2. 2
TOPICS
Where We Are with AI Today
What Is Artificial Intelligence, ML and DL
How Deep Learning Can Be Applied
Industry Use Cases:
Healthcare, Automotive, Finance, Retail
How Do We Get Started?
3. 3
“Find where I parked
my car”
AI IS EVERYWHERE
TOUCHING OUR LIVES
“Find the bag I just saw
in this magazine”
“What movie should
I watch next?”
4. 4Source: Gartner, “Architecting the On-Demand Digital Business”; Drue Reeves, Kyle Hilgendorf, Kirk Knoernschild, August 16, 2016
6. 6
GPU DEEP LEARNING
IS A NEW COMPUTING MODEL
TRADITIONAL APPROACH
Requires domain experts
Time consuming
Error prone
Not scalable to new problems
Algorithms that learn from examples
DEEP LEARNING APPROACH
Learn from data
Easily to extend
Speedup with GPUs
Expert Written
Computer Program
Car
Vehicle
Coupe
Car
Vehicle
Coupe
Deep Neural Network
8. 8
Every day, pathologists are tasked with providing
definitive cancer diagnosis to guide patient
treatment. However, keeping pace with the
massive volume of data and the variety of analysis
methods makes reliable predictions difficult. By
combining GPU deep learning and CUDA with
traditional pathology, PathAI’s approach is able to
reduce error rates by 85% in breast cancer
diagnosis.
AI: HELPING
DOCTORS DIAGNOSE
BREAST CANCER
9. 9
AI SEES THE
UNSEEN – COULD
REDUCE THE NEED
FOR BRAIN BIOPSIES
Brain tumors can be spotted by today’s MRIs, but
determining the right way to treat them requires
information about the tumor’s genomic makeup — data
that can only come from highly invasive brain biopsies.
Researchers at the Mayo Clinic may have found another
way. Using AI, Mayo discovered that the same genomic
data can be found in the MRIs themselves, hidden from
traditional analysis methods. Mayo used GPU-accelerated
deep learning with CUDA to train its systems where to
look and how to extract the information. The new system
has greater than 90% accuracy and has the potential to
greatly reduce the need for brain biopsies.
11. 11
THE MODERN
WAREHOUSE
BUILT ON AI
Worldwide retail e-commerce sales are expected
to reach $2 trillion in 2016, according to
eMarketer. With thousands of orders placed
every hour, data scientists at Zalando, Europe’s
leading online fashion retailer, applied deep
learning and GPUs to develop the Optimal Cart
Pick algorithm. Applying the algorithm resulted
in an 11% decrease in workers’ travel time per
item picked. The work is a good example of the
efficiencies that AI can discover for e-commerce,
manufacturing and other large-systems-based
industries.
12. 12
AI-DRIVEN
SMART SHOPPING
According to Forrester E-Commerce was a
$390B market in 2016 and is expected to double
by 2024. E-commerce company Jet.com
(acquired by Walmart) partners with multitudes of
suppliers with different offerings at different
prices. Jet uses GPU-accelerated AI to drive its
smart cart solution that fulfills orders at the lowest
prices though the smart bundling of supplier
offers. The platform finds the ideal merchant and
warehouse combination to lower the total order
cost. The bigger the shopping cart, the greater
the savings that can be generated.
14. 14
AI-DRIVEN ASSET
MANGEMENT
AI has led to break-through innovations across all
industries and the finance industry is no exception.
qplum, an online asset management firm, uses
quantitative trading techniques and invests using
data and GPU-powered deep learning. qplum blends
the mathematics of data-driven decision-making,
the science of behavioral economics, and the art of
effective communications. In the speed trade
category, qplum has been an innovation leader
having started with a $10,000 risk limit and, over
the last 10 years, making more than $1.4B in profits.
16. 16
Autonomous vehicles can reduce accidents,
improve the productivity of trucks and taxis,
and enable new mobility services —
transforming the $10 trillion transportation
industry. WEpods is piloting an autonomous
shuttle that leverages GPUs to compute data
and build a complete picture of the
environment, enabling it to safely navigate
traffic and other obstacles. It’s a revolutionary
new kind of transportation that offers the
convenience of a personal vehicle, without the
hassles of car ownership.
