2. 2
Doctors typically use their eyes to examine CT
scan images, looking for small nodules in an
attempt to deduce whether they’re benign or
malignant. When the nodules are small, they’re
harder to spot. And the result is that lung cancer
is often detected too late, leading to a dismal
17% survival rate.
Trained on DGX Station, 12 Sigma Technologies
could reduce the time-consuming workload for
both diagnostic and reporting, as well as change
the existing lung disease diagnostic practice
from being dependent on the subjective
experience of radiologists to being based on
objective clinical data.
AI TO SPOT LUNG
CANCER EARLY
3. 3
In the healthcare industry, there exists mountains
of underutilized data. 16 Bit is using AI to unlock
the value hidden in this data to augment
diagnostics and improve the quality and
accessibility of healthcare.
16 Bit is using GPU-accelerated deep learning and
big data to assist radiologists in detecting breast
cancer, analyze CT scans of the brain to exclude
acute diseases, and accurately measure pediatric
bone age. Their Bone Age analyzer has an accuracy
rate of +/- 4 months and returns results in
milliseconds―earning 16 Bit 1st place in the 2017
RSNA Machine Learning Challenge.
16 BIT.AI
RSNA BONE AGE
CHALLENGE WINNER
Try the 16 Bit algorithm at 16bit.ai/bone-age
4. 4
MRIs can take 20 minutes to 2 hours. Subsampled
data speeds scanning time but contributes to
inaccurate image reconstruction and diagnostics.
Researchers from the MGH/Martinos Center for
Biomedical Imaging and Harvard University are
working to speed MRI image reconstruction.
Powered by the NVIDIA DGX-1 AI supercomputer,
they’ve created AUTOMAP (Automated Transform by
Manifold Approximation).
AUTOMAP uses deep learning to reconstruct images
directly from sensor data using a cascade of densely
connected and sparse convolutional neural network
layers—it filters out noise and defects to
reconstruct images 100x faster and with 5x higher
accuracy to deliver more accurate diagnostic
outcomes.
AI IMPROVES
DIAGNOSTIC SPEED
AND ACCURACY
5. 5
Retinopathy of prematurity (ROP) affects preterm
babies born before 31 weeks, weighing <2¾ pounds.
It’s treatable if caught early but diagnosing the
severity of ROP is subjective—doctors compare the
infant’s retina with photos selected by experts in the
1980s.
Using a dataset of 6,000 images matched with expert
ROP diagnoses and a DGX-1 with cuDNN-accelerated
DL frameworks, researchers at Athinoula A. Martinos
Center for Biomedical Imaging trained a deep neural
network to differentiate ROP severity.
Still work in progress, this method could be deployed
in countries where access to specialists is lacking,
and help make a difference in reversing preventable
blindness worldwide.
AUTOMATING THE
DIAGNOSIS OF
INFANT BLINDNESS
6. 6
AI-ACCELERATED
CYBER DEFENSE
Our daily life, economic vitality, and national security
depend on a stable, safe and resilient cyberspace. But
attacks on IT systems are becoming more complex and
relentless, resulting in loss of information and money
and disruptions to essential services. Accenture’s
dedicated cyber security lab uses NVIDIA GPUs,
CUDA libraries, and machine learning to
accelerate the analysis and visualization
of 200M-300M alerts daily so analysts
can take timely action.
7. 7
SETTING THE
GOLD STANDARD
FOR NOTE TAKING
After downloading Otter from your Appstore, you can go
to your next meeting and focus 100% on the conversation
while your new assistant captures comments with clear
labels, and produces voice and text content you can
search and extract meaningfulinsights from.
This incredible work on smart speech recognition and
processing at over 90% accuracy involves terabytes of
data and complex speech AI models, made possible today
by the computationalstrength of NVIDIA Tesla GPUs and
formidable infrastructure on the Google Cloud.
8. 8
BREAKING DOWN
BUSINESS BARRIERS
The Alibaba Group services 8.3B translation
requests daily, enabling international
business trade. Alibaba uses neural machine
translation (NMT) which improves translation
quality significantly but drives up latency and
computation cost. To remedy this and
accelerate online NMT-based services,
Alibaba deployed NVIDIA Tesla GPUs
and achieved a 3x increase in
number of requests serviced
while cutting latency by
a factor of three.
9. 9
APPLYING STYLE
TRANSFER TO
3D TEXTURES
Using the popular texture synthesis and style
transfer technique, Artomatix is applying a
high-level of detail of one texture to a lower-
res texture which provides game artists
with a quick automated process to
generate usable textures with only
minimal user input.
10. 10
New drug development can cost billions and take up
to 14 years—and still only ~8% of drugs make it to
market. Atomwise helps its customers make smarter
decisions about which potential medicines to develop.
With NVIDIA GPUs to power training and inference,
Atomwise’s AtomNet deep learning software
understands the interactions of millions of molecules
and analyzes simulations to determine whether a
potential treatment works against a target.
AtomNet explored 8.2 million molecules and
identified several candidates that could prove to be
cures for Multiple Sclerosis. Effective in animal trials,
those candidates are now undergoing further
exploration.
SPEEDING THE
PATH TO MARKET
The Janus kinas 3 protein, which has been implicated in cancer and immune function. Image courtesy of Atomwise.
11. 11
SMARTER INSPECTION
SERVICES
In business, ensuring equipment uptime and
meeting safety and regulatory compliance is
non-negotiable. Using deep neural networks
developed on NVIDIA DGX-1 in the data center
that can easily extend to NVIDIA DGX Station in
the field, Avitas Systems delivers inspection
services using robotic-based autonomous
inspection and advanced data analytics.
In addition to safeguarding workers,
Avitas Systems AI solutions can
reduce inspection costs by 25%
and reduce maintenance
downtime by 15%.
The robots can handle the heat and use infrared cameras and chemical and other sensing
technologies to inspect assets under dangerous conditions and keep production running.
Image credit: Avitas Systems
12. 12
PUMPING AI INTO THE
OIL & GAS INDUSTRY
NVIDIA and Baker Hughes (BHGE) are using AI and
GPU-accelerated computing to help companies distill
oceans of data and reduce the cost of finding,
extracting, processing and delivering oil. BHGE’s
applied AI services and NVIDIA’s end-to-end AI
supercomputing solutions ―from the NVIDIA
DGX-1 in data centers, to the NVIDIA DGX
Station deskside or at remote locations,
to NVIDIA Jetson at the edge— can unlock
insights from data that was previously
as hidden as the oil underground.
13. 13
ACCELERATING
DISCOVERIES
WITH AI
New drugs typically take 12-14 years and $2.6
billion to bring to market. BenevolentAI is using
GPU deep learning for NLP to bring new therapies
to market quickly and more affordably. They’ve
automated the process of identifying patterns
within large amounts of research literature,
enabling scientists to form hypotheses and draw
conclusions quicker than any human researcher
could. And using the NVIDIA DGX-1 AI
supercomputer with CUDA, they identified two
potential drug targets for Alzheimer’s in less than
one month.
14. 14
If a picture is worth a thousand words, imagine
being able to skip the text and show your search
engine what you’re looking for. This is the powerful
experience Bing offers its users with automated
object detection for accurate search results
delivered quickly.
No more manually cropping the object you’re
interested in. Bing users simply click on the hotspot
that’s over the object of interest and the bounding
box automatically positions over the object and
triggers the search.
