lost pets as well as cataloguing stray animals
Cookit Track ingredients purchased and suggest recipes based on the list of items
19copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (3) Deduplication engine a software based solution to this problem
using a Support Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
20192020 Market Price Suggestion Multi-Modal Emotion Classifier Automatic Piano fingering through data-driven knowledge Gun Detection on images via MRCNN Sharesio (Secure and convenient photo sharing) An Intelligent System for Stock Market Prediction 20202021 Face Interpolator Midi to Tab Automatic generation of guitar tablatures Money Tracker
20copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Analysing and interpreting tweets related to weather talking aboutthe past or about the future making some sentimentmood analysis
21
Our objective is to create a system capable to analyze tweet data related to weather By analyzing we really mean that the system should be able to understand some of the information in the tweet
This information is clustered into three tasks
bull The first main task we want to be able to do is to detect when the information of a tweet happened is the tweet talking about the past the present or is it making a prediction
bull The second task is to understand if the tweet is positive negative or neutral This task is usually referred as sentiment or mood analysis
bull The third and last task is to infer what kind of weather is a tweet refer-ring to Is the tweet talking about the rain wind etc
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (2)
Midterm Document (MS2-D2) Specifications of the Problem to be solved Requirement Analysis of the System Initial Architecture of the System Proposed Solution Design
Initial Task Analysis Methods and alternatives for each task
Final Project Scheduling (Gantt diagram) Final Task Assignment
The Project Time Sheet Conclusions and Future Work
Final Software Delivery (MS3-D4) Final Software
Code Executable object
15copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (5)
Presentation slides for the Defense of the project (MS4-D5)
16copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects
17copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (1)20132014 Analysing and interpreting tweets related to weather talking
about the past or about the future making some sentimentmoodanalysis
Detecting talks about topics of interest related to some businessin the Linkedin Network (WhoTalk)
20142015 A prediction system for bike and spot availabilities (Bicing
predictor)20152016 A recommendation engine for movies (BAGmovies) An Image Search Engine based on Artistic Style PCC - Parrot Communication with children - An intelligent
interaction system for children with difficult emotion expressing skills through a flying drone with camera
18copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (2)
20162017 PartnerTIP Online dating system based on vkcom 20172018 Robust Euro Notes Classification (Adversarial Attack-Defense) Meeting the right people A Classification System for fictional stories20182019 Answering Machine (Question Answering) Finding Lost Pets A web platform that works as an animal
profiling service to help people find their lost pets as well as cataloguing stray animals
Cookit Track ingredients purchased and suggest recipes based on the list of items
19copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (3) Deduplication engine a software based solution to this problem
using a Support Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
20192020 Market Price Suggestion Multi-Modal Emotion Classifier Automatic Piano fingering through data-driven knowledge Gun Detection on images via MRCNN Sharesio (Secure and convenient photo sharing) An Intelligent System for Stock Market Prediction 20202021 Face Interpolator Midi to Tab Automatic generation of guitar tablatures Money Tracker
20copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Analysing and interpreting tweets related to weather talking aboutthe past or about the future making some sentimentmood analysis
21
Our objective is to create a system capable to analyze tweet data related to weather By analyzing we really mean that the system should be able to understand some of the information in the tweet
This information is clustered into three tasks
bull The first main task we want to be able to do is to detect when the information of a tweet happened is the tweet talking about the past the present or is it making a prediction
bull The second task is to understand if the tweet is positive negative or neutral This task is usually referred as sentiment or mood analysis
bull The third and last task is to infer what kind of weather is a tweet refer-ring to Is the tweet talking about the rain wind etc
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (3)
Final Document (MS3-D3) Executive Summary Userrsquos Manual
Description of the Systemrsquos purpose Start-upShutdown of the system Examples of use Interactions of the system
Input output List of Error messages
TechnicalSystem Manual Architecture of the System Solution Design
Task Analysis Methods implementing each task
14copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Final Project Scheduling (Gantt diagram) Final Task Assignment
The Project Time Sheet Conclusions and Future Work
Final Software Delivery (MS3-D4) Final Software
Code Executable object
15copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (5)
Presentation slides for the Defense of the project (MS4-D5)
16copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects
17copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (1)20132014 Analysing and interpreting tweets related to weather talking
