sas and artificial intelligence · pattern recognition prediction classification image ......
TRANSCRIPT
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
SAS and Artificial Intelligence
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
OBJECTIVE
Tactics StrategyFuture Vision Current Reality
Leading with Intelligence
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
Artificial Intelligence (AI) is the science of training computers to perform tasks that typically require human intelligence to complete.
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
Performance Assessment
Extract real-time insights during games
Object Detection
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
Strategic AI Applications
Retail
Conversational Chat Bots
Contextual Marketing
Personalized Virtual Shopper
Fraud Detection
Credit Analysis
Automated Financial Advisors
Banking
Smart Cities
Sensor Fusion
Facial Recognition
Government
Predictive Diagnostics
Biomedical Imaging
Health Monitor
Heath and Life Sciences
Supply Chain Optimization
Automated Defect Detection
Energy Forecasting
Manufacturing and Energy
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
A Tactical Approach to AI
Machine Learning Natural Language Processing
Deep Learning Natural Language Understanding
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
EnergyForecasting
Use short and long-term variability to improve accuracy
Deep Learning
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
A Tactical Approach to AI
Machine Learning Natural Language Processing
Deep Learning Natural Language Understanding
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
Supply Chain Stability
Improve response to changes in the market
Natural Language Understanding
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
A Tactical Approach to AI
Machine Learning Natural Language Processing
Deep Learning Natural Language Understanding
Pattern Recognition ● Prediction ● Classification ● Image Recognition ● Speech to Text ●Cognitive Search ● Natural Language Interaction ● Natural Language Generation
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
Manufacturing Optimization
Identify defects during production
Pattern Recognition
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
A Tactical Approach to AI
Machine Learning Natural Language Processing
Deep Learning Natural Language Understanding
Pattern Recognition ● Prediction ● Classification ● Image Recognition ● Speech to Text ●Cognitive Search ● Natural Language Interaction ● Natural Language Generation
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
Wildlife Conservation
Provide non-invasive monitoring of
endangered species
Image Recognition
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
We provide elements and capabilities
for people that aspire to build an AI system.
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
These AI elements are being embedded in the SAS platform.
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
Our approach to AI is
augmenting human efforts
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
Change the World
Advancing analytics by embedding AI in the
SAS Platform
Apply analytics to data
The SAS Platform
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
Input
Deep learning
Learning
MilitarySurveillance
Crop Yields
Fraud MedicalImages
Autonomous Vehicles
CustomerTargeting
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
Introduction to deep Learning
Agenda
4
1Deep Learning Defined
2Sample Business Applications
3Deep Learning Architectures
Comparison with other Algorithms
5Competitive Landscape
High Level Planning6
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
The Basics Deep Learning Defined
Deep learning is a sub-class of machine learning based on a set of algorithms that attempt to model high-level abstractions in data by using multiple processing layers with complex structures.
Deep learning platform is a platform which help users to build different deep learning architectures or facilitate users to apply deep learning to a wide range of business applications with apps and services.
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
Deep learning Why Deep learning?
• Very flexible structures of networks;
• Learning often improves with more data;
• Requires minimal feature engineering;
• Perform best in speech, text, image and video recognition;
Problems in Deep learning?
• Long training time;
• Hard to train due to the unstable gradient, numerous hype-
parameters and flexible architectures;
• Overfitting;
• The extracted features may be non-interpretable;
C op yr i g h t © 2015 , SAS Ins t i t u te Inc . A l l r i g h ts r eser v ed .
deep Learning WHY IS DEEP LEARNING SO HOT?
• Drastically increased chip processing abilities such as general purpose graphical processing units (GPGPUs) and MPP platforms;
• Significantly increased size of data for training, making the overfitting problem much
less severe;
• Great advances of machine learning algorithms for deep learning such as min-batch
random gradient descent;
Google Trends over time – Deep Learning
C op yr i g h t © 2015 , SAS Ins t i t u te Inc . A l l r i g h ts r eser v ed .
Business applications
UNLIMITED BOUNDS
Finance: Time series prediction and forecasting; fraud detection
Cybersecurity: Protection of information systemsMedical applications: Diagnostics, image recognitionAgriculture: Image recognition and feature codingMilitary Surveillance: Video recognition, planning and
optimizationComputer recognition: Image tagging and speech, text
recognitionAutonomous vehicles – cars, drones, robots, etc.
And much, much more.
C op yr i g h t © 2015 , SAS Ins t i t u te Inc . A l l r i g h ts r eser v ed .
Deep Learning ARCHITECTURES
Deep Forward Nets
Convolutional Networks
Recurrent Networks
Auto-encoders4
Copyr ight © SA S Inst i tute Inc . A l l r ights reserved .
What’s your
vision?