And the search is fast. With NVIDIA GPUs on Azure
cloud, Bing speeds up object detection 60X to
40ms—well under the threshold for an excellent
user experience.
SMARTER, FASTER
VISUAL SEARCH
Try Bing: https://www.bing.com/images/
15. 15
FASTER
PROCESSING,
HAPPIER
CUSTOMERS
The insurance industry still relies largely on
evidence-based, non-standardized documents
such as paper, scans, and photos for contract
management. Processing this type of
documentation is often manual, tedious, and
time consuming for the insurer and the insured.
‘Cardif Forward’ is BNP Paribas Cardif’s
innovative digitization plan and AI is a key
element. The company—known for leading-edge
customer service—is developing GPU-accelerated
deep learning image recognition algorithms to
automatically recognize and process documents
digitized by its clients. The AI solution will
significantly reduce the complexity of contract
management and speed the process.
16. 16
To speed advances in the fight against cancer, the
Cancer Moonshot initiative unites the Department
of Energy, the National Cancer Institute and other
agencies with researchers at Oak Ridge, Lawrence
Livermore, Argonne, and Los Alamos National
Laboratories. NVIDIA is collaborating with the labs
to help accelerate their AI framework called
CANDLE as a common discovery platform, with
the goal of achieving 10X annual increases in
productivity for cancer researchers.
AI PLATFORM TO
ACCELERATE
CANCER RESEARCH
17. 17
The number of mobile banking users is estimated to
reach 2 billion by the year 2021. So it’s not surprising
to see the rising popularity of Chatbots in the finance
industry. Through convenience and ease-of-use,
Chatbots optimize digital services at scale.
Capital One is piloting an SMS text-based intelligent
assistant named Eno. Eno uses GPU-powered deep
learning to respond to natural language text messages
from customers inquiring about their accounts.
Customers text Eno to track their balance, recent
charges, or to pay their bill. Eno takes mobile
banking to the next level, which is just a text
message away.
AI CHATBOT THAT
SPEAKS EMOJI
18. 18
“AUTO”MATIC
RECOGNITION AND
CATEGORIZATION
Online retailer Carsales handles ~20,000 uploaded
images each day. Having sellers input details on
each car and having Carsales staff manually
categorize photos was time-consuming and
inefficient. Carsales implemented Cyclops,
a GPU-powered AI tool that automatically
recognizes cars in photos, categorizes angles
and notifies sellers of missing angles and
poor quality photos. With 97% accuracy,
Cyclops exceeds human performance
and saves Carsales 55 hours/day
of valuable staff time.
19. 19
AI ADVANCES THE
FIGHT AGAINST
BREAST CANCER
Breast cancer is the second leading cause of cancer
death for women worldwide. Genomic tests help
doctors determine a cancer’s aggressiveness so
they can prescribe appropriate treatment. But
testing is expensive, tissue-destructive, and
takes 10-14 days. Case Western Reserve is
using GPU-based deep learning with
CUDA to develop an automated
assessment of cancer risk at
1/20th the cost of current
genomic tests.
20. 20
AI POWERED
INSIGHTS FOR
EFFECTIVE MEDICAL
DECISION MAKING
Electronic health records (EHR) offer huge volumes
of data that doctors can combine with real-time
analytics to improve their patient outcomes. Using
NVIDA GPUs with CUDA, CloudMedx is creating a
clinical AI platform that leverages both structured
and unstructured data fields from EHRs, reads
doctors’ patient-specific clinical notes and using
evidence based guidelines, helps identify risks and
correlates treatments. It’s a combination of
machine learning and natural language processing
that gives doctors the power to make time-critical
decisions that has the potential to save lives.
21. 21
SEEING BEYOND IMAGE
RECOGNITION
CloudSight democratizes the power of AI and
gives its users the ability to gain more
insights from their images. Its API is a
powerful visual cognition engine of >600m
images. Powered by the NVIDIA DGX-1 to
speed training and inference. CloudSight
accelerates deep neural net training runs
by 61x—enabling its engineers to
experiment freely and improve
services for its
customers.
22. 22
Weather forecasting involves processing vast
amounts of data to derive predictions that can save
lives and protect property. Colorful Clouds is using
GPU deep learning with CUDA to speed the
processing of data by 30-50x. It’s location-based
reporting tool can forecast and communicate
weather and air-quality conditions with high-
accuracy in real-time.
AI-POWERED
WEATHER
FORECASTING
23. 23
REINVENTING TV
WITH AI
Comcast and AI are improving the TV viewing
experience. With NVIDIA GPUs and the NVIDIA DGX-1
to power deep learning, Comcast parses millions of
voice commands received daily through its Xfinity
X1 platform. Each command is automatically
processed so customers find the content they
love in an instant. Add a recommendation
engine that suggests programming based
on individual preferences and Comcast
delivers a truly personalized
TV experience.
24. 24
AI INCREASES RETURN
ON ADVERTISING SPEND
In the Criteo commerce marketing ecosystem Big
Data is put into action throughout the purchase
journey. Criteo processes 600TB of shopper data
per day and serves >900B ads that drive $550B in
commerce sales per year. Criteo is piloting
GPU-powered AI for data analytics to better
understand buying habits and predictive
search to enable its customers to
propose the best product(s)
at the best time.
25. 25
REDEFINING
CYBERSECURITY
We depend on a safe cyberspace for just about
every aspect of our lives. Cyber attacks can be
devastating, and in today’s world mutations have
become the rule not the exception. Cylance
leverages GPU-powered deep learning with CUDA
to predict and prevent malicious code execution
by identifying indicators of an attack.
CylancePROTECT immediately prevented
the execution of the May 2017
WannaCry attack on 100%
of its customers’
endpoints.
26. 26
Even more than its meticulous engineering,
Mercedes-Benz is defined by its continuous
innovation. Since inventing the car in 1886, the
company has never stopped reinventing it. And
now Mercedes-Benz is using AI to enhance the
user experience behind the wheel by having its
cars predict where drivers want to go.
Trained on driver behavior data from 24,000 road
trips, the NVIDIA GPU-accelerated destination
prediction AI learns the driver’s habits over time
in order to make better suggestions.
AI-POWERED
DESTINATION
PREDICTION AND
ROUTE PLANNING
27. 27
ENABLING
AI SOLUTIONS
EDGE TO CLOUD
DarwinAI optimizes deep learning neural
networks ― making them faster, scalable,
portable and understandable. It’s Generative
Synthesis AI-driven technology observes the
inner workings of a neural network, then
generates an optimized version that is
orders of magnitude smaller and faster.
And, when combined with TensorRT
running on NVIDIA GPUs, DarwinAI
solutions run up to 7000 times
faster than CPUs.
RapidID enables real-time detection and
identification of objects in scenes on
embedded and mobile devices.
28. 28
DELIVERING GREATER
BUSINESS INSIGHTS
Insights from call center audio can help
companies increase sales, enhance employee
training and improve customer satisfaction,
but most companies are only able to
manually harness data from ~2% of their
recorded calls. Powered by NVIDIA DGX
Systems for deep learning training and
inference, Deepgram recognizes speech
more completely and precisely,
enabling companies to
utilize 100% of their
recorded calls.
