+ All Categories
Home > Software > MACHINE LEARNING – THE WHY, WHAT AND HOW

MACHINE LEARNING – THE WHY, WHAT AND HOW

Date post: 23-Jan-2018
Category:
Upload: sas-in-deutschland-oesterreich-und-der-schweiz
View: 1,985 times
Download: 0 times
Share this document with a friend
12
Copyright © SAS Institute Inc. All rights reserved. MACHINE LEARNING – THE WHY, WHAT, AND HOW Dr. Andreas Becks, SAS @becks_andreas
Transcript
Page 1: MACHINE LEARNING –  THE WHY, WHAT AND HOW

Copyright © SAS Inst itute Inc. A l l r ights reserved.

MACHINE LEARNING –THE WHY, WHAT, AND HOWDr. Andreas Becks, SAS@becks_andreas

Page 2: MACHINE LEARNING –  THE WHY, WHAT AND HOW

Copyright © SAS Inst itute Inc. A l l r ights reserved.

Maschine Learning – An Example

Everybody uses machinelearning today – in yourphoto app it identifiesfaces and persons on

your images.

Page 3: MACHINE LEARNING –  THE WHY, WHAT AND HOW

Copyright © SAS Inst itute Inc. A l l r ights reserved.

Cheap Compute Powerand Parallel Processing

“2.5 Exabytes of data are produced every day – that’s 90 years of HD videos.”

“90 % of the worlds data today has been created in the last 2 years alone.”

March 2015, DN Capital

Availability of Data R&D on Algorithms

Machine Learning – Why is ML so HOT?

Three combining trends: bigdata, massive compute

power, and better algorithms.

Page 4: MACHINE LEARNING –  THE WHY, WHAT AND HOW

Copyright © SAS Inst itute Inc. A l l r ights reserved.

Cognitive Computing

ArtificialIntelligence

Machine Learning

Neural Networks

Deep Learning

There is a lot of buzz aroundsome related terms. Oftenconfused, are they neither

separted nor the same.

Page 5: MACHINE LEARNING –  THE WHY, WHAT AND HOW

Copyright © SAS Inst itute Inc. A l l r ights reserved.

Supervised Learning

10 Algorithms Machine Learning Engineers Need to KnowTwo major groups of algorithms form a set of machine learning tools.

Source: KDnuggets based on Udacity’s Intro to Machine Learning

Unsupervised Learning

Naïve Bayes Classific.

Linear Regression

Logistic Regression

Decision Tree

Support Vector

MachinesEnsemble Methods

Unsupervised Learning

Principal Component

Analysis

Cluster Algorithms

Singular Value

Decompo-sition

Indepen-dent

Component Analysis

Page 6: MACHINE LEARNING –  THE WHY, WHAT AND HOW

Copyright © SAS Inst itute Inc. A l l r ights reserved.

The Principle: Recognizing ActivitiesLearning dependent patterns of movements

Different sports, different movements –algorithms

learns characteristic patternsin sensor data.

Page 7: MACHINE LEARNING –  THE WHY, WHAT AND HOW

Copyright © SAS Inst itute Inc. A l l r ights reserved.

Learning dependent patterns of movements

X -

Wri

st

X - Ankle

Motion Trajectories

Raw data by motionand inertial sensors

X -

Wri

st

True Activity Labels

X - Ankle

Training data: movement in context

Machine Learning

Support Vector MachinesNeural Networks

Learned classification

The Principle: Recognizing Activities

Combination of algorithms clusters the different raw data and connects it todifferent activties. Doing that, new and previously unknown incoming data can be

appropriately classified.

Page 8: MACHINE LEARNING –  THE WHY, WHAT AND HOW

Copyright © SAS Inst itute Inc. A l l r ights reserved.

Use Cases for Machine Learning

Manufacturing

• Predictive Maintenance

• Warranty reserve estimation

• Propensity to buy

• Demand forecasting

• Telematics

Healthcare

• Alerts and diagnostics from real-time patient data

• Risk stratification

• Proactive health management

Retail

• Predictive inventory planning

• Recommendation engines

• Upsell and cross-channel marketing

• Market segmentation

• ROI and customer value

Travel and Hospitality

• Aircraft scheduling

• Dynamic pricing

• Consumer feedback (social media analysis)

• Customer complaint resolution

Energy

• Smart grid management

• Power usage analytics

• Energy demand and supply optimization

• Seismic data analysis

• Carbon emission and trading

Sources: Forbes Magazine, July 2016, Harvard Business Review, February 2016

Financial Services

• Risk analytics and regulation

• Customer segmentation

• Cross-/Up-Sell

• Campaign management

• Credit worthiness evaluation

More than a third of earlymovers also saw gains in bottom-line performance using

machine-reengineering to slash 15% to 70% of costs from certain processes.

Business processes in

every industrywill be affected.

Page 9: MACHINE LEARNING –  THE WHY, WHAT AND HOW

Copyright © SAS Inst itute Inc. A l l r ights reserved.

Example: Predict Maintenance of Computer Tomographs

Thousands of devices,

Tens of thousands event codes per day,sensor data

1,000s of predictive models

Challenge: Predict failures of components 5 to 10 days in advance >70% precisionand <20% false positives

Impact on operative processes

Real applications tendto become complex.

It‘s not only 1 model / algorithm!

Page 10: MACHINE LEARNING –  THE WHY, WHAT AND HOW

Copyright © SAS Inst itute Inc. A l l r ights reserved.

Managing the Analytical Life CycleOnly integration of best models into business processes generates value

Discover Deploy

Prepare data

Explore

Model

Integrate into business processes

Execute

Evaluate

Ask

IT, Business Analyst, LoB

Robust

Automation

Actions

Decisions

Operations

Experiments

Data Science

New data

Innovation

Explorative

Data Scientist, LoB

DATA

€Analytics needs two things: building

best models AND bringing theminto production. Automation

needed due to complexity / scale.

Page 11: MACHINE LEARNING –  THE WHY, WHAT AND HOW

Copyright © SAS Inst itute Inc. A l l r ights reserved.

Summary & Next StepsCollect & explore data, learn patterns, and automate decisions.

Data Science PLUS lines of businesses

Only the integration in new processes will lead to

business value. Everyone needs to understand the

possibilities of ML.

Machine Learning improves analytics

Machine learning is an essential part

of Advanced Analytics –and every corporate

strategy.

Operationalization requiresdiscovery and action

Automation and integration are more

important than algorithms

Machine Learning

This e-book provides a primer on these innovative techniques as well as 10 best practices and a checklist for machine learning readiness.

Page 12: MACHINE LEARNING –  THE WHY, WHAT AND HOW

Copyright © SAS Inst itute Inc. A l l r ights reserved.

Get in Contact with me!

Dr. Andreas Becks, SAShttps://twitter.com/becks_andreashttps://www.linkedin.com/in/andreas-becks-10998058/


Recommended