Bharatkumar Sharma, Senior Solution Architect
AI IN HEALTHCARE AND RETAIL
AGENDA
1. Where I come from?
2. AI and Deep Learning
3. AI in Healthcare
4. Intelligent Video Analytics in Retail
3
NVIDIA
The AI Computing Company
TRANSPORTATION
HEALTHCARE
MACHINE LEARNINGHPC DEEP LEARNING
GAMING
DESIGN
4
1980 1990 2000 2010 2020
GPU-Computing perf
1.5X per year
1000X
by
2025
RISE OF GPU COMPUTING
Original data up to the year 2010 collected and plotted by M. Horowitz, F. Labonte, O. Shacham, K.
Olukotun, L. Hammond, and C. Batten New plot and data collected for 2010-2015 by K. Rupp
102
103
104
105
106
107
Single-threaded perf
1.5X per year
1.1X per year
APPLICATIONS
SYSTEMS
ALGORITHMS
CUDA
ARCHITECTURE
5
AI REVOLUTION
Big Data GPU AccelerationBetter Algorithms
350 million
images uploaded
per day
Petabytes of
customer data
hourly
300 hours of video
uploaded every
minute
“The Three Breakthroughs that have
Finally Unleashed A.I. on the World”
6
EXPERT SYSTEMS
EXECUTE HAND-WRITTEN
ALGORITHMS AT HIGH SPEED
Accelerate with
GPU accelerated Libraries
OpenACC Directives
CUDA Kernels
INCREASING COMPLEXITY AND AUTONOMY OVER TIME
EXPERT SYSTEMS
EXECUTE HAND-WRITTEN
ALGORITHMS AT HIGH SPEED
THREE ROADS TO AI
Three main flavors of AI, and each can be GPU accelerated
• There are 3 main types of AI
• Expert systems accelerated through libraries, OpenACC, CUDA
• ML is accelerated with NVIDIA’s RAPIDS
• DL is accelerated via cuDNN in most DL frameworks
7
INCREASING COMPLEXITY AND AUTONOMY OVER TIME
EXPERT SYSTEMS
EXECUTE HAND-WRITTEN
ALGORITHMS AT HIGH SPEED TRADITIONAL ML
LEARN FROM EXAMPLES USING
HAND-CRAFTED FEATURES Accelerate with
NVIDIA RAPIDS
THREE ROADS TO AI
Three main flavors of AI, and each can be GPU accelerated
• There are 3 main types of AI
• Expert systems accelerated through libraries, OpenACC, CUDA
• ML is accelerated with NVIDIA’s RAPIDS
• DL is accelerated via cuDNN in most DL frameworks
8
INCREASING COMPLEXITY AND AUTONOMY OVER TIME
EXPERT SYSTEMS
EXECUTE HAND-WRITTEN
ALGORITHMS AT HIGH SPEED TRADITIONAL ML
LEARN FROM EXAMPLES USING
HAND-CRAFTED FEATURES Accelerate with
NVIDIA RAPIDS
EXPERT SYSTEMS
EXECUTE HAND-WRITTEN
ALGORITHMS AT HIGH SPEED
Accelerate with
GPU accelerated Libraries
OpenACC Directives
CUDA Kernels
EXPERT SYSTEMS
EXECUTE HAND-WRITTEN
ALGORITHMS AT HIGH SPEED
LEARNS BOTH OUTPUT AND
FEATURES FROM DATA
EXPERT SYSTEMS
EXECUTE HAND-WRITTEN
ALGORITHMS AT HIGH SPEED TRADITIONAL ML
LEARN FROM EXAMPLES USING
HAND-CRAFTED FEATURES
THREE ROADS TO AI
Three main flavors of AI, and each can be GPU accelerated
• There are 3 main types of AI
• Expert systems accelerated through libraries, OpenACC, CUDA
• ML is accelerated with NVIDIA’s RAPIDS
• DL is accelerated via cuDNN in most DL frameworks
9
A NEW COMPUTING MODEL
Algorithms that Learn from Examples
Expert Written
Computer
Program
Traditional Approach
➢ Requires domain experts
➢ Time consuming
➢ Error prone
➢ Not scalable to new
problems
Deep Neural Network
Deep Learning Approach
✓ Learn from data
✓ Easily to extend
✓ Speedup with GPUs
10
11
THE EVOLUTION OF PROGRAMMING
Classical coding
Input data
Logic
Output data
 Machine learning
12
6 QUESTIONS FACING EVERY AI ENTERPRISE
Top Challenges for AI, Big Data, and Enterprise Transformation
Is your data doubling each year?