29. 29
AI TOOL BOOSTS
CUSTOMER SERVICE
KLM’s 350+ social media service agents
handle 15K requests/week. To support the
volume of incoming messages, KLM uses GPU-
accelerated deep learning to predict the best
response. Service agents review and either
approve or personalize each response. The
resulting time savings allows agents to
focus on customers with more pressing
needs and handle more questions
while maintaining high levels of
customer satisfaction.
30. 30
The field of AI holds tremendous promise to
improve lives. Facebook A.I. Researchers
(FAIR) are advancing the field of machine
intelligence by creating new technologies that
give people better ways to communicate. To
manage the huge variety of projects, datasets,
and ever-changing workloads, FAIR needed to
update its research cluster. 128 NVIDIA DGXs
with CUDA are the main component of the new
cluster and deliver the extreme performance
and flexibility FAIR needs to advance AI.
128 NODE DGX-1
CLUSTER SCALES
DEEP LEARNING
INFRASTRUCTURE
31. 31
HUNTING “GHOST
PARTICLES” WITH
DEEP LEARNING
Tiny particles called neutrinos are the most
abundant form of matter in the universe and
understanding their properties is the focus of a
world-wide campaign of experiments. Observing
these ‘ghost particles’ in action requires
instruments of incredible size and scale. Fermilab’s
NOvA experiment applies two enormous detectors
with a total weight of 30 Million pounds spaced 500
miles apart. It is effectively one of the world’s
largest cameras, snapping 2 million images per
second and analyzing them for neutrino activity.
NOvA’s scientists developed deep neural networks
trained on NVIDIA GPUs with CUDA to improve the
machine’s detection rate by 33% - increasing the
discovery potential of NOvA and other large scale
experiments probing fundamental questions of the
universe.
32. 32
SPEEDING UP NASCAR
WITH AI
In NASCAR, the difference between victory and
defeat can be measured in milliseconds. That’s
why Ford Motorsports is optimizing aerodynamics
with AI powered by the NVIDIA DGX-1 running
TensorRT. Hours before a race, Ford trains its
model to recognize the field. During the race,
the AI analyzes video feeds and assesses
performance in real-time. Ford teams
receive immediate feedback and
adjust their cars as needed to
stay ahead of the
competition.
33. 33
Automakers use Computational Fluid Dynamics
(CFD) to help speed development, but Ford takes it
up a notch. The auto giant applies GPU-accelerated
AI to CFD to prepare its NASCAR racing teams for
optimal performance.
Augmenting traditional CFD, Ford developed a
virtual wind tunnel Neural Network on the DGX-1 to
simulate configurations of its racing teams’ cars and
predict how each car will perform under different
scenarios.
Simulations are 99% accurate and completed in a
few hours vs. 3-4 days, enabling the Ford teams to
adjust their vehicles before every race.
FORD & AI TAKE
NASCAR TO THE
FINISH LINE
34. 34
Health and life insurance administrators spend
countless hours manually reviewing statements,
physician notes, and other forms to correct any
discrepancies. Digitizing this paperwork is complex―it
involves extracting heterogeneous data from
documents and using multiple techniques to interpret
it accurately.
With NVIDIA Tesla V100 GPUs on AWS for training and
inference, the Friendly AI platform transforms medical
and insurance paperwork into a structured, analyzable
format and then uses natural language processing to
extract information to create digital claims.
Friendly clients process claims 40% faster (in minutes
vs. days) and with 20% fewer errors―equaling annual
cost savings of up to hundreds of thousands of dollars.
NEXT GENERATION
CLAIMS PROCESSING
35. 35
Customizing dental restorations to get the perfect
color, shape, and fit means happier patients and
fewer returns. Glidewell combines automation
with AI to speed the design process and meet a
demanding manufacturing workload of 10,000+
units per day.
Trained on >100k patient cases and 3D data,
Glidewell’s AI design automation tool uses a
convolutional neural network to accurately
determine if designs will produce products to
each patient’s satisfaction.
Additionally, patients no longer wait for weeks to
receive their custom restorations as production
time has been reduced to 2-3 days.
AI-BASED PRECISION
DENTAL CARE
36. 36
Heart disease, the world’s biggest killer, is responsible
for ~9 million deaths annually. Until recently, the best
test to diagnose heart disease was an angiogram―an
invasive and costly procedure.
HeartFlow applies GPU-accelerated deep learning to
the analysis of coronary blood vessels to deliver non-
invasive diagnostics. Trained on CT scans and
computational fluid dynamics, HeartFlow’s AI creates a
personalized 3D model of a patient’s coronary arteries
and analyzes the impact of blockages.
With HeartFlow, clinicians can provide personalized
treatment for each patient and improve quality of life.
It also means 61% of patients can avoid an angiogram,
reducing healthcare system costs by 26%.
HEART SMART:
AI TO DETECT
HEART DISEASE
Image is not representative of actual product. Courtesy of HeartFlow, Inc.
37. 37
THE AI BRAIN OF
SMART CITIES
A key technology component of any Smart City is
video surveillance. With millions of cameras in
large cities, analyzing the massive amount of
video content generated is a huge undertaking.
Hikvision is using deep learning and the
NVIDIA DGX-1 to speed intelligent video
content analysis for Smart Cities to
safeguard citizens and property,
manage traffic, and increase
criminal investigation
efficiencies.
38. 38
AI PREDICTS
AND PREVENTS
DISEASE
GPU deep learning is giving doctors a life-
saving edge by identifying high-risk patients
before diseases are diagnosed. Icahn School of
Medicine at Mount Sinai built an AI-powered
tool, “Deep Patient,” based on NVIDIA GPUs
and the CUDA programming model. Deep
Patient can analyze a patient’s medical
history to predict nearly 80 diseases up to 1
year prior to onset.
39. 39
AI CHANGES
THE GAME
Sports team managers have limited data.
ICEBERG’s automated AI system uses cameras
to capture millions of data points describing
hockey players’ positions, speed, plays and
tactics at every point in the game. With
this new level of data-driven understanding,
managers can make changes to their
game plan during periods to gain a
competitive advantage.
40. 40
SCALE SERVICES,
REDUCE TCO
Speech translation breaks language barriers for
travelers, businesses, students, and more.
When iFLYTEK wanted to scale its Mandarin
speech service to serve multiple accents and
dialects, the company expanded its use of GPUs
by moving its inference operations to Tesla
GPUs and TensorRT. iFLYTEK now handles
10x the number of concurrent requests,
has improved accuracy by 20%,
and yielded a 20% reduction
in operational
TCO.
41. 41
NEXT GEN AI-POWERED
IMAGE RECOGNITION
Fast and easy to implement functionalities are vital
in helping organizations enhance services. Imagga
helps developers and enterprises optimize projects
and quickly build scalable, image-intensive cloud
apps. With an API and solutions built on the
NVIDIA DGX Station to speed training and
inference, Imagga speeds service delivery
up to 88%, while maintaining high
levels of accuracy.
42. 42
IMAGINE NO
LIMITATIONS TO
DRUG DISCOVERY
It takes 12 years and $2.6B to bring a new drug to
market. Insilico Medicine takes an innovative approach
to speed drug discovery and uses GPU-accelerated
Generative Adversarial Networks (GANs) to "invent"
new molecular structures on demand. With GANs, in
only a few months’ time, Insilico did what would have
taken years with conventional research methods –
identified 69 new molecules. To date, Insilico has
generated 5,000+ new molecules, integrated GANs
into a comprehensive drug discovery framework, and
established working collaborations with large
pharmaceutical companies. This unique approach
could dramatically reduce the time and cost of
developing novel substances with medicinal properties.