REVOLUTIONIZING
TRANSPORTATION
WITH AI
17. 17
Deep neural networks require a huge amount of
computational power and tremendous
amounts of data, which is particularly true with
safety critical systems, like self-driving
cars, where detection accuracy requirements
are extremely high. Zenuity is tackling this with
the combined power of DGX-1 and FlashBlade,
which is enabling them to make ground-
breaking progress in reducing training run
intervals, to the extent that they expect to be
able to iterate on their models.
DEVELOPING THE
VEHICLES OF THE
FUTURE
18. 18
AIRI: AI-READY INFRASTRUCTURE
18
• NVIDIA DGX-1 | 4x DGX-1 Systems | 4 PFLOPS
• PURE FLASHBLADE™ | 15x 17TB Blades | 1.5M IOPS
• ARISTA | 2x 100Gb Ethernet Switches with RDMA
• NVIDIA GPU CLOUD DEEP LEARNING STACK | NVIDIA
Optimized Frameworks
• AIRI SCALING TOOLKIT | Multi-node Training Made
Simple
HARDWARE
SOFTWARE
Extending the power of DGX-1 at-scale in every enterprise
20. 20
DO YOU HAVE ENOUGH LABELED DATA?
The Achilles heel of deep learning: You need a lot of labeled data.
Based on a presentation from Bryan Catanzaro
Without a large dataset, deep learning isn’t likely to succeed.
Labels:
Getting someone to decide the “right” answer can be hard (think about medical
imaging)
If a dataset requires skilled labor to produce labels, this limits scale / affects the
cost
21. 21
DO YOU HAVE ENOUGH LABELED DATA?
“As of 2016, a rough rule of thumb is that a supervised deep learning algorithm will
generally achieve acceptable performance with around 5,000 labeled examples per
category, and will match or exceed human performance when trained with a
dataset containing at least 10 million labeled examples.”
Ian Goodfellow, Yoshua Bengio, Aaron Courville
How much data is enough?
Goodfellow, Ian, Yoshua Bengio, and Aaron Courville. Deep learning. MIT Press, 2016.
22. 22
WHAT LEVEL OF ACCURACY DO YOU NEED?
How much accuracy you need? (mortgage risk calculation - high, celebrity portal - low)
Aim for lowest acceptable for the product
What is the measure:
• Accuracy (% correct)
• Coverage (% of examples processed)
• Precision (% of detections that are right)
• Recall (% of objects that are detected)
• Amount of error (for regression problems)
• What protective mechanisms to you need to safeguard the system from unavoidable
prediction error?
Defining and measuring accuracy
Goodfellow, Ian, Yoshua Bengio, and Aaron Courville. Deep learning. MIT Press, 2016.
23. 23
BEST PRACTICE FOR STARTING A DL PROJECT
Hypothesis for the
business outcome you
believe DL can solve
Current, needed
Data – enough to train?
Current AI & DL skills
People training plan
Current IT Infrastructure
(Cloud, On-premise)
ASSESS DESIGN & SELECT LEARN DEPLOY
Analyze data to train
(e.g. text, video,
images, structure)
Plan research (Data
Scientist) & deployment
models (IT Architect)
Select DNN Network,
Libraries & Frameworks
Begin training
Feedback on outputs so
the network can learn
Achieve training state
that provides actionable
data for business
decisions
Performance
monitoring
Optimization of trained
DNN for deployment
performance
Move trained outcomes
to inferencing platform
Begin inferencing (e.g.
search, speak, translate,
classify, segment,
predict, recommend)
Expand DL Training to
adjacent areas
Performance
monitoring
24. 24
CLOUD, ON-PREMISE OR HYBRID?
Cloud
Pre-trained models
Ease of integration into
your app development
Cloud scale & efficiency
Cloud billing
On – Premise
Instant productivity
Desktop to data center
Tuned /optimized perf.
Data security
Hybrid
Any compute environment
Common software stack
Flexibility (e.g. train
local, inference in cloud)
25. 25
BE READY FOR THE RACE FOR TALENT
• Freedom, flexibility and
challenges attract talents
• Provide great tools and
infrastructure
• Data Science + Business +
IT have to partner
together
26. 26
DEEP LEARNING INSTITUTE
DLI Mission: Help the world to solve the most challenging
problems using AI and deep learning
We help developers, data scientists and engineers to get
started in architecting, optimizing, and deploying neural
networks to solve real-world problems in diverse industries
such as autonomous vehicles, healthcare, robotics, media
& entertainment and game development.
https://www.nvidia.co.uk/deep-learning-ai/education/