DATA DELUGE
Are you an intelligent enterprise needing
real time predictive analytics?
DELAYED INTELLIGENCE
Is your CAPEX budget shrinking amidst
escalating infrastructure demand?
SHRINKING BUDGET
Is ML training prohibitively long, delaying
time-to-predictions?
PROLONGED TRAINING TIME
Is Spark workloads creating relentless
infrastructure sprawl?
COMPLEX WORKLOADS
$Do you have oceans of data, that take
lifetimes to wrangle?
TEDIOUS DATA PREP
13
SO WHAT?
WHAT DOES IT MEAN FOR HEALTHCARE
RISE OF GPU COMPUTING
20182013
25K
20182013
8M
CUDA Downloads — 5X in 5 YrsGTC Attendees — 7X in 5 Yrs
MEDICAL IMAGING BIOINFORMATICS
COMPUTATIONAL FLUID DYNAMICS NUMERICAL ANALYTICS
DEEP LEARNINGIMAGING AND COMPUTER VISION
COMPUTATIONAL STRUCTURAL MECHANICS
DATA SCIENCE
COMPUTATIONAL CHEMISTRY
RAY TRACING
WEATHER AND CLIMATE
MATERIALS
2
10 YEARS INNOVATING IN HEALTHCARE
15
RADIOLOGY CHALLENGES
>7% increase in images read
https://www.itnonline.com/content/increases-imaging-procedures-chronic-diseases-
spur-growth-medical-imaging-informatics-market
TOO MANY IMAGES
#1 challenge in operational efficiency is inter- &
intra- department sharing & access of images
https://www.radiologybusiness.com/topics/imaging-informatics/6-issues-pacs-
radiology-departments-imaging
INEFFICIENT ACCESS TO IMAGES
10 to 15% of cases are misdiagnosed (in US)
https://appliedradiology.com/articles/diagnostic-errors-in-medicine-a-critical-role-
for-diagnostic-imaging-in-finding-and-facilitating-solutions
DIAGNOSTIC ERRORS
16
17
AI IMPROVES
MEDICAL IMAGING
MRIs can take up to two hours. Subsampled data speeds
scanning time but contributes to inaccurate image
reconstruction.
Researchers from Harvard University and the A.A.
Martinos Center for Biomedical Imaging are using
deep learning to speed up image reconstruction
without compromising accuracy.
Their AI framework, powered by the NVIDIA
DGX-1, reconstructs images directly from
sensor data. It filters out noise and defects
to reconstruct images 100x faster and
with 5x higher accuracy.
Bo Zhu et al, Image reconstruction by domain-transform manifold learning,
Nature (2018). DOI: 10.1038/nature25988
18
SUPERCHARGING
GENOMIC ANALYTICS
China’s healthcare industry is turning to AI to address the
needs of its elderly population. Genetics giant BGI—which
has over 1PB of data—is classifying targetable peptides
for personalized immunotherapy for cancer patients.
By running the open source RAPIDS data processing and
machine learning libraries built on CUDA X AI on an
NVIDIA DGX-1 AI supercomputer, BGI sped up
analysis 18x using cuDF, and 10x using XGBoost.
The company is now expanding analysis
to millions of peptide
candidates.
19
DIGITAL HEALTH
MANAGEMENT
Data helps us make decisions about consumer purchases,
how to navigate traffic, where to eat, etc. By applying
data science to correlate microbiome features and
Type-2 diabetes, iCarbonX helps people make
decisions that can improve their health such
as diet and treatments.
The open source RAPIDS data processing and
machine learning libraries built on CUDA-X AI
deployed on Tencent Cloud P40 servers
resulted in a 6x speed up
of data analytics.
20
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.