43. 43
AI CAD TOOL
SPEEDS DESIGN
Design for Manufacturability (DFM) is the process of
designing products so they're easy to manufacture.
Traditional rule-based DFM relies on the experience
and training of individual designers which can vary
widely. Scientists at ISU are using CUDA on GPUs to
develop an AI DFM decision-support tool to help
designers optimize their models for manufacturability.
Trained on a database of CAD models, the tool predicts
non-manufacturability, identifies features that are the
cause, and provides recommendations for
improvement. Early results show the ISU tool can:
accelerate design and manufacturing cycles by
replacing manual iterative reviews; establish
consistency by eliminating inter-expert variability; and
speed a product’s time-to-market by >10x.
44. 44
DELIVERING FAMILY
FRIENDLY CONTENT
Growth in online video traffic means
companies are monitoring more content than
ever to filter inappropriate material. JD.com
uses NVIDIA’s DeepStream SDK and TensorRT
on Tesla P40 GPUs to identify and filter 1,000
channels of live-streamed full-HD videos. The
company has increased throughput 20x in
inference-based video content
filtering and is simultaneously
Inferencing 20 videos per
Tesla-equipped
server.
45. 45
SIGNED, SEALED
DELIVERED BY AI
JD X brings AI to logistics & delivery with
intelligent machines powered by the NVIDIA
Jetson supercomputer. The JDrone reduces
logistics fees by 70%, while delivering fresh
food & medicine to remote locations. The
JDrover robot easily navigates through
pedestrians & traffic to deliver packages to
select locations. JD X opened the world’s
first autonomous sorting center with
Pick & Place Robots that sort up to
16,000 parcels/hour with
99.999% accuracy.
46. 46
SHOPPING SMARTER
According to Forrester, the $390B E-Commerce
market will double by 2024. Jet.com (acquired by
Walmart) 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.
47. 47
REAL-TIME ANOMALY
DETECTION WITH MULTI-
SENSOR ANALYTICS
Frauds and intrusions in modern information
systems can cause significant harm. KickView’s
applied AI automates processing and anomaly
detection using signal data, image data, and
sensors such as video, radio frequency, and
IoT devices. With the NVIDIA DGX Station,
KickView reduced training time by 3x,
while maintaining >90% detection and
classifications accuracy with
multi-sensor systems,
all in real-time.
48. 48
The high cost of drug discovery is driving researchers
and pharmaceutical companies to turn to AI as a
faster, more efficient way to develop new drugs.
Professor Okuno, Kyoto University and RIKEN, have
formed the Life INtelligence Consortium (LINC) to
build an AI drug discovery ecosystem in Japan. LINC
uses the NVIDIA DGX-1 AI supercomputer―the DGX-1
delivers the extreme performance LINC needs to
solve complex problems and speed drug discovery.
DGX-1 POWERS
FASTER, MORE
EFFICIENT DRUG
DISCOVERY
49. 49
IDENTIFYING THREATS
Lawrence Livermore Labs (LLNL), a national AI project
leader, is using AI to make the world safer.
Understanding changing patterns of vehicles
or other objects over time may help identify
security concerns. Using NVIDIA GPUs, CUDA,
and deep learning to analyze overhead imagery,
LLNL has developed a “one-look” method
that counts objects directly, bypassing
the detection phase, while achieving
over 90% accuracy.
50. 50
AN AI POWERED
SUPPLY CHAIN
IN THE CLOUD
Machine Shop Services is a $43B industry according to
IBISWorld, and MakeTime — an online platform where
manufacturers in need of machining are matched
with pre-qualified shops — is poised to disrupt the
industry with a new data and cloud-driven approach
to machining.
MakeTime’s GPU-accelerated algorithms ensure
orders get placed with the best machine shop, every
time. With a nation-wide supplier network of 700+
computer numerical control (CNC) machine shops,
MakeTime brokers new win-win relationships
for manufacturers and their suppliers — shop owners
fill gaps in machine schedules and manufacturers
speed time to execution.
51. 51
AI DETECTS
GROWTH PROBLEMS
IN CHILDREN
Detecting growth-related problems in children
requires calculating their bone age. But it’s an
antiquated process that requires radiologists to
match X-rays with images in a 1950s textbook.
Massachusetts General Hospital, which conducts
the largest hospital-based research program in
the United States, developed an automated
bone-age analyzer built on the NVIDIA DGX-1
with CUDA. The system is 99% accurate and
delivers test results in seconds versus days.
52. 52
NOT ENOUGH DATA?
NOT A PROBLEM
Deep Learning holds enormous promise to
advance medical discoveries, but adequate
training data can be a challenge. Scientists
at the MGH & BWH Center for Clinical Data
Science are using the NVIDIA DGX Station
to train GANs that create and validate
synthetic brain MRI images. Combining
the manufactured images with real
MRI images enables the team
to train its neural network
with 75% less data.
53. 53
AI HELPS PERSONALIZE
IMMUNOTHERAPY
Immunotherapy has a success rate of only 40% and a
risk that it may attack healthy cells. Max Kelsen is
using sophisticated AI approaches with NVIDIA V100
GPUs to integrate genomic, transcriptomic
and patient information to identify
a classifier and develop a test
that can predict treatment
response.
54. 54
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.
55. 55
Arctic sea ice is melting as the Earth warms, but
instead of making seafaring safer, melting ice can
unexpectedly block channels, damage vessels and
cargo, or completely trap ships. To make seafaring
safer, scientists at Memorial University Newfoundland
are using GPU-powered AI with CUDA to predict when
and where sea ice is likely to melt and refreeze.
Unlike current forecast methods, which look ahead
7-10 days, the team is working to predict conditions
up to 6 weeks ahead for ice, and 6 months ahead for
icebergs. In addition to safeguarding lives and
livelihoods, the AI forecasting tool will help shippers
make the most of the short season.
ON THIN ICE:
HOW AI PREDICTS
MELTING OF SEA ICE
56. 56
AI TOOL LETS YOU
APPLY BEFORE
YOU BUY
Testing different types of makeup can take hours
and be a frustrating experience. ModiFace is using
GPUs with CUDA and facial modeling technology to
help consumers explore and select the ideal
products. ModiFace developed the ‘Sephora Virtual
Artist’, an online tool that allows consumers to
virtually experiment with new makeup without
having to leave their computer screen. With
technology on skin analysis and facial visualization,
ModiFace and its AI features have introduced a more
efficient way to style oneself.
57. 57
SUPER DRUGS TO
COMBAT SUPER BUGS
New Robot Arm from ARC
Image: One of Dr Matt Belousoff and Professor Trevor Lithgow ribosome structures illustrating complicated details
that can be determined using MASSIVE and the Titan Krios at the Ramaciotti Centre for Cryo Electron Microscopy.
In the race to design more effective drugs and new
treatments for diseases, Australian scientists are using
HPC and advanced imaging to visualize changes in
ribosomes that occur in response to antibiotics. With
MASSIVE’s M3 supercomputer powered by >160 NVIDIA
GPUs and 2 DGX-1V’s to accelerate data processing,
the team strives to identify new drugs that
are lethal to bacteria despite
structural changes in
ribosomes.