about the past or about the future making some sentimentmoodanalysis
Detecting talks about topics of interest related to some businessin the Linkedin Network (WhoTalk)
20142015 A prediction system for bike and spot availabilities (Bicing
predictor)20152016 A recommendation engine for movies (BAGmovies) An Image Search Engine based on Artistic Style PCC - Parrot Communication with children - An intelligent
interaction system for children with difficult emotion expressing skills through a flying drone with camera
18copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (2)
20162017 PartnerTIP Online dating system based on vkcom 20172018 Robust Euro Notes Classification (Adversarial Attack-Defense) Meeting the right people A Classification System for fictional stories20182019 Answering Machine (Question Answering) Finding Lost Pets A web platform that works as an animal
profiling service to help people find their lost pets as well as cataloguing stray animals
Cookit Track ingredients purchased and suggest recipes based on the list of items
19copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (3) Deduplication engine a software based solution to this problem
using a Support Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
20192020 Market Price Suggestion Multi-Modal Emotion Classifier Automatic Piano fingering through data-driven knowledge Gun Detection on images via MRCNN Sharesio (Secure and convenient photo sharing) An Intelligent System for Stock Market Prediction 20202021 Face Interpolator Midi to Tab Automatic generation of guitar tablatures Money Tracker
20copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Analysing and interpreting tweets related to weather talking aboutthe past or about the future making some sentimentmood analysis
21
Our objective is to create a system capable to analyze tweet data related to weather By analyzing we really mean that the system should be able to understand some of the information in the tweet
This information is clustered into three tasks
bull The first main task we want to be able to do is to detect when the information of a tweet happened is the tweet talking about the past the present or is it making a prediction
bull The second task is to understand if the tweet is positive negative or neutral This task is usually referred as sentiment or mood analysis
bull The third and last task is to infer what kind of weather is a tweet refer-ring to Is the tweet talking about the rain wind etc
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Final Project Scheduling (Gantt diagram) Final Task Assignment
The Project Time Sheet Conclusions and Future Work
Final Software Delivery (MS3-D4) Final Software
Code Executable object
15copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (5)
Presentation slides for the Defense of the project (MS4-D5)
16copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects
17copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (1)20132014 Analysing and interpreting tweets related to weather talking
about the past or about the future making some sentimentmoodanalysis
Detecting talks about topics of interest related to some businessin the Linkedin Network (WhoTalk)
20142015 A prediction system for bike and spot availabilities (Bicing
predictor)20152016 A recommendation engine for movies (BAGmovies) An Image Search Engine based on Artistic Style PCC - Parrot Communication with children - An intelligent
interaction system for children with difficult emotion expressing skills through a flying drone with camera
18copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (2)
20162017 PartnerTIP Online dating system based on vkcom 20172018 Robust Euro Notes Classification (Adversarial Attack-Defense) Meeting the right people A Classification System for fictional stories20182019 Answering Machine (Question Answering) Finding Lost Pets A web platform that works as an animal
profiling service to help people find their lost pets as well as cataloguing stray animals
Cookit Track ingredients purchased and suggest recipes based on the list of items
19copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (3) Deduplication engine a software based solution to this problem
using a Support Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
20192020 Market Price Suggestion Multi-Modal Emotion Classifier Automatic Piano fingering through data-driven knowledge Gun Detection on images via MRCNN Sharesio (Secure and convenient photo sharing) An Intelligent System for Stock Market Prediction 20202021 Face Interpolator Midi to Tab Automatic generation of guitar tablatures Money Tracker
20copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Analysing and interpreting tweets related to weather talking aboutthe past or about the future making some sentimentmood analysis
21
Our objective is to create a system capable to analyze tweet data related to weather By analyzing we really mean that the system should be able to understand some of the information in the tweet
This information is clustered into three tasks
bull The first main task we want to be able to do is to detect when the information of a tweet happened is the tweet talking about the past the present or is it making a prediction
bull The second task is to understand if the tweet is positive negative or neutral This task is usually referred as sentiment or mood analysis
bull The third and last task is to infer what kind of weather is a tweet refer-ring to Is the tweet talking about the rain wind etc
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deliverables of the IS Project (5)
Presentation slides for the Defense of the project (MS4-D5)
16copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects
17copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (1)20132014 Analysing and interpreting tweets related to weather talking
about the past or about the future making some sentimentmoodanalysis
Detecting talks about topics of interest related to some businessin the Linkedin Network (WhoTalk)
20142015 A prediction system for bike and spot availabilities (Bicing
predictor)20152016 A recommendation engine for movies (BAGmovies) An Image Search Engine based on Artistic Style PCC - Parrot Communication with children - An intelligent
interaction system for children with difficult emotion expressing skills through a flying drone with camera
18copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (2)
20162017 PartnerTIP Online dating system based on vkcom 20172018 Robust Euro Notes Classification (Adversarial Attack-Defense) Meeting the right people A Classification System for fictional stories20182019 Answering Machine (Question Answering) Finding Lost Pets A web platform that works as an animal
profiling service to help people find their lost pets as well as cataloguing stray animals
Cookit Track ingredients purchased and suggest recipes based on the list of items
19copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (3) Deduplication engine a software based solution to this problem
using a Support Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
20192020 Market Price Suggestion Multi-Modal Emotion Classifier Automatic Piano fingering through data-driven knowledge Gun Detection on images via MRCNN Sharesio (Secure and convenient photo sharing) An Intelligent System for Stock Market Prediction 20202021 Face Interpolator Midi to Tab Automatic generation of guitar tablatures Money Tracker
20copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Analysing and interpreting tweets related to weather talking aboutthe past or about the future making some sentimentmood analysis
21
Our objective is to create a system capable to analyze tweet data related to weather By analyzing we really mean that the system should be able to understand some of the information in the tweet
This information is clustered into three tasks
bull The first main task we want to be able to do is to detect when the information of a tweet happened is the tweet talking about the past the present or is it making a prediction
bull The second task is to understand if the tweet is positive negative or neutral This task is usually referred as sentiment or mood analysis
bull The third and last task is to infer what kind of weather is a tweet refer-ring to Is the tweet talking about the rain wind etc
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects
17copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (1)20132014 Analysing and interpreting tweets related to weather talking
about the past or about the future making some sentimentmoodanalysis
Detecting talks about topics of interest related to some businessin the Linkedin Network (WhoTalk)
20142015 A prediction system for bike and spot availabilities (Bicing
predictor)20152016 A recommendation engine for movies (BAGmovies) An Image Search Engine based on Artistic Style PCC - Parrot Communication with children - An intelligent
interaction system for children with difficult emotion expressing skills through a flying drone with camera
18copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (2)
20162017 PartnerTIP Online dating system based on vkcom 20172018 Robust Euro Notes Classification (Adversarial Attack-Defense) Meeting the right people A Classification System for fictional stories20182019 Answering Machine (Question Answering) Finding Lost Pets A web platform that works as an animal
profiling service to help people find their lost pets as well as cataloguing stray animals
Cookit Track ingredients purchased and suggest recipes based on the list of items
19copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (3) Deduplication engine a software based solution to this problem
using a Support Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
20192020 Market Price Suggestion Multi-Modal Emotion Classifier Automatic Piano fingering through data-driven knowledge Gun Detection on images via MRCNN Sharesio (Secure and convenient photo sharing) An Intelligent System for Stock Market Prediction 20202021 Face Interpolator Midi to Tab Automatic generation of guitar tablatures Money Tracker
20copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Analysing and interpreting tweets related to weather talking aboutthe past or about the future making some sentimentmood analysis
21
Our objective is to create a system capable to analyze tweet data related to weather By analyzing we really mean that the system should be able to understand some of the information in the tweet
This information is clustered into three tasks
bull The first main task we want to be able to do is to detect when the information of a tweet happened is the tweet talking about the past the present or is it making a prediction
bull The second task is to understand if the tweet is positive negative or neutral This task is usually referred as sentiment or mood analysis
bull The third and last task is to infer what kind of weather is a tweet refer-ring to Is the tweet talking about the rain wind etc
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (1)20132014 Analysing and interpreting tweets related to weather talking
about the past or about the future making some sentimentmoodanalysis
Detecting talks about topics of interest related to some businessin the Linkedin Network (WhoTalk)
20142015 A prediction system for bike and spot availabilities (Bicing
predictor)20152016 A recommendation engine for movies (BAGmovies) An Image Search Engine based on Artistic Style PCC - Parrot Communication with children - An intelligent