21
RAPIDS
SOLUTION OVERVIEW
22
12
6
39
GPU
POWERED
WORKFLOW
DAY IN THE LIFE OF A DATA SCIENTIST
Train Model
Validate
Test Model
Experiment with
Optimizations and
Repeat
Go Home on Time
Dataset
Downloads
Overnight
Start
GET A COFFEE
Stay Late
Restart Data Prep
Workflow Again
Find Unexpected Null
Values Stored as String…
Switch to Decaf
12
6
39
CPU
POWERED
WORKFLOW
Restart Data Prep
Workflow
@*#! Forgot to Add
a Feature
ANOTHER…
GET A COFFEE
Start Data Prep
Workflow
GET A COFFEE
Configure Data Prep
Workflow
Dataset
Downloads
Overnight
Dataset Collection Analysis Data Prep Train Inference
23
THE RAPIDS VALUE PROPOSITION
High Performance, Easy-to-use
Data Scientist Data Science Leader
Reduced Training Time
Drastically improve your productivity with
near-interactive data science
Hassle-Free Integration
Accelerate your Python data science toolchain with
minimal code changes and no new tools to learn
Open Source
Customizable, extensible, interoperable — the
open-source software is supported by NVIDIA and
built on Apache Arrow
Top Model Accuracy
Increase machine learning model accuracy by iterating
on models faster and deploying them more frequently
TCO Reduction
Decrease the server costs, footprint, power consumption
of your ML workloads reducing the TCO
Increased Data Scientist Productivity
Reduce training time, allow data scientists to be more
productive
24
THE RAPIDS ECOSYSTEM
RAPIDS
Open Source
Community
Enterprise Data Science
Platforms
Startups
Deep Learning
Integration
GPU Servers Storage Partners
25
NOT ENOUGH DATA?
NO 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 power 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.
26
MEGA TECHNOLOGY TRENDS
ARTIFICIAL INTELLINGENCESERVICE-ORIENTED ARCHITECTURE
Software Defined Imaging + AI Redefine Radiology
Monolith Modular
27
Partially connected, manual processes
RADIOLOGY INFRASTRUCTURE
28
NVIDIA CLARA PLATFORM
Universal Compute Platform for Medical Imaging
NVIDIA HW
CLARA SDK
Image
Reconstruction
Artificial
Intelligence
Rendering
& Viz
CUDA
GPU
Accelerated Libraries | Engines & Containers | Management Tools
29
CLARA AGX XAVIER
Unify your computing infrastructure
NVIDIA CLARA PLATFORM
High Performance
Computing
Artificial
Intelligence
Rendering
& Viz
Compute | System Libraries | SDKs | Reference Workflows
Deploy anywhere
Scalable infrastructure
Get started quickly
30
CLARA AI
Build, Manage, & Deploy AI in the Clinic
Rapid Data Annotation
Reference Pipelines
created by Data Scientists
Integration into
existing clinical
workflows
Accurate model with less data
31
EXAMPLE
WORKFLOWS
ENGINES &
CONTAINERS
NVIDIA
LIBRARIES
VISUALIZATION
INDEXOPTIX
CUDA | VULKAN
ARTIFICIAL INTELLIGENCE
cuDNN DALI TensorRT
TESLA GPUs
& SYSTEMS
SYSTEM OEM CLOUDTESLA GPU NVIDIA HGXNVIDIA DGX FAMILYVIRTUAL GPU
VISUALIZATION
WEB UI
RENDER
SERVER
COMPUTE
OPEN SOURCE
CT RECON
ARTIFICIAL INTELLIGENCE
AI INFERENCE
ENGINE
TensorRT INF
SERVER
SAMPLE
AI MODELS
COMPUTE
CuBLAS NPPCuFFT NCCL
NVIDIA CLARA SDK
COMPUTED TOMOGRAPHY ULTRASOUNDMAGNETIC RESONANCE
32
DELIVERING AI-ASSISTED
ANNOTATION
The largest research hospital in America, the National
Institutes of Health Clinical Center and NVIDIA scientists
used Clara AI to develop a domain generalization method
for the segmentation of the prostate from surrounding
tissue on MRI.
The localized model achieved performance similar
to that of a radiologist and outperformed other
state-of-the-art algorithms that were
trained and evaluated on data from
the same domain.