58. 58
AN AI MONITOR OF
EARTH’S VITALS
The Earth’s climate has changed throughout
history, but in recent years there have been record
increases in temperature, glacial retreat and rising
sea levels. NASA Ames is using satellite imagery to
measure the effects of carbon and greenhouse gas
emissions on the planet. To do so, they developed
DeepSat―a deep learning framework for satellite
image classification trained on a GPU-powered
supercomputer. The enhanced satellite imagery
will help scientists plan to protect ecosystems and
farmers improve crop production.
NASA: Late summer 2016, forest fires in Africa produce plumes of CO2
Left: CO2 - 10/14/2016 / Right: CO2 - 12/24/2016
Source: https://climate.nasa.gov/climate_resources/142/
59. 59
DEFENDING
THE PLANET
The U.S. government’s Asteroid Grand Challenge
seeks to identify asteroid threats to human
populations. The team at NASA Frontier
Development Labs picked up the challenge
by employing GPU deep learning with CUDA
to identify threats and their unique
characteristics. The resulting “Deflector
Selector” achieved a 98% success rate
in determining which technology
produced the most successful
deflection.
60. 60
“SEEING” GRAVITY
FOR THE FIRST TIME
In September 2015, 100 years after Einstein
predicted them, gravitational waves were
observed for the first time. Astronomers at
the Laser Interferometer Gravitational-wave
Observatory have since used GPU-powered
deep learning to process gravitational wave
data 100x faster than previous methods,
making real-time analysis possible and
putting us one step closer to
understanding the
universe’s oldest
secrets.
Physics Letters B - Deep learning for real-time gravitational wave detection
and parameter estimation: Results with advanced LIGO data
Daniel George, E.A. Huerta
61. 61
A RADIOLOGY LIBRARY
FOR THE WORLD
AI holds enormous promise for advancing medical imaging,
but well-annotated datasets are hard to come by.
Researchers at the National Institutes of Health
have created an auto-annotation system leveraging
deep learning, NVIDIA GPUs and the CUDA
programming model. The NIH research
could lead to the creation of a
global library of datasets for
medical researchers.
62. 62
GPU Job Management System
AI CAMERA HELPS USERS
TAKE BEAUTIFUL PHOTOS
EVERY TIME
PHOTO BACK-UP CLOUD
RETRIEVES PHOTOS
IN AN INSTANT
NTT DOCOMO ACCELERATES DEVELOPMENT OF AI WITH
NVIDIA DGX-1, TESLA GPUS, JETSON TX-1
AI CONCIERGE FINDS
FASHION ITEMS TO
SUIT YOU
TAXI DEMAND AI
FORECASTING TOOLTELLS
DRIVERS WHERE THEY’RE
NEEDED MOST
63. 63
AI ALARM SYSTEM
SAFEGUARDS
HONEYBEES
Habitat loss and pesticides are the primary
threats to bee populations, but the Varroa mite
can devastate entire colonies. To combat the
Varroa, student Jade Greenberg turned to AI.
Her solution —NVIDIA’s Jetson TX2, DGX
Station, TensorRT, Microsoft’s Cognitive
Toolkit, Kinetica— uses sensors and
cameras to feed a convolutional neural
network that assess hive health in
real-time and converts the data
into a visual early warning
system for beekeepers.
Image courtesy of Piscigate
AI detects a mite in frame 24
64. 64
THE NEW
SCIENCE OF SPORTS
Predictive analytics, commonly used in business to
identify risks and opportunities, is increasingly
used by the sports industry to tap into massive
amounts of data. Scientists at NYU are applying
deep learning and the NVIDIA DGX-1 AI super-
computer to analyze unprecedented amounts
of Major League Baseball data —four
years-worth of every player’s every
move— to help improve
the game.
65. 65
A 21ST CENTURY
PLANNING TOOL
BUILT ON AI
With the Earth's population at 7 billion and growing,
understanding population distribution is essential to
meeting societal needs for infrastructure, resources
and vital services. Using GPUs and deep learning,
Oak Ridge National Laboratory quickly processes
high-resolution satellite imagery to map human
settlements. With the ability to process a
major city in minutes, ORNL can provide
emergency response teams critical
information that used to take
days to create.
66. 66
AI IMPROVES
THE CUSTOMER
EXPERIENCE
AI is dramatically changing the online shopping
experience with tangible improvements to retailers
and consumers. In 2016 online British grocery giant
Ocado improved customer service with their AI-
enhanced contact center, and is applying machine
learning and NVIDIA GPUs with CUDA to develop
humanoid robotics to assist maintenance
technicians, and advanced computer vision for
image classification and recognition to replace
barcode systems. Computer vision will expedite the
picking process and better ensure orders are filled
correctly so customers receive exactly what they
ordered.
67. 67
The phrase “Time is Brain” means every minute
counts after a stroke. A typical patient loses
almost 2 million neurons per minute in which a
stroke is untreated. Immediate treatment
minimizes brain damage.
To help Radiologists diagnose the most urgent
cases and enhance critical care, the OSU
Department of Radiology used GPU-accelerated
deep learning to develop an Automated Critical
Test-Findings Identification and Online Notification
System (ACTIONS). With GPUs, ACTIONS was
trained in minutes vs. days. It identifies in seconds
the most urgent cases of stroke, hydrocephalus,
hemorrhage, and large tumors with an accuracy
rate of 81% (stroke) and 91% (hydrocephalus,
hemorrhage, large tumors), speeding time to
critical care.
AI SPEEDS TIME
TO CRITICAL CARE
Examples of head CT examinations containing critical findings.
A) A patient with a recent stroke involving the left cerebral hemisphere (green arrows).
B) A patient with a large left frontal tumor compressing adjacent structures (orange
arrows).
http://pubs.rsna.org/doi/pdf/10.1148/radiol.2017162664
68. 68
PERSONALIZING
PRODUCTS
Olay is arming consumers with knowledge to make
informed purchase decisions. Its Olay Skin Advisor—a
GPU-accelerated AI tool that works on any mobile
device—assesses a user-provided selfie and
advises how to improve trouble areas using
a daily regime of recommended Olay
products. After four weeks 94% of
Skin Advisor users continued
to use the products
it recommended.
69. 69
B O X B A G
D E N I M J A C K ET
A VIATOR SUNGLASSES
L O O K: S T REE T
L O O S E F I T
W A S H E D
B R O W N
S W E A T P A N TS
B L A C K
W H I T E
S T R I P E
L E A T H E R
M E T A L L I C
AI TAGGING API
IS ON TREND
Fashion moves and changes quickly. To help its B2B
customers stay ahead of the fashion curve, Omnious
offers an AI tagging API that reduces manual work.
Powered by a DGX Station, the automated fashion
image tagging tool is 100x faster than manual
tagging, reduces 90% of operation costs,
and is more accurate than human
labelling by fashion experts.
70. 70
TEACHING
COMPUTERS TO
SPEAK HUMAN
Natural language processing, a research area
of AI, will help bridge the worlds of humans
and machines. OpenAI, a non-profit founded
by Elon Musk to ensure AI is safe, is using the
NVIDIA DGX-1 AI supercomputer with CUDA to
advance its research, including developing an
AI agent with natural language understanding.
With DGX-1, OpenAI researchers can explore
problems that they couldn't previously pursue
and, ultimately, advance their mission to
widely distribute the benefits of AI.