interaction system for children with difficult emotion expressing skills through a flying drone with camera
18copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (2)
20162017 PartnerTIP Online dating system based on vkcom 20172018 Robust Euro Notes Classification (Adversarial Attack-Defense) Meeting the right people A Classification System for fictional stories20182019 Answering Machine (Question Answering) Finding Lost Pets A web platform that works as an animal
profiling service to help people find their lost pets as well as cataloguing stray animals
Cookit Track ingredients purchased and suggest recipes based on the list of items
19copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (3) Deduplication engine a software based solution to this problem
using a Support Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
20192020 Market Price Suggestion Multi-Modal Emotion Classifier Automatic Piano fingering through data-driven knowledge Gun Detection on images via MRCNN Sharesio (Secure and convenient photo sharing) An Intelligent System for Stock Market Prediction 20202021 Face Interpolator Midi to Tab Automatic generation of guitar tablatures Money Tracker
20copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Analysing and interpreting tweets related to weather talking aboutthe past or about the future making some sentimentmood analysis
21
Our objective is to create a system capable to analyze tweet data related to weather By analyzing we really mean that the system should be able to understand some of the information in the tweet
This information is clustered into three tasks
bull The first main task we want to be able to do is to detect when the information of a tweet happened is the tweet talking about the past the present or is it making a prediction
bull The second task is to understand if the tweet is positive negative or neutral This task is usually referred as sentiment or mood analysis
bull The third and last task is to infer what kind of weather is a tweet refer-ring to Is the tweet talking about the rain wind etc
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (2)
20162017 PartnerTIP Online dating system based on vkcom 20172018 Robust Euro Notes Classification (Adversarial Attack-Defense) Meeting the right people A Classification System for fictional stories20182019 Answering Machine (Question Answering) Finding Lost Pets A web platform that works as an animal
profiling service to help people find their lost pets as well as cataloguing stray animals
Cookit Track ingredients purchased and suggest recipes based on the list of items
19copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (3) Deduplication engine a software based solution to this problem
using a Support Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
20192020 Market Price Suggestion Multi-Modal Emotion Classifier Automatic Piano fingering through data-driven knowledge Gun Detection on images via MRCNN Sharesio (Secure and convenient photo sharing) An Intelligent System for Stock Market Prediction 20202021 Face Interpolator Midi to Tab Automatic generation of guitar tablatures Money Tracker
20copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Analysing and interpreting tweets related to weather talking aboutthe past or about the future making some sentimentmood analysis
21
Our objective is to create a system capable to analyze tweet data related to weather By analyzing we really mean that the system should be able to understand some of the information in the tweet
This information is clustered into three tasks
bull The first main task we want to be able to do is to detect when the information of a tweet happened is the tweet talking about the past the present or is it making a prediction
bull The second task is to understand if the tweet is positive negative or neutral This task is usually referred as sentiment or mood analysis
bull The third and last task is to infer what kind of weather is a tweet refer-ring to Is the tweet talking about the rain wind etc
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Examples of Past ISP Projects (3) Deduplication engine a software based solution to this problem
using a Support Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
20192020 Market Price Suggestion Multi-Modal Emotion Classifier Automatic Piano fingering through data-driven knowledge Gun Detection on images via MRCNN Sharesio (Secure and convenient photo sharing) An Intelligent System for Stock Market Prediction 20202021 Face Interpolator Midi to Tab Automatic generation of guitar tablatures Money Tracker
20copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Analysing and interpreting tweets related to weather talking aboutthe past or about the future making some sentimentmood analysis
21
Our objective is to create a system capable to analyze tweet data related to weather By analyzing we really mean that the system should be able to understand some of the information in the tweet
This information is clustered into three tasks
bull The first main task we want to be able to do is to detect when the information of a tweet happened is the tweet talking about the past the present or is it making a prediction
bull The second task is to understand if the tweet is positive negative or neutral This task is usually referred as sentiment or mood analysis
bull The third and last task is to infer what kind of weather is a tweet refer-ring to Is the tweet talking about the rain wind etc
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Analysing and interpreting tweets related to weather talking aboutthe past or about the future making some sentimentmood analysis
21