33
34
KEY
TAKEAWAYS
Disruptive Technology has created an opportunity
NVIDIA Clara unifies your compute infrastructure
Clara AI provides tools to build, manage, and deploy AI
Transform your business today with AI
35
SO WHAT?
WHAT DOES IT MEAN FOR RETAIL
36
IVA IN INDUSTRIES
Industrial
Product Quality Inspection
Smart Cities
Security and Surveillance
Finance
Security and Fraud Detection
Retail
Loss Prevention
37
LOSS PREVENTION
A major retailer is losing $500K per store, per year to
shrinkage. Using IVA for loss prevention,
the company is able to in real-time identify shoppers
who are switching tickets, double scanning, and
mis-scanning products. The IVA inference is detecting
the theft at the store using deep learning-based
software powered by NVIDIA GPUs and notifies
employees for immediate intervention.
Each store is equipped with the IVA software
installed on a server in the back
of the store, processing 30 frames per
second from cameras above each
checkout stand.
The solution has been tested in
multiple stores successfully saving
millions of dollars in shrinkage
per store.
38
Computer Vision software for standard traffic
counters measure traffic into and out of the store.
Enabling retailers to:
Detect unique Identity
Segment by age / ethnicity
Track shopping behavior
Monitor traffic patterns
Retailers can integrate app-based
recommendation logic and launch targeted
promotions based on proximity, past
purchase, and consumer profile.
Multiple camera signals can be stitched
together to detect in store patterns.
Exterior cameras can determine
shopper density based on parking.
Origin tracking can identify
external traffic sources, and/or
co-marketing opportunities.
CONSUMER MONITORING
39
Using existing cameras, a retailer can install highly
effective computer vision algorithms to detect
shopper traffic patterns and prevent loss.
In the US, Loss Prevention is a $50B problem
impacting all retailers. At the same time, investment
in Loss Prevention staff is flat of shrinking.
While the average cost of shoplifting incidents is
doubling to $798, 30% of inventory shrinkage
is an inside-job. Using computer vision can
identify theft, shrinkage, and shoplifting
incidents. This new technology can
invigorate a longstanding problem
for retail.
LOSS PREVENTION
40
AI is changing the way merchandising decisions are
made. Advanced math and science combined with
NVIDIA GPUs power simulations are the future of
retail and deliver smarter, more profitable decisions
that previously, were unattainable.
Just like in the game of GO, Daisy Intelligence’s
Theory of Retail™ models a retailer’s environment
using their existing POS data; taking into
consideration merchandising objectives,
strategies and constraints, to deliver results
that are beyond human capacity.
Daisy’s clients are seeing tremendous
revenue gains by leveraging AI
powered decisions.
PROMOTIONS, PRICING AND
DEMAND FORECASTING
41
Store Associates are representatives of the brand,
and therefore it makes sense to reduce the time
they spend performing tasks that are not
customer-facing.
Fellow Robots created a solution to scan
shelves, monitor misplaced items, and act
as a way-finder kiosk for consumers. This
allows associates to interact with the
shopping public, improving consumer
satisfaction and raising revenue through
larger shopping baskets.
As an Inception partner, Fellow is
closely aligned with NVIDIA and is
poised to deliver incredible impact
on retail business processes.
VIEW THE VIDEO
SHELF SCANNING AND
WAY-FINDING ROBOTS
42
43
Retailers are looking for new ways to enhance the
consumer experience whiling reducing costs. The
next generation of vending machines need to be cost-
effective and connect consumers with desired products.
Malong Technologies’ Smart Cabinet offers
a computer vision based smart retail solution
that is cost-effective and can support
a diverse selection of products. With their
unique supervised learning techniques, they
have achieved outstanding accuracy in
product recognition and are able to train
a new SKU with only dozens of images.
As an Inception partner, Malong
Technologies is working closely
with NVIDIA to power the
future of retail.
