2. 2
• Why is Artificial Intelligence such
a hot topic now?
• What are some applications of AI
in the Enterprise?
• How to get started on your AI
journey?
• Conclusions
Agenda
5. 5
The Resurgence of Artificial Intelligence
• Significant advances in hardware
capability
• Rapid progress in research and
applications using neural networks
• Significant technology investments
By 2019, deep learning will provide best-
in-class performance for demand, fraud,
and failure prediction. - Gartner
6. 6
“
By 2020 AI will be a top five
investment priority for more
than 30% of CIOs.
—Gartner BI Summit,
February, 2017
“The Resurgence of AI
7. 7
Deep Learning Innovation in Computer Vision
Continuous Improvement in Supervised Learning Methods
Recent Image-Net Results
8. 8
Deep Learning is driving the resurgence of AI
How is it different?
• Multiple layers in neural network with intermediate data
representations to facilitate dimensional reduction.
• Interpret non-linear relationships in the data.
• Derive patterns from data with very high dimensionality.
Why do we care?
• Ability to create value with little or no
domain knowledge required.
• Ability to incorporate data from across
multiple, seemingly unrelated sources.
• Ability to tolerate very noisy data.
9. 9
Classic Artificial Neural Network
• Simple Feed Forward Network
• Linear Classifier
Single Layer Perceptron Multi Layer Perceptron
• Feature Learning instead of Feature
Engineering
• Highly dimensional data is organized into
learning features as network is connected
from layer to layer
• Solving Activation Function of hidden layer:
Gradient Descent (best weights) assisted by
Back Propagation (partial derivatives)
10. 10
AI Technology Landscape
Point Solutions Specialized APIs General Purpose Frameworks
VISUAL VISION
ASSISTIVE SPEECH
OPERATIONS LANGUAGE
Teradata Partner
12. 12
Industry Specific Use Cases
High-Dimensional Data
Image
Video
Audio
Time Series
Text
• Many already have working solutions using non-DL Machine Learning Techniques
• Deep Learning is delivering improvement in performance on complex problems
Automotive Retail
• Navigation, Guidance, Assistance
• Predictive Maintenance
• Visual Search
• Recommendation
• Text Analytics
• Assistants
• Brand Analytics
Manufacturing & High-Tech Health Care
• Image/Audio/Video
• Reinforcement Learning – Systems
Optimization
• Plant Operations Optimization
• Image-based Analysis
• Drug Discovery
Financial Services & Insurance Cross-Industry
• Anti-Fraud
• Portfolio Optimization
• Damage Assessment
• Cyber Security
• Call Center Audio
13. 13
Mobile Personalization
• Generalize rules (e.g., categories of interest)
• Memorize exceptions (e.g., common pairs)
• Google Play Store production and other leading digital companies
• Projects in banking, telco, retail
Source: Google
14. 14
Banking Anti-Fraud: Business Drivers
• Goal: fraud detection across products
• Trends
– Evolution of new payment methods
– Mobile payments exploding
– Fraud evolving rapidly, increased sophistication
• Traditional approach is hand-written rules
• Cost, delay and customer impact of false positives
15. 15
• Phased implementation
approach
– Simulated result
– Champion/challenger testing
– Production deployment
• Significant improvements over
traditional rules-based
techniques
• Techniques
– Boosted Decision Trees
– Recurrent Neural Networks
– Generative Adversarial Networks
• Tools: Spark, Hadoop, TensorFlow
Banking Anti-Fraud: Solution Approach
16. 16
Key Requirement: Model Interpretation
• Technique used: LIME (Locally Interpretable Model Explanation)
https://github.com/marcotcr/lime
Source: Ribeiro et al.
