7. 7
Sensors
● Colour video camera
(1280 x 960) + infrared
camera (640 x 480)
● Pointcloud with height
and width data
● Continuous
streaming/changing data
11. AMAZING ACHIEVEMENTS IN AI
Play Go Play Doom Learn Paint Style Synthesize Voice
Write Captions Learn Motor Skills Learn to Walk Drive
12. 2016 – Baidu Deep Speech 2
Superhuman Voice Recognition
2015 – Microsoft ResNet
Superhuman Image Recognition
2017 – Google Neural Machine Translation
Near Human Language Translation
100 ExaFLOPS
8700 Million Parameters
20 ExaFLOPS
300 Million Parameters
7 ExaFLOPS
60 Million Parameters
To Tackle Increasingly Complex Challenges
NEURAL NETWORK COMPLEXITY IS EXPLODING
13. 13
DEEP LEARNING FOR SELF DRIVING CARS
Multi-class detection (DriveNet)
OpenRoadNet LaneNet 3D Bounding Boxes
14. 14
Driving is a learned behavior that people do
as second nature. Yet one that is impossible
to program a computer to perform. Using all
of the AI capabilities of NVIDIA DRIVE PX 2,
our research AI car, BB8, watches humans
drive, and has learned to drive in all kinds
of conditions — on highways and dirt roads,
through obstacle courses, at night, and in the
rain. Processing data from multiple cameras,
BB8 can even look both ways before safely
crossing a busy road on its own.
NVIDIA BB8 AI CAR —
LEARNING BY EXAMPLE
Center CameraLeft Camera Right Camera
19. 19
Long short-term memory (LSTM)
Hochreiter (1991) analysed vanishing gradient “LSTM falls out of this almost naturally”
Gates control importance of
the corresponding
activations
Training
via
backprop
unfolded
in time
LSTM:
input
gate
output
gate
Long time dependencies are preserved until
input gate is closed (-) and forget gate is open (O)
forget
gate
Fig from Vinyals et al, Google April 2015 NIC Generator
Fig from Graves, Schmidhuber et al, Supervised
Sequence Labelling with RNNs
25. 25
Generative Adversarial Networks
Tutorial on Celebs dataset: http://bit.ly/2lEgHDZ
Encoder-decoder
CNNs learn the
mapping from input
to output.
Provides the loss
function so we don’t
have to.
Wasserstein GANs
solve the stability
problem ie can train
to convergence.
T-SNE clustering =>
the GAN learns the
notion of what makes
images similar and
how to make them
close in latent space.
Totally unsupervised
the model learns
discriminative
features of the
dataset without ever
being told what they
are.
26. 26
Stacked GANs for text to SuperRes image
https://arxiv.org/pdf/1612.03242v1.pdf
Han Zhang et al, Rutgers University
Lehigh University, The Chinese University of Hong Kong & University of North Carolina at Charlotte
27. 27
Using HPC and deep learning,
InSilico Medicine created the
DeepPharma platform which
provides pharmaceutical companies
rapid and accurate analysis of
massive amounts of data.
DeepPharma paves the way for
personalized medicine.
AI DRIVES NEW
DISCOVERIES
28. 28
A PLETHORA OF HEALTHCARE STORIES
Molecular Energetics
For Drug Discovery
AI for Drug Discovery
Medical Decision
Making
Treatment Outcomes
Reducing Cancer
Diagnosis Errors by
85%
Predicting Toxicology
Predicting Growth
Problems
Image Processing Gene Mutations Detect Colon Polyps
Predicting Disease from
Medical Records
Enabling Detection of
Fatty Acid Liver Disease
38. 38
TTQ: Trained Ternary Quantization
“as easy as full precision”
ICLR 2017 https://openreview.net/pdf?id=S1_pAu9xl
Chenzhuo Zhu, Tsinghua University & Song Han, Stanford University
William J Dally, NVIDIA
Ternary trades-off model size, accuracy, performance vs. binarized weights
39. 39
In September 2015, 100 years after Einstein predicted
them, gravitational waves were observed for the first
time by astronomers at the Laser Interferometer
Gravitational-wave Observatory (LIGO) originating from a
pair of merging Black Holes 1.3B light years away.
“Seeing” gravity —a feat that Einstein doubted would
ever be achieved— opened the door to a new class of
astrophysics, along with a daunting new challenge:
observing these waves in parallel with electromagnetic
waves, radio waves and visible light, and analyzing the
combined data in real time.
Scientists at NCSA are using GPU-powered deep learning
to make this computationally intensive approach
possible. Using a deep Convolutional Neural Network
(CNN), NCSA trained its system to process gravitational
wave data more than 100 times faster than its previous
machine learning methods — making real-time analysis
possible and putting us one step closer to understanding
the universe’s oldest secrets.