VIEW THE VIDEO
Cameras on shopping cart
On shelf cameras
Haier Smart Cabinets
AUTONOMOUS SHOPPING
WITH SMART CABINETS
44
45
THE STORE OF THE FUTURE
Future-Proofed IVA Infrastructure
Loss Prevention
Stock Out Reduction
Store Analytics
Autonomous Shopping
Security
DL-BASED IVA EDGE USE CASES
Server (T4s)
ServerBackofStore
Jetson AGX Xavier / Nano
T
4
T
4
T
4
T
4
T
4
T
4
In-Store
Cameras Sensors
46
JETSON NANO
Small, low-power AI Computer
128 CUDA Cores | 4 Core CPU
4 GB Memory
472 GFLOPs
70x45mm
5W | 10W
48
49
CUDA • Linux4Tegra • ROS
JETSON SOFTWARE
NsightDeveloperTools
Jetson Computer
Deepstream
Modules
Depth
estimation
Path
planning
Object
detection
Gesture
recognition
Ecosystem
modules
Pose
estimation
…
TensorRT
cuDNN
VisionWorks
OpenCV
libargus
Video API
GraphicsComputer Vision Accel. ComputingDeep Learning
Drivers
Ecosystem
JetpackSDK
cuBLAS
cuFFT
Jetson software: developer.nvidia.com/jetson
Vulkan
OpenGL
SensorsMultimedia
52
JETSON NANO DEVELOPER KIT
$99 AI Computer
128 CUDA Cores | 4 Core CPU
472 GFLOPs
5W | 10W
Available from nvidia.com and
distributors worldwide
53
THANK YOU

Deep Learning & AI for Healthcare and Retail

  • 1.
    Bharatkumar Sharma, SeniorSolution Architect AI IN HEALTHCARE AND RETAIL
  • 2.
    AGENDA 1. Where Icome from? 2. AI and Deep Learning 3. AI in Healthcare 4. Intelligent Video Analytics in Retail
  • 3.
    3 NVIDIA The AI ComputingCompany TRANSPORTATION HEALTHCARE MACHINE LEARNINGHPC DEEP LEARNING GAMING DESIGN
  • 4.
    4 1980 1990 20002010 2020 GPU-Computing perf 1.5X per year 1000X by 2025 RISE OF GPU COMPUTING Original data up to the year 2010 collected and plotted by M. Horowitz, F. Labonte, O. Shacham, K. Olukotun, L. Hammond, and C. Batten New plot and data collected for 2010-2015 by K. Rupp 102 103 104 105 106 107 Single-threaded perf 1.5X per year 1.1X per year APPLICATIONS SYSTEMS ALGORITHMS CUDA ARCHITECTURE
  • 5.
    5 AI REVOLUTION Big DataGPU AccelerationBetter Algorithms 350 million images uploaded per day Petabytes of customer data hourly 300 hours of video uploaded every minute “The Three Breakthroughs that have Finally Unleashed A.I. on the World”
  • 6.
    6 EXPERT SYSTEMS EXECUTE HAND-WRITTEN ALGORITHMSAT HIGH SPEED Accelerate with GPU accelerated Libraries OpenACC Directives CUDA Kernels INCREASING COMPLEXITY AND AUTONOMY OVER TIME EXPERT SYSTEMS EXECUTE HAND-WRITTEN ALGORITHMS AT HIGH SPEED THREE ROADS TO AI Three main flavors of AI, and each can be GPU accelerated • There are 3 main types of AI • Expert systems accelerated through libraries, OpenACC, CUDA • ML is accelerated with NVIDIA’s RAPIDS • DL is accelerated via cuDNN in most DL frameworks
  • 7.
    7 INCREASING COMPLEXITY ANDAUTONOMY OVER TIME EXPERT SYSTEMS EXECUTE HAND-WRITTEN ALGORITHMS AT HIGH SPEED TRADITIONAL ML LEARN FROM EXAMPLES USING HAND-CRAFTED FEATURES Accelerate with NVIDIA RAPIDS THREE ROADS TO AI Three main flavors of AI, and each can be GPU accelerated • There are 3 main types of AI • Expert systems accelerated through libraries, OpenACC, CUDA • ML is accelerated with NVIDIA’s RAPIDS • DL is accelerated via cuDNN in most DL frameworks
  • 8.
