AI IN FINANCIAL SERVICES
FIAZ MOHAMEDINTEL AI PRODUCTS GROUP
ANALYTICS NEEDS AI
HindsightWhat Happened
InsightWhat Happened and Why
What Will Happen, When, and Why
Simulation-Driven Analysis and Decision-Making
Self-Learning and Completely Automated Enterprise
Mature Data Lake
Computerized Human Thought Simulation and Actions Towards Autonomic Enterprise
DescriptiveAnalytics
DiagnosticAnalytics
PredictiveAnalytics
PrescriptiveAnalytics
CognitiveAnalytics
Advanced Analytics
Operational Analytics
To
day
Em
erg
ing AI
is a vital tool for
reaching higher
maturity & scale
Foresight
VARIETY OF AI APPROACHES
CLASSICAL ML DEEP LEARNING REASONING EMERGING
Statistical problems, recommendation
engines, transparency
requirement, etc.
Image/speech recognition, natural
language processing, pattern
recognition/detection, etc.
Multivariate supply chain probe, full database fraud
detection, whole CRM churn analysis, etc.
AI research: ‘sequence alignment’
in computational biology, ‘binary
neural network based inferencing’, etc.
EX
AM
PLES
CLASSICAL ML VS. DEEP LEARNING
Random Forest
Support Vector Machines
Regression
Naïve Bayes
Hidden Markov
K-Means Clustering
Ensemble Methods
More…
CLASSICAL ML
Using optimized functions or algorithms to extract insights from data
TrainingData*
Inference, Clustering, orClassification
New Data*
DEEP LEARNING
Using massive labeled data sets to train deep (neural) graphs that can make inferences about
new data
Step 1: Training
Use massive labeled dataset (e.g. 10M tagged images) to iteratively adjust weighting of neural network connections
Step 2: Inference
Form inference about new input data (e.g. a photo) using trained neural network
Hours to Days in Cloud
Real-Time at Edge/Cloud
New Data
Untrained Trained
Algorithms
CNN,RNN,RBM..
*Note: not all classic machine learning functions require training
REASONING SYSTEMS
MEMORY BASED
Using associations between concepts from multiple data types to make sense of
complex situations
Flexibility to handle ALL data types at once
Incorporate new data in real-time
Transparent and explainable
e.g. under what system conditions should I perform preventive maintenance to avoid a failure?
LOGIC BASED
Using a rule-based reasoning engine, usually hand-created or maintained, to
perform logical inferencing steps
Explicit encoding of knowledge
Repeatable, reversible, deterministic
Transparent and explainable
e.g. should I maintain or alter my equity portfolio given my risk profile?
INTEL® NERVANA™ PLATFORM
A full stack, user-friendly & turnkey system
that enables businesses to develop and
deploy high-accuracy AI solutions in record
time:
-or-
Compress the Development Cycle
PureAcceleration
Benefits Beyondthe Box
*Other names and brands may be claimed as the property of others.
DEEP LEARNING IN PRACTICE
TABULAR DATA ANALYSIS
IMAGE CLASSIFICATION
Arjun
DOCUMENT ANALYSIS
VIDEO ACTIVITY DETECTION
Baby Crawling 0.91
TIME SERIES DATA
SPEECH ANALYSIS
INTELLIGENT KNOWLEDGE MANAGEMENT
Highly diversified financial services and asset management firm with >$750B
Deep learning-based clustering of documents by topic
Faster and more accurate insights by reducing the number of documents needed to drive decisions
Reduce time to insight for portfolio managers
Knowledge management system based on DL natural language processing engine to:
• Ingest vast stores of unstructured data
• Predict topical relevance of paragraphs
• Recommend relevant articles with Q&A system
Client
Challenge
Solution
Advantages
INTELLIGENT ORDER MATCHING
Global financial services and asset management firm
• Streamlining of the trade execution process by automatically assessing quality of prices
• Learnings can drive better trading strategies and increased profitability
In trading system, quotes are placed by counterparties for instruments at different prices. The prices on the system are indicative and not firm and this results in different possible outcomes and different liquidity.
DL system to predict the probability of fixed income orders clearing at a user specified price level and trading partner• Model was trained on historical trading data • Allowed the client to select a counterparty and a price that had the
highest probability of being filled
Client
Challenge
Solution
Advantages
ORDER BOOK SEARCH AND PREDICTION
Leading U.S. equity marketplace
• More accurate matches than non-deep learning approaches
• Enables new use cases for fraud detection, anomaly detection, and other future intelligent applications
Drive future investment activity based upon similar historical data patterns
DL system to: • Ingest public order book data and
automatically learn patterns of activity• Enable search queries for similar
historical patterns
Client
Challenge
Solution
Advantages
Limit order book search
THANK YOU