“SEEING” GRAVITY
IN REAL TIME
40. 40
AI IS SPEEDING
THE PATH TO
FUSION ENERGY
Fusion is the future of energy on Earth. But it’s a
highly sensitive process where even small
environmental disruptions can stall reactions and even
damage multi-billion $$ machines. Current models can
predict the disruptions with 85% accuracy, but ITER
will need something more precise. Researchers at
Princeton have developed the Fusion Recurrent Neural
Network (FRNN) using deep learning and NVIDIA GPUs
to predict disruptions and make adjustments to
minimize damage and downtime. Even a 1%
improvement in the prediction accuracy can be
transformative considering the immense scale and
cost of fusion science. Today, FRNN is on the path to
achieve 95% accuracy for ITER’s tests.
41. 41
NVIDIA INDEX
Leading HPC Tool to Analyze Large-Scale Data for Faster Discoveries
Interactive
Built for Large-Scale DataRemote Visualization
Performance @ Scale
Visualize Anywhere
42. 42
WHAT’S NEW IN DIGITS 5
IMAGE SEGMENTATION MODEL STORE
Partition images into regions of interest Download pre-trained neural networks
43. 43
COMPUTER VISION SPEECH AND AUDIO NATURAL LANGUAGE PROCESSING
Object Detection Voice Recognition Language Translation
Recommendation
Engines
Sentiment Analysis
DEEP LEARNING
cuDNN
MATH LIBRARIES
cuBLAS cuSPARSE
MULTI-GPU
cuFFT
Image Classification
Accelerated Deep Learning TRAINING Stack
Productivity Layer/Rapid experimentation: DIGITS, NVIDIA GPU Cloud
DEEP LEARNING FRAMEWORKS
PADDLE PADDLE
48. Max-Q operating mode (< 7.5 watts) delivers up to 2x energy efficiency vs. Jetson TX1 maximum performance
Max- P operating mode (< 15 watts) delivers up to 2x performance vs. Jetson TX1 maximum performance
ANNOUNCING JETSON TX2
JETSON TX2
EMBEDDED AI SUPERCOMPUTER
Advanced AI at the edge
JetPack SDK
7.5 watts full module
Up to 2X performance or 2X energy efficiency
49. Jetson TX1 Developer Kit reduced to €549/£459 – Jetson TX1/TX2 Developer kits have same price for education
JETSON TX2
DEVELOPER KIT
€649/£544 Web or retail
€350/£300 education
Order now US and Europe, shipping now
APAC / other regions starting in April
51. 51
TESLA V100
THE MOST ADVANCED DATA CENTER GPU EVER BUILT
5,120 CUDA cores
640 NEW Tensor cores
7.5 FP64 TFLOPS | 15 FP32 TFLOPS
120 Tensor TFLOPS
20MB SM RF | 16MB Cache | 16GB HBM2 @ 900 GB/s
300 GB/s NVLink
https://devblogs.nvidia.com/parallelforall/inside-volta/
52. 52
Productivity That Follows You
From Desk to Data Center to Cloud
Access popular deep learning
frameworks, NVIDIA-optimized
for maximum performance
DGX containers enable easier
experimentation and
keep base OS clean
Develop on DGX Station, scale on
DGX-1 or the NVIDIA Cloud
52
EFFORTLESS
PRODUCTIVITY
53. 53
Training organizations and individuals to solve challenging problems using Deep Learning
On-site workshops and online courses presented by certified experts
Covering complete workflows for proven application use cases
Image classification, object detection, natural language processing, recommendation systems, and more
www.nvidia.com/dli
Hands-on Training for Data Scientists and Software Engineers
NVIDIA Deep Learning Institute
54. 54
NVIDIA
INCEPTION
PROGRAM
Accelerates AI startups with a boost of
GPU tools, tech and deep learning expertise
Startup Qualifications
Driving advances in the field of AI
Business plan
Incorporated
Web presence
Technology
DL startup kit*
Pascal Titan X
Deep Learning Institute (DLI) credit
Connect with a DL tech expert
DGX-1 ISV discount*
Software release notification
Live webinar and office hours
*By application
Marketing
Inclusion in NVIDIA marketing efforts
GPU Technology Conference (GTC)
discount
Emerging Company Summit (ECS)
participation+
Marketing kit
One-page story template
eBook template
Inception web badge and banners
Social promotion request form
Event opportunities list
Promotion at industry events
GPU ventures+
+By invitation
www.nvidia.com/inception
55. 55
October 10-12, 2017 | Munich, Germany
#GTC17EU www.gputechconf.eu
CONNECT
Connect with technology
experts from NVIDIA and
other leading organizations
LEARN
Gain insight and valuable
hands-on training through
sessions & instructor-led labs
DISCOVER
See how GPUs are creating
amazing breakthroughs in
important fields such as
deep learning and AI
INNOVATE
Hear about disruptive
innovations from startups
25% DISCOUNT CODE : AlisonLowndesGTCEU17
3 Days | 150+ Sessions | 100+ Press & Analysts | 60+ Exhibitors