    8 INCREASING COMPLEXITY ANDAUTONOMY OVER TIME EXPERT SYSTEMS EXECUTE HAND-WRITTEN ALGORITHMS AT HIGH SPEED TRADITIONAL ML LEARN FROM EXAMPLES USING HAND-CRAFTED FEATURES Accelerate with NVIDIA RAPIDS EXPERT SYSTEMS EXECUTE HAND-WRITTEN ALGORITHMS AT HIGH SPEED Accelerate with GPU accelerated Libraries OpenACC Directives CUDA Kernels EXPERT SYSTEMS EXECUTE HAND-WRITTEN ALGORITHMS AT HIGH SPEED LEARNS BOTH OUTPUT AND FEATURES FROM DATA EXPERT SYSTEMS EXECUTE HAND-WRITTEN ALGORITHMS AT HIGH SPEED TRADITIONAL ML LEARN FROM EXAMPLES USING HAND-CRAFTED FEATURES THREE ROADS TO AI Three main flavors of AI, and each can be GPU accelerated • There are 3 main types of AI • Expert systems accelerated through libraries, OpenACC, CUDA • ML is accelerated with NVIDIA’s RAPIDS • DL is accelerated via cuDNN in most DL frameworks
  • 9.
    9 A NEW COMPUTINGMODEL Algorithms that Learn from Examples Expert Written Computer Program Traditional Approach ➢ Requires domain experts ➢ Time consuming ➢ Error prone ➢ Not scalable to new problems Deep Neural Network Deep Learning Approach ✓ Learn from data ✓ Easily to extend ✓ Speedup with GPUs
  • 10.
  • 11.
    11 THE EVOLUTION OFPROGRAMMING Classical coding Input data Logic Output data  Machine learning
  • 12.
    12 6 QUESTIONS FACINGEVERY AI ENTERPRISE Top Challenges for AI, Big Data, and Enterprise Transformation Is your data doubling each year? DATA DELUGE Are you an intelligent enterprise needing real time predictive analytics? DELAYED INTELLIGENCE Is your CAPEX budget shrinking amidst escalating infrastructure demand? SHRINKING BUDGET Is ML training prohibitively long, delaying time-to-predictions? PROLONGED TRAINING TIME Is Spark workloads creating relentless infrastructure sprawl? COMPLEX WORKLOADS $Do you have oceans of data, that take lifetimes to wrangle? TEDIOUS DATA PREP
  • 13.
    13 SO WHAT? WHAT DOESIT MEAN FOR HEALTHCARE
  • 14.
    RISE OF GPUCOMPUTING 20182013 25K 20182013 8M CUDA Downloads — 5X in 5 YrsGTC Attendees — 7X in 5 Yrs MEDICAL IMAGING BIOINFORMATICS COMPUTATIONAL FLUID DYNAMICS NUMERICAL ANALYTICS DEEP LEARNINGIMAGING AND COMPUTER VISION COMPUTATIONAL STRUCTURAL MECHANICS DATA SCIENCE COMPUTATIONAL CHEMISTRY RAY TRACING WEATHER AND CLIMATE MATERIALS 2 10 YEARS INNOVATING IN HEALTHCARE
  • 15.
    15 RADIOLOGY CHALLENGES >7% increasein images read https://www.itnonline.com/content/increases-imaging-procedures-chronic-diseases- spur-growth-medical-imaging-informatics-market TOO MANY IMAGES #1 challenge in operational efficiency is inter- & intra- department sharing & access of images https://www.radiologybusiness.com/topics/imaging-informatics/6-issues-pacs- radiology-departments-imaging INEFFICIENT ACCESS TO IMAGES 10 to 15% of cases are misdiagnosed (in US) https://appliedradiology.com/articles/diagnostic-errors-in-medicine-a-critical-role- for-diagnostic-imaging-in-finding-and-facilitating-solutions DIAGNOSTIC ERRORS
  • 16.
  • 17.
    17 AI IMPROVES MEDICAL IMAGING MRIscan take up to two hours. Subsampled data speeds scanning time but contributes to inaccurate image reconstruction. Researchers from Harvard University and the A.A. Martinos Center for Biomedical Imaging are using deep learning to speed up image reconstruction without compromising accuracy. Their AI framework, powered by the NVIDIA DGX-1, reconstructs images directly from sensor data. It filters out noise and defects to reconstruct images 100x faster and with 5x higher accuracy. Bo Zhu et al, Image reconstruction by domain-transform manifold learning, Nature (2018). DOI: 10.1038/nature25988
  • 18.