17. 17
• Provide smart assistance to
drivers
– Navigation and safety
– Realtime Pricing
– Vehicle comfort
– Parking assistance
• Leverage video and other
sensors
Connected Car Assistance
Real-Time
Streaming
Streaming
Results
Traffic Data
Service
Navigation
Update
Object Detection
Object
Segmentation
Motion Detection
GPU Training
TF Serving
Online
Inference
Model
Update
s
18. 18
• Large Tier 1 automotive supplier
including navigation
• Objective: To build a system that
detects objects and scenes from
in car video to improve
navigation and guidance
Customer and Team Project Details
• Pilot: Dec. 2016 – March 2017
• Deliverables:
• Annotation Tool
• Bounding Box Demo
• Research Report &
Presentation
• Next Phase In Progress:
• Live Demo at Internal Car
Show
• Object Tracking
• Road Segmentation
19. 19
Customer Challenge
Customer Data
Video Files
Frames
Bounding Box
Scene Labeling
Bounding Box & Scene
Labeling
XML
JSON
XML
JSON
XML
JSON
Create File
Create File
Create File
Create File
• Detect 6 classes in car driving video e.g., car,
truck, bike, pedestrian
• Detect roads/lanes
• Next phase:
• Identify stopped vs. moving vehicles
• Identify if classes are on/off the road
20. 20
Our Approach
• Survey of State of the Art research and open source
• Open Source code often does not run out-of-the-box, so we worked to
overcome quirks and limitations
• Run available pre-trained models on Customer data.
– Trained and created new models
• Visual inspection of results
• Mean Average Precision performance metric calculated identically for
the tested methods on three test datasets
• Determining appropriate infrastructure for initial experimentation
26. 26
Focus First on Pilot into Production
Sets up Phase Two: Scale COE, Standardize Capabilities
Investigate
Test
Engineer
SimulateIntegration
Analyze
Data
Go Live
Handover
Validate
Activities: Define business
opportunity, understand data
available, test model
approaches, potentially
generate data
Outcome: Proposed solution
approach
Discovery/Insights
Activities: Architecture
selection, software engineering
of model and simulation
Outcome: Predicted impact of
model
Live Test
Activities: Integration into
live business process
(Champion/Challenger),
analysis, iteration
Outcome: Benefit
measurement, live learnings,
improvement
Production
Activities: Go Live, Analytics
Ops integration, Hand Over
Outcome: System scaled,
application teams and ops
trained and operating
Assessment
Insights
Production
Live Test
Cross-Functional
Teams
Cross-Functional Teams
27. 27
Analytics Ops for Cross-Functional AI Teams
Constant
Monitoring
Test and
Deploy-
ment
A/B Testing
Automated
Training &
Scoring
Application
Integration
28. 28
Our Approach
Teradata Deep Learning CommunityTeradata Labs
Dozens of Experts in Deep Learning,
Image/Audio/Video Processing,
Computer Vision, GPU
200+ Practitioners delivering
Artificial Intelligence Business Value
on Customer Projects
500+ Solution Architects, Business
Consultants and Software Engineers with
knowledge of Artificial Intelligence Tools,
Techniques and Technologies. Deep
expertise in retail and across industries.
Experts
Practitioners
Interest
Industry
Collaborations
Academic
Collaborations
Analytics
Ops
Data
Management
33. 33
• Handwritten check volume is decreasing
however processing checks has many
fixed costs
• Handwriting recognition to reduce
manual processing and fraud
examination resulting in cost savings
• Techniques:
– Convolutional Neural Network
– Image Processing
– Natural Language Processing
• Tools: Spark, Hadoop, TensorFlow
Automated Check Processing
Check Images
To Hadoop
ImageMagick
Processing
Handwriting
Recognition
34. 34
Software
Open Source
Frameworks from
digital giants will
dominate notably
TensorFlow, Keras
Framework
fragmentation will
remain way of life (see
JavaScript)
Integration with
enterprise analytics
environments will
emerge
Hardware
General purpose
chips will dominate on
price/performance
(compared to ASICs)
Model scoring/AI
deployment will be a
big area of hardware
innovation
Cloud training is path
of least resistance
now but AI appliances
on the horizon
Datat&Algorithms
Training & labeled
data services will
have a short
lifespan
Unsupervised &
Transfer Learning
goes mainstream
Algorithms and
code will
continue to
remain open
Hypotheses