Our objective is to create a system capable to analyze tweet data related to weather By analyzing we really mean that the system should be able to understand some of the information in the tweet
This information is clustered into three tasks
bull The first main task we want to be able to do is to detect when the information of a tweet happened is the tweet talking about the past the present or is it making a prediction
bull The second task is to understand if the tweet is positive negative or neutral This task is usually referred as sentiment or mood analysis
bull The third and last task is to infer what kind of weather is a tweet refer-ring to Is the tweet talking about the rain wind etc
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (1)
22
Interview with the client (user)Taking requirementsDefinition of scope and risksAccess to customer data (Books Keywords LinkedIn profile)
Books analysisPreprocessing of the booksImplement frequency algorithmsFrequency analysis of keywords (keywords Client)Implement the features extraction algorithm (keywords)Extraction of featuresFrequency analysis of the extracted keywordsMake Collecting of keywords frequency
Database creationDefinition of structureCreate databasePopulate Database
Creation and Analysis of classification algorithmsStudy the proposed classification algorithm (antispam methods)Algorithm implementationTraining the AlgorithmTesting the ImplementationTesting the Algorithm
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Detecting talks about topics of interest related to somebusiness in the Linkedin Network (WhoTalk) (2)
23
Creation of Service RESTful (Server BackEnd)Install ServerCreating restful structuresCreating the Rest serviceImplementation of the service in serverIntegration with LinkedIn ApiImplementation of Algorithm for periodic reading the linkedIn comments of users contacts
Creation of FrontEndLayout of the FrontCreation of MobileFirst AppCreation of Restful front structureIntegration with LinkedIn ApiIntegration with Restful Server (app server)
Management and monitoringPrepare documentationTrack project stagesContinue Communication with client
copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A prediction system for bike and spot availabilities (Bicing predictor)
24copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
25copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
26copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A recommendation engine for movies (BAGmovies)
27copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
28
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
29copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Image Search Engine based on Artistic Style
30
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
31copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
32copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (1) The Problem
It is not common for the applications based on the social networks to use the combined information from several accounts to provide an analysis of relation with the other members of the platform
The most popular social network are able to suggest to the possible friends using the information about the friends of friends
In the dating recommendation problem the information about user preferences is not contained in the relations itself with the exception of relationship status
The dating recommendation application based on the relationship information is thus a challenging and at the same time interesting problem from machine learning point of view and from the point of view of the user interest
33copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (2)
The Solution Our team approached the problem of dating recommendation
system with a preliminary analysis of prediction power of classification models in this type of tasks
The result of the project is an implementation of a recommendation system in a form of a web application called Partner-Tip
The application uses the social network relationships of the user friends and calculate the recommendation score with all the friends of opposite sex
34copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
PartnerTIPOnline dating system based on vkcomldquo (3)
35copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
The Problem Main objective of the project is to build a robust intelligent
system that counts money by classifying images of euro banknotes even when exposed to adversarial attacks
Making good classifier is not enough today it has to defend itself from malicious agents that try to manipulate with data
36copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Robust Euro Notes Classification(Adversarial Attack-Defense)
37copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
The problem Help every attendee of an event to connect with the right
people in an efficient time saving and personal manner The solution
The approach has both an offline section which takes the profile information of each attendee process it and calculate similarities between each pair of attendees
The online section which is a web application including a chatbot integrated in it The web application will be the main interface with the userattendee and it will recommend people to meet with to each attendee of the event
38copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Meeting the right people
39copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The problem
Text classification applied to an on-line book-store Writers upload their stories on the on-line platform Unordered arrangement of stories Not automatically grouped Inefficient access to stories
40copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
The solution A classifier system able to do the following was