    18 SUPERCHARGING GENOMIC ANALYTICS China’s healthcareindustry is turning to AI to address the needs of its elderly population. Genetics giant BGI—which has over 1PB of data—is classifying targetable peptides for personalized immunotherapy for cancer patients. By running the open source RAPIDS data processing and machine learning libraries built on CUDA X AI on an NVIDIA DGX-1 AI supercomputer, BGI sped up analysis 18x using cuDF, and 10x using XGBoost. The company is now expanding analysis to millions of peptide candidates.
  • 19.
    19 DIGITAL HEALTH MANAGEMENT Data helpsus make decisions about consumer purchases, how to navigate traffic, where to eat, etc. By applying data science to correlate microbiome features and Type-2 diabetes, iCarbonX helps people make decisions that can improve their health such as diet and treatments. The open source RAPIDS data processing and machine learning libraries built on CUDA-X AI deployed on Tencent Cloud P40 servers resulted in a 6x speed up of data analytics.
  • 20.
    20 AI PREDICTS AND PREVENTS DISEASE GPUdeep 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.
  • 21.
  • 22.
    22 12 6 39 GPU POWERED WORKFLOW DAY IN THELIFE OF A DATA SCIENTIST Train Model Validate Test Model Experiment with Optimizations and Repeat Go Home on Time Dataset Downloads Overnight Start GET A COFFEE Stay Late Restart Data Prep Workflow Again Find Unexpected Null Values Stored as String… Switch to Decaf 12 6 39 CPU POWERED WORKFLOW Restart Data Prep Workflow @*#! Forgot to Add a Feature ANOTHER… GET A COFFEE Start Data Prep Workflow GET A COFFEE Configure Data Prep Workflow Dataset Downloads Overnight Dataset Collection Analysis Data Prep Train Inference
  • 23.
    23 THE RAPIDS VALUEPROPOSITION High Performance, Easy-to-use Data Scientist Data Science Leader Reduced Training Time Drastically improve your productivity with near-interactive data science Hassle-Free Integration Accelerate your Python data science toolchain with minimal code changes and no new tools to learn Open Source Customizable, extensible, interoperable — the open-source software is supported by NVIDIA and built on Apache Arrow Top Model Accuracy Increase machine learning model accuracy by iterating on models faster and deploying them more frequently TCO Reduction Decrease the server costs, footprint, power consumption of your ML workloads reducing the TCO Increased Data Scientist Productivity Reduce training time, allow data scientists to be more productive
  • 24.
    24 THE RAPIDS ECOSYSTEM RAPIDS OpenSource Community Enterprise Data Science Platforms Startups Deep Learning Integration GPU Servers Storage Partners
  • 25.
    25 NOT ENOUGH DATA? NOPROBLEM 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 power 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.
  • 26.
    26 MEGA TECHNOLOGY TRENDS ARTIFICIALINTELLINGENCESERVICE-ORIENTED ARCHITECTURE Software Defined Imaging + AI Redefine Radiology Monolith Modular
  • 27.
    27 Partially connected, manualprocesses RADIOLOGY INFRASTRUCTURE
  • 28.
    28 NVIDIA CLARA PLATFORM UniversalCompute Platform for Medical Imaging NVIDIA HW CLARA SDK Image Reconstruction Artificial Intelligence Rendering & Viz CUDA GPU Accelerated Libraries | Engines & Containers | Management Tools
  • 29.
    29 CLARA AGX XAVIER Unifyyour computing infrastructure NVIDIA CLARA PLATFORM High Performance Computing Artificial Intelligence Rendering & Viz Compute | System Libraries | SDKs | Reference Workflows Deploy anywhere Scalable infrastructure Get started quickly
  • 30.
    30 CLARA AI Build, Manage,& Deploy AI in the Clinic Rapid Data Annotation Reference Pipelines created by Data Scientists Integration into existing clinical workflows Accurate model with less data
  • 31.