71. 71
AI HELPS DOCTORS
DIAGNOSE
BREAST CANCER
Every day, pathologists are tasked with providing
cancer diagnosis to guide patient treatment.
However, sifting through millions of normal cells
to identify a few malignant cells is extremely
laborious using conventional methods. PathAI
combines GPU deep learning with traditional
pathology to improve accuracy,
speed diagnosis, and
reduce error rates
by 85%.
72. 72
Understanding protein structural variability and
disorder is paramount for advancements in protein
applications and drug design. Researchers at Peptone
recently unleashed the power of big data and AI to
understand protein structural variability at the
building block level through statistical analyses of
protein NMR data. Peptone’s dSPP, is the world’s first
interactive repository of structure features of
proteins for the next generation machine learning
problems with seamless integration for Keras and
Tensorflow frameworks.
Researchers harnessed the computational power of
the DGX-1 with CUDA to unravel the sequence-
dynamics relationships in 7200+ proteins of medical
significance through Bayesian Deep Learning and
Hybrid Statistical Thermodynamics.
AI ACCELERATES
PROTEIN RESEARCH
73. 73
DISCOVER MORE
WITH DEEP
LEARNING
Online shopping can be convenient but searching
through multiple websites can be arduous and time-
consuming. Pinterest makes it easy for users to
quickly discover things they love. Automatic object
detection lets users search for products within a Pin’s
image, and Shop the Look lets users buy items seen
in fashion and home décor Pins. Scientists on
Pinterest’s visual search team use GPU-accelerated
deep learning with CUDA to teach their system to
recognize image features using a dataset of billions of
Pins and compute similarity scores to identify the best
matches. One visual search study reports a 50%
improvement in user engagement and traffic.
74. 74
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
damage multi-billion machines. Current models
can predict the disruptions with 85% accuracy,
but ITER will need something more precise.
Researchers at Princeton University have
developed the Fusion Recurrent Neural Network
(FRNN) using deep learning and NVIDIA GPUs with
CUDA 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. FRNN has achieved 90% accuracy and is
on the path to achieving its goal of 95% accuracy
for ITER’s tests.
Visualization courtesy of Jamison Daniel,
Oak Ridge Leadership Computing Facility
75. 75
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.
76. 76
The Research Computing Centre (RCC) at UQ are using the
Wiener supercomputer to expedite research in a diverse
range of imaging-intensive sciences. Using deconvolution
algorithms, machine learning and pattern recognition
techniques, Wiener —with its NVIDIA Tesla V100
accelerators— provides near real-time outputs
of deconvolved, tagged and appropriately
characterized data, providing researchers
with immediate feedback on data
quality and allowing for faster
interpretation of
microscopy data.
UQ's new Wiener supercomputer will add sharp detail to microscopic imagery
FASTER INTEPRETATION
OF MICROSCOPY DATA
77. 77
AI ROBOT
WASTES NOT,
WANTS NOT
Humanity produces ~1.3B metric tons of waste
a year. Most ends up in landfills. Much of it
could be recycled, but the process of sorting
and recycling waste material is often cost
prohibitive for manufacturers. Sadako is
turning trash into cash with its AI-powered
robot. With NVIDIA GPUs powering machine
learning, the Max-AI robot removes
recyclable materials from
the waste stream
cost-effectively.
Inset upper right corner: Max-AI
combines computer vision and AI
to identify recyclables
78. 78
IDC predicts that by 2021 AI will boost global
business revenue by $1.1 trillion. Businesses are
embracing AI to optimize processes, decrease
costs, and remain competitive.
SalesHero's AI assistant, Robin, performs tedious
sales tasks that detract from productivity. Using
GPUs for deep learning and inference, Robin learns
from CRM systems, customer interactions, and
SalesHero’s Account Graph database to automate
sales processes such as updating sales records,
prospect mining, and predicting close dates. With
the time savings Robin delivers, sales teams can
increase productivity up to 30%.
AI SALES ASSISTANT
INCREASES
PRODUCTIVITY
79. 79
A NEW WAVE OF AI
BUSINESS APPS
Many companies sponsor televised events to
promote their brands, yet ROI can take weeks,
or even months, to measure. SAP Brand
Impact, powered by NVIDIA deep learning,
measures brand attributes in near real-time
with superhuman accuracy. With deep neural
networks trained on the NVIDIA DGX-1, and
with the TensorRT inference engine,
SAP improves performance by 40X,
reduces hourly costs by 32X, and
delivers immediate, accurate,
and auditable
results.
80. 80
CONTROLLING AIR
TRAFFIC WITH AI
From autopilot systems to customer service to
predicting weather, AI is transforming aviation. With
Aimee—a GPU-powered framework for AI solutions
from Searidge Technologies—Air Traffic Control no
longer needs a direct sightline. Aimee analyzes
video feeds from hundreds of cameras, enabling
ATC to look past occlusions and “see” every
runway, taxiway, tarmac, and gate without
looking away from their workstations.
81. 81
AI HELPS CREATE
BEAUTIFUL MUSIC
Music plays an impactful role in the experience of
interactive media, but it’s challenging for composers
to produce music for dynamic, user driven situations.
Researchers at SensiLab used the NVIDIA DGX-1 to
develop a new deep learning framework for
adaptively scoring interactive media
that’s been successfully
implemented in two
video games.
82. 82
NAME THAT TUNE
IN REAL-TIME
Shazam maximizes music recognition with
NVIDIA GPUs on Google Cloud for intensive
calculations and fast throughput. Each time
someone Shazams a song, they send an
acoustic fingerprint of the audio that their
device recorded. Shazam uses that
fingerprint and GPUs for the intensive
operation of searching for and instantly
matching songs in its catalog of
>11 million songs.
83. 83
ACCELERATING IVA
FOR SMART CITIES
Intelligent video analysis (IVA) can safeguard citizens
and property and is a key element of smart cities but
analyzing data from millions of cameras in real-time
requires deep learning and intensive computing
power. SK Telecom uses NVIDIA GPUs to power
T View, its AI VSaaS (Video Surveillance as a
Service) solution. With Tesla GPUs, SKT
speeds training 5x, and with TensorRT
to scale its inference engine, SKT
achieves cost-efficiencies
without sacrificing
accuracy.
84. 84
AI SHEDS LIGHT
ON MYSTERIES OF
THE UNIVERSE
Gravitational Lensing generates an image of a
distant light source that is distorted by the gravity
of a massive object, such as a galaxy cluster. The
distortions provide clues about how mass is
distributed in space―and how that distribution
changes over time―to measure properties of dark
matter, galaxy size, and the expansion of the
universe. Today there are roughly 200 known
gravitational lenses and scientists expect to uncover
another 200,000 over the next decade. Analyzing a
single lens image with traditional approaches can
take 2 days to 3 months.
Scientists at SLAC are using GPU-powered artificial
neural networks to analyze gravitational lenses in
just 10 milliseconds. This speed up provides
researchers with opportunities for new discoveries
to shed light on the mysteries of our universe.
85. 85
The human retina contains diagnostic markers for
many diseases, but testing requires the skills of
specialists who are in short supply. Built on GPU
deep learning with CUDA, the Mobile Autonomous
Retinal Evaluation (MARVIN), from SocialEyes, can
help transform healthcare systems worldwide. With
MARVIN, tens of millions of community healthcare
workers and physicians can diagnose a wide range
of conditions immediately with low-cost mobile
devices for timely and effective intervention.