constructed Automatically categories stories Improved and optimized accessing of stories Stories are arranged orderly
41copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
A Classification System for fictional stories
42copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (1)
The problem Automatically answer questions posed by humans
The solution Building a system that uses information retrieval and
natural language processing to automatically answers questions
43copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Answering Machine (Question Answering) (2)
44copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (1)
The problem People lose their pets and the platform helps find them Many cities have undocumented stray animals roaming
their streets and the platform helps catalogue them
The solution A web platform that works as an animal profiling service to
help people find their lost pets as well as cataloguing stray animals
45copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Finding Lost Pets (2)
46copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (1)
The problem What to cook with available ingredients
The solution Track ingredients purchased with the ticket of the
purchase and suggest recipes based on the list of items
47copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Cookit (2)
48copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (1)
The problem Managing duplicates (expressed in several forms) in
databases
The solution A software based solution to this problem using a Support
Vector Machine (SVM) classifier that will learn duplicates of the provided dataset using active learning
49copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Deduplication engine (2)
50copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Market Price Suggestion (1)
The problem Online auctionmarket services have gotten extremely
popular (eBay Mercari Wallapophellip) Two main scenarios
Selling an item For how much list an item What should be the starting price How to list and describe the item in order to earn more
Buying an item How much should we pay
51copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (2)
The proposed solution Simple price prediction application - web and mobile
Ease of use Clearly defined goals
The user inputs the basic characteristics of an item Item name - natural language Description - natural language Brand name - categorical Category - categorical Item condition - categorical Shipping payer - categorical
The system outputs the predicted price
52copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research CenterMarket Price Suggestion (3)
53copy Miquel Sagravenchez i Marregrave KEMLG-IDEAI 2020
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (1) The problem
Input data in two different modalities text + image that provide complementary information
The solution Output predicted class of emotion associated to the input Potentially useful for inter-user interaction platforms
(social media forums etc) Users share their experiences (through comments photos etc)
which communicate emotions Analyzing the emotion of the user can improve user experience
and personalization by providing content suitable for the current state of the user (sad angry happy etc)
Benefits Keep the interest and satisfaction of users (user retention) Encourage users to recommend the platform (user acquisition)
54copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Multi-Modal Emotion Classifier (2)
55copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (1)
The problem Piano fingering is useful to improve reading speed and
accuracy Usually not present in the sheets (needs to be deduced) The deduction is non-trivial for non-experts
56copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (2)
The solution An intelligent tool able to automatically add the fingering
to any upcoming piano sheet
57copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Automatic Piano fingering through data-driven knowledge (3)
58copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (1)
The problem In the USA 100 people
die each day from gun violence
In 2017 the number of civilian-owned firearms was 120 guns per 100 residents in the USA
There is a correlation between gun deaths and the number of guns
59copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (2)
The solution Development of a Gun Detection System to help
alerting when a violent event occurs Detecting guns in images provided by users Sending an alert message when a gun is detected with
reliable confidence Taking advantage of state-of-the-art Neural Network
architectures Using Transfer Learning techniques to adapt the
system to our goal Creation of a user-friendly interface to ease the userrsquos
interaction
60copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Gun Detection on images via MRCNN (3)
61copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (1)
The problem Secure photo sharing For events parties trips etc Keeping user data private
The solution An intelligent system for secure photo sharing Intuitive use of application Free to use
62copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Sharesio (Secure and convenient photo sharing) (2)
63copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (1)
The problem Prediction of future stock market values is complex Hidden patterns occur in short-term predictions News are considered to predict fluctuations due to world events
The solution An intelligent system capable of predicting future stock market values
of a given company The system consists of the combination of different Machine Learning
and statistical models Generate a standalone model for predicting how the stock values will