    31 EXAMPLE WORKFLOWS ENGINES & CONTAINERS NVIDIA LIBRARIES VISUALIZATION INDEXOPTIX CUDA |VULKAN ARTIFICIAL INTELLIGENCE cuDNN DALI TensorRT TESLA GPUs & SYSTEMS SYSTEM OEM CLOUDTESLA GPU NVIDIA HGXNVIDIA DGX FAMILYVIRTUAL GPU VISUALIZATION WEB UI RENDER SERVER COMPUTE OPEN SOURCE CT RECON ARTIFICIAL INTELLIGENCE AI INFERENCE ENGINE TensorRT INF SERVER SAMPLE AI MODELS COMPUTE CuBLAS NPPCuFFT NCCL NVIDIA CLARA SDK COMPUTED TOMOGRAPHY ULTRASOUNDMAGNETIC RESONANCE
  • 32.
    32 DELIVERING AI-ASSISTED ANNOTATION The largestresearch hospital in America, the National Institutes of Health Clinical Center and NVIDIA scientists used Clara AI to develop a domain generalization method for the segmentation of the prostate from surrounding tissue on MRI. The localized model achieved performance similar to that of a radiologist and outperformed other state-of-the-art algorithms that were trained and evaluated on data from the same domain.
  • 33.
  • 34.
    34 KEY TAKEAWAYS Disruptive Technology hascreated an opportunity NVIDIA Clara unifies your compute infrastructure Clara AI provides tools to build, manage, and deploy AI Transform your business today with AI
  • 35.
    35 SO WHAT? WHAT DOESIT MEAN FOR RETAIL
  • 36.
    36 IVA IN INDUSTRIES Industrial ProductQuality Inspection Smart Cities Security and Surveillance Finance Security and Fraud Detection Retail Loss Prevention
  • 37.
    37 LOSS PREVENTION A majorretailer is losing $500K per store, per year to shrinkage. Using IVA for loss prevention, the company is able to in real-time identify shoppers who are switching tickets, double scanning, and mis-scanning products. The IVA inference is detecting the theft at the store using deep learning-based software powered by NVIDIA GPUs and notifies employees for immediate intervention. Each store is equipped with the IVA software installed on a server in the back of the store, processing 30 frames per second from cameras above each checkout stand. The solution has been tested in multiple stores successfully saving millions of dollars in shrinkage per store.
  • 38.
    38 Computer Vision softwarefor standard traffic counters measure traffic into and out of the store. Enabling retailers to: Detect unique Identity Segment by age / ethnicity Track shopping behavior Monitor traffic patterns Retailers can integrate app-based recommendation logic and launch targeted promotions based on proximity, past purchase, and consumer profile. Multiple camera signals can be stitched together to detect in store patterns. Exterior cameras can determine shopper density based on parking. Origin tracking can identify external traffic sources, and/or co-marketing opportunities. CONSUMER MONITORING
  • 39.
    39 Using existing cameras,a retailer can install highly effective computer vision algorithms to detect shopper traffic patterns and prevent loss. In the US, Loss Prevention is a $50B problem impacting all retailers. At the same time, investment in Loss Prevention staff is flat of shrinking. While the average cost of shoplifting incidents is doubling to $798, 30% of inventory shrinkage is an inside-job. Using computer vision can identify theft, shrinkage, and shoplifting incidents. This new technology can invigorate a longstanding problem for retail. LOSS PREVENTION
  • 40.
    40 AI is changingthe way merchandising decisions are made. Advanced math and science combined with NVIDIA GPUs power simulations are the future of retail and deliver smarter, more profitable decisions that previously, were unattainable. Just like in the game of GO, Daisy Intelligence’s Theory of Retail™ models a retailer’s environment using their existing POS data; taking into consideration merchandising objectives, strategies and constraints, to deliver results that are beyond human capacity. Daisy’s clients are seeing tremendous revenue gains by leveraging AI powered decisions. PROMOTIONS, PRICING AND DEMAND FORECASTING
  • 41.
    41 Store Associates arerepresentatives of the brand, and therefore it makes sense to reduce the time they spend performing tasks that are not customer-facing. Fellow Robots created a solution to scan shelves, monitor misplaced items, and act as a way-finder kiosk for consumers. This allows