AI-POWERED
HEALTHCARE
AT SCALE
86. 86
With over 1B people living in extreme poverty,
ending poverty tops the list of the UN’s sustainable
development goals. But data that identifies
impoverished areas is scarce. Researchers at
Stanford University’s AI Lab used GPU deep learning
with CUDA and satellite imagery to map areas of
extreme poverty. Not only is the approach
effective, scalable and cost efficient, the poverty
maps will help world organizations locate those
most in need of relief.
MAPPING AN END
TO POVERTY
WITH AI
87. 87
Skin cancer is the most common form of all cancers.
The 5-year survival rate of melanoma is ~98% when
detected early, but drops to <20% when the disease
is detected in its later stages.
Scientists at Stanford University’s AI lab are using
dermatology photos to help diagnose skin cancer.
Using GPU-powered deep learning and 130K images
representing >2K skin diseases, the team trained a
convolutional neural network to recognize cancerous
lesions.
When tested, the AI tool was 90% accurate in its
diagnosis and could help clinicians speed detection
and diagnosis—key to a skin cancer patient’s
outcome.
AI TOOL
HELPS DIAGNOSE
SKIN CANCER
Image credit: Matt Young
88. 88
In fashion, styles change quickly but the fundamental
customer experience—brick-and-mortar stores and
traditional online shopping sites—hasn’t changed much
in the past decade. Stitch Fix broke that mold with a
fashion styling service that combines the art of
personal styling with data analytics insights powered
by GPU-accelerated deep learning.
Backed by a team of 70+ data scientists, Stitch Fix
builds style recommendation algorithms to make its
clients look their best.
Take the Stitch Fix algorithms tour:
http://algorithms-tour.stitchfix.com/
REINVENTING RETAIL
BY COMBINING
ART AND AI
89. 89
AI UNLOCKS
SCIENTIFIC MYSTERIES
Gravitationally lensed galaxies ―a prediction of the
General Theory of Relativity― are rare and lie amongst
billions of galaxies. Swinburne used astrophysical
simulations to train a GPU-powered CNN to recognize
gravitational lenses in astronomical image data,
then applied those CNNs to the Dark Energy
Survey image data to discover dozens of
new gravitational lenses.
90. 90
AI: THE TICKET TO
DELIVERING WORLD-
CLASS LOGISTICS
The Swiss Federal Railway (SBB) system connects all
European railways with 300+ tunnels, 6,000 bridges,
80+ trains, and 30,000 switches. To ensure seamless
logistics, SBB turned to simulation. Powered by the
NVIDIA DGX-1 and its integrated software stack,
SBB now simulates the physics of all train
traffic in Switzerland for one day
in just 0.3 seconds.
91. 91
AI IS ON TRACK TO
SAFEGUARD RAILWAY
INTEGRITY
To maintain the integrity of its 3,232 km of tracks,
the Swiss Federal Railways (SBB) runs diagnostics
trains to photograph and monitor tracks in real-
time. But traditional data processing methods
produce false positives/negatives. To remedy
this, SBB and CSEM (Swiss Research and
Development Center) launched the
Railcheck project which applies deep
learning, powered by the NVIDIA DGX
Station, to improve the automatic
detection and classification
of faults.
92. 92
In 2003 the Human Genome Project successfully
decoded the human genome and unlocked the door
to new genetic discoveries. With 3 billion nucleotide
pairs in the human DNA, genome analysis is
computationally intensive. The Tohoku Medical
Megabank Organization (ToMMo) is using the power
of its DGX-1 AI supercomputer cluster to accelerate
understanding the complicated correlations
between human genotype and phenotype.
And, to further deep learning based genomics
research, ToMMo will open its DGX-1 supercomputer
cluster to external contracted researchers.
DGX 3 NODE CLUSTER
TO ADVANCE
GENOMIC RESEARCH
93. 93
According to the World Health Organization, TB is
one of the top 10 causes of death worldwide. 1.7M
people died from the disease in 2016 with 95% of
those deaths occurring in developing countries
where access to radiological expertise is limited.
Armed with >1,000 TB images, NVIDIA GPUs, Caffe,
CUDA, and cuDNN, scientists at Philadelphia’s
Thomas Jefferson University trained a deep learning
model to read chest x-rays.
With GPUs delivering a 40x increase in speed up over
CPUs, the research could expand to include other
lung diseases and possibly lead to the development
of a centralized global chest x-ray library for
healthcare providers in developing countries to use
to accurately diagnose anomalies.
FIGHTING TB WITH
GPU-POWERED AI
Image credit: Yale Rosen. Licensed via Creative Commons 2.0
94. 94
It takes an average of 12 years and >$2B to bring a
new drug to market. Molecular dynamics (MD)
simulation is a powerful tool to calculate potential
efficacy. One key to accelerating drug discovery is
the ability to run more MD simulations but this
requires substantial computing power.
Tokyo Tech’s Smart Drug Discovery Research Unit is
using its GPU-powered TSUBAME supercomputer to
accelerate drug discovery through massive MD
simulation. With GPUs, Tokyo Tech has achieved a
>50x speedup as compared to CPU-powered
simulations, and they’ve already discovered a new
drug candidate for Chagas disease, one of the
Neglected Tropical Diseases.
ACCELERATING
DRUG DISCOVERIES,
IMPROVING LIVES
Tokyo Institute of Technology TSUBAME3.0 Supercomputer
95. 95
OPTIMIZED LOGISTICS
MANAGEMENT
The United States Post Office delivers >1.5 billion
pieces of mail each year. To analyze and optimize
routes of 200,000+ mail carriers in real-time the USPS
turned to Kinetica and its GPU-accelerated data
analytics platform. The USPS can now modify
routes for optimal efficiencies—it’s using
fewer trucks, handling more
deliveries, and narrowing
delivery windows.
96. 96
TEACHING A
ROBOT TO STAND
UP FOR ITSELF
New approaches to AI promise to help
scientists build machines with greater
autonomy. Researchers at UC Berkeley are
tapping into the processing power and
integrated software of NVIDIA’s DGX-1 to
advance robotics using reinforcement
learning. DGX-1 with the CUDA programming
model allows them to iterate faster and
ultimately build robots that are able to
understand and navigate a diverse and
changing world on their own.
97. 97
ACCELERATING AI
WITH GAME-BASED
VIRTUAL WORLDS
Deep Learning speeds up autonomous driving
research, but the human task of keeping up with the
image data can slow it down. Inspired by Grand Theft
Auto V, scientists at the University of Michigan are
replacing the tedious process with a simulation
engine that rapidly generates annotated data in a
game-based virtual world. The technique is powered
by NVIDIA DGX-1 with CUDA and blows the doors off
traditional approaches. Deep learning projects that
took weeks now take days, and Michigan’s scientists
can now focus less on note-taking and more on
breakthroughs in self-driving car technology.
Image courtesy of Les Nuits Photographiques
98. 98
Molecular energetics studies can lead to
breakthroughs in drug discovery and materials
science, but traditional computing approaches are
time-consuming and expensive. Researchers at the
University of Florida and the University of North
Carolina leveraged GPU deep learning and CUDA to
develop ANAKIN-ME, which can reproduce
molecular energy surfaces with super speed,
extremely high accuracy, and at 1-10/millionth the
cost of current computational methods.