evolve as a time series (next k days) Download news articles from New York Times Reddit and Wall Street
Journal parse news articles and generate a model for predicting how an article may affect the stock market value
Generate a final model for joining both predictions
64copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
An Intelligent System for Stock Market Prediction (2)
65copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (1) The problem
Building a face generation and edition software component Multiple applications such as
Creating virtual avatars Observe how we would look with a different style before getting a
new haircut
The solution Deep learning system to build the face interpolator itself Need to
get meaningful features Access to the model through an API offered as service Graphical user interface (web app)
66copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Face Interpolator (2)
67copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (1)
68copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem The tablature format (TAB) is a useful notation when learning to play a
new instrument The tablature gives the finger placement rather than the musical
pitches However this format is unavailable on many songs The solution
A method that translate the more available (and standardised) MIDI format into the corresponding tablature
The converters that are already available are faulty in the sense that they do a rule-based translation that can not capture the intuitive finger placement of a correct tablature
The proposed solution instead trains multiple LSTM models on correctly translated tablatures to capture the intuitive translation
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Midi to Tab Automatic generation of guitar tablatures (2)
69copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (1)
70copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
The problem Quite often we find ourselves over-spending the money that weearn and left wondering what the money was spent on in order to reduce unnecessary expenditures
The solution Create a money tracker application that would allow a person to know
exactly how much money they are spending and where they are spending it
The core idea is to allow a user to take a picture of their receipt after every purchase and later be able to view the different types of items that they bought in a series of graphs
This will allow the user to know which categories of items they are spending an unnecessary amount of money on
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
Money Tracker (2)
71copy Miquel Sagravenchez i Marregrave KEMLG IDEAI-UPC 2021
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
Money Tracker (1)
Money Tracker (2)
Nuacutemero de diapositiva 72
Inte
llige
ntSy
stem
Pro
ject
Intelligent Data Science andArtificial Intelligence Research Center
72httpskemlgupcedu
Intelligent Data Science and Artificial Intelligence(IDEAI-UPC)
Miquel Sagravenchez-Marregravemiquelcsupcedu
Intelligent System Project
PART 1 ndash INTRODUCTION
Contents (1)
Contents (2)
Contents (3)
Contents (4)
Contents (5)
INTRODUCTION
Aims of the course
Description of the Teamwork
Intelligent System Project Timeline (1)
Deliverables of the IS Project (1)
Deliverables of the IS Project (2)
Deliverables of the IS Project (3)
Deliverables of the IS Project (4)
Deliverables of the IS Project (5)
Examples of Past ISP Projects
Examples of Past ISP Projects (1)
Examples of Past ISP Projects (2)
Examples of Past ISP Projects (3)
Analysing and interpreting tweets related to weather talking about the past or about the future making some sentimentmood analysis
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (1)
Detecting talks about topics of interest related to some business in the Linkedin Network (WhoTalk) (2)
A prediction system for bike and spot availabilities (Bicing predictor)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
A recommendation engine for movies (BAGmovies)
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
An Image Search Engine based on Artistic Style
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PCC - Parrot Communication with children - An intelligent interaction system for children with difficult emotion expressing skills through a flying drone with camera
PartnerTIPOnline dating system based on vkcomldquo (1)
PartnerTIPOnline dating system based on vkcomldquo (2)
PartnerTIPOnline dating system based on vkcomldquo (3)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Robust Euro Notes Classification (Adversarial Attack-Defense)
Meeting the right people
Meeting the right people
A Classification System for fictional stories
A Classification System for fictional stories
A Classification System for fictional stories
Answering Machine (Question Answering) (1)
Answering Machine (Question Answering) (2)
Finding Lost Pets (1)
Finding Lost Pets (2)
Cookit (1)
Cookit (2)
Deduplication engine (1)
Deduplication engine (2)
Market Price Suggestion (1)
Market Price Suggestion (2)
Market Price Suggestion (3)
Multi-Modal Emotion Classifier (1)
Multi-Modal Emotion Classifier (2)
Automatic Piano fingering through data-driven knowledge (1)
Automatic Piano fingering through data-driven knowledge (2)
Automatic Piano fingering through data-driven knowledge (3)
Gun Detection on images via MRCNN (1)
Gun Detection on images via MRCNN (2)
Gun Detection on images via MRCNN (3)
Sharesio (Secure and convenient photo sharing) (1)
Sharesio (Secure and convenient photo sharing) (2)
An Intelligent System for Stock Market Prediction (1)
An Intelligent System for Stock Market Prediction (2)
Face Interpolator (1)
Face Interpolator (2)
Midi to Tab Automatic generation of guitar tablatures (1)
Midi to Tab Automatic generation of guitar tablatures (2)