AN AI QUANTUM
BREAKTHROUGH
99. 99
AI PRODUCES THE 5TH
STATE OF MATTER
Bose-Einstein Condensate (BEC) is a state of matter
formed by cooling a gas to near-zero absolute
temperatures. BEC matter creation has proven
useful in exploring superconductive material
breakthroughs and creating extremely precise
measurements of the Earth’s gravity. Researchers
at the University of New South Wales used
GPU-powered AI to create BEC matter
14x faster than conventional
methods.
100. 100
AI ACCELERATES
ASTEROSEISMOLOGY
Light years away, aging stars are blazing into a fiery
stage of life as red giants. Classifying the evolutionary
stage of red giants was a slow manual process, so
scientists from Australia and Denmark trained a
GPU-powered convolutional neural network to
learn visual features of red giants and predict
a star’s age. Tests showed the AI system
classified 7,655 red giants in real-time
and achieved 99% accuracy.
101. 101
AI ACCELERATES
DRUG DISCOVERY
The discovery phase of drug development involves
exploring different possible combinations of
protein molecules (targets) and drug chemical
compounds to ensure the drug will do what it’s
designed to do. Classic Molecular Dynamics
simulations are time-consuming and expensive.
Machine Learning models help predict probability
of the target molecules interacting with the drug
chemical compounds, but still require significantly
greater performance to deliver improved accuracy.
Researchers at the University of Pittsburgh are
improving model performance and prediction
accuracy. Their convolutional neural network,
accelerated with NVIDIA GPU’s and CUDA,
improved prediction accuracy from ~52% to 70%
compared to other machine learning-based models.
102. 102
Studying the size, shape, age and location of
moon craters provides insight into the history of
our solar system.
Counting and determining characteristics of
craters was a manual process until researchers
at the University of Toronto and Penn State
University developed a convolutional neural
network, powered by NVIDIA Tesla P100 GPUs on
the SciNet P8 supercomputer, to detect and
classify characteristics of craters from Lunar
digital elevation maps.
Upon implementation, the AI system identified
6,000 new craters in just a few hours—orders of
magnitude faster than human counting—and is
now being applied to study craters on Mercury.
MAPPING THE
MOON’S CRATERS
Image courtesy of NASA
103. 103
AI-enabled transformations such as autonomous
vehicles, personal assistants, and medical
breakthroughs can greatly benefit society, but
demand for applied AI is growing faster than the
talent pool.
UnternehmerTUM is on a mission through its
Applied.AI Initiative to accelerate the delivery
of AI solutions by educating and connecting
talent with state-of-the-art technology &
industry companies. The government-backed
initiative—which expects 3,000 participants and
>30 new AI startups its first year—has selected
the NVIDIA DGX-1V and DGX Station with CUDA,
and the Deep Learning Institute to realize its
vision for the Applied.AI Initiative as the leading
innovation hub for AI in Germany and one of the
top three centers in the world.
ACCELERATING
THE DELIVERY
OF AI SOLUTIONS
104. 104
EXTRACTING NEW
VALUE FROM VIDEO
Imagine being able to quickly find any scene
in any video. Valossa AI provides unparalleled
capabilities to capture new value from video
with advanced audio-visual content search
and recognition. Powered by the NVIDIA GPU
hardware and software stack for deep
learning and inference, Valossa cut training
time from weeks to hours and processes
videos 30X faster and more accurately
than CPU-based methods, enabling
a new generation of video
analytics and
monetization.
105. 105
BETTER DATA,
SMARTER BUILDINGS
Verdigris is on a mission to help businesses eliminate
wasteful energy spend. By harnessing the power of
data and GPU-powered deep learning, Verdigris’
Smart Building optimization solutions continually
audit and analyze electronic signatures of
individual devices to learn what's normal
and what’s energy waste. Furthermore,
with real-time monitoring and alerts,
operations teams can proactively
respond before a situation
becomes problematic.
106. 106
THE BRAINS BEHIND
SMART CITIES
Verizon’s Smart Communities Group is on a
mission to make cities safer, smarter and
greener. Using NVIDIA Metropolis, an edge-
to-cloud video platform for building smarter,
faster AI-powered applications, Verizon is
working to collect and analyze multiple
streams of video data to improve traffic
flow, enhance pedestrian safety,
optimize parking
and more.
107. 107
REAL-TIME SPEECH
SERVICES AT SCALE
WeChat, a leading Chinese social media
platform with ~1B users, wanted to improve
its speech to context services. But as the
company deployed its new acoustic model,
its CPU-only servers were unable to
effectively run the new version. WeChat
deployed servers equipped with Tesla P4
GPU inference accelerators and increased
speech inference throughput by 2.5X and
in-model accuracy by 20%—all
while staying within its
low latency
budget.
108. 108
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
109. 109
Wildbook is protecting endangered animals by
blending wildlife research with AI, citizen science,
computer vision, and data analytics to speed
population analysis and develop new insights.
Wildbook harnesses publicly shared videos,
photos, text and audio, and with the power of
GPUs speeds up the task of identifying animals of
all kinds.
GPUs reduce the time it takes to not only detect
species but each individual animal, from a few
seconds per image to a fraction of a second—a
significant time savings as Wildbook deals with
thousands of images per species.
USING AI TO
COMBAT
EXTINCTION
110. 110
AI DEMOCRATIZES
MOTION CAPTURE
Motion capture has revolutionized the entertainment
industry, but traditionally has required specialized
equipment and expertise. wrnch democratizes
motion capture with AI that makes any camera
capable of digitizing human movement. wrnch’s
GPU-powered deep learning software recognizes
human motion in 2D RGB video, then creates 3D
motion data. With NVIDIA TensorRT, wrnch’s AI
performs 2x faster during inference—it
Delivers results in real-time and
Creates endless possibilities for
new applications.
111. 111
AI-BASED NEXT
GEN WEATHER
FORECASTS
Improving the speed and accuracy of weather
forecasts is big business, with the potential to
save billions across global supply chains,
commercial air travel, agriculture and more. To
create the world’s first widely available
‘nowcasting’ service, researchers at Russia’s
online search engine Yandex created a prediction
model trained on more than 800,000 sequences
of Russian radar data and satellite images.
Applying NVIDIA GPUs with CUDA to its
convolutional neural network sped its system’s
learning by more than 40% - the speed they
needed to alert users to sudden rainstorms just
10 minutes away.
112. 112
MODERNIZING THE
WAREHOUSE
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 with CUDA 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.
113. 113
The demand for medical imaging services is
continuously increasing, outpacing the supply of
qualified radiologists and stretching them to
produce more output, without compromising
patient care. It’s not atypical for hospitals to have
a large backlog of x-rays waiting to be routed.
Zebra is using GPU-powered AI to augment the
capabilities of radiologists. Its low-cost AI1
assistant instantly detects diseases of the lung,
breast, liver, cardiovascular system, and bones to
help radiologists manage the ever increasing
workload while continuing to deliver quality care.
AI TRANSFORMS
PATIENT CARE
114. 114
DEVELOPING
THE VEHICLES
OF THE FUTURE
Zenuity, a joint venture of Volvo and Veoneer,
aims to build autonomous driving software for
production vehicles by 2021. They chose to
build their deep learning infrastructure
with NVIDIA DGX-1 servers and Pure
FlashBlade systems to accelerate
their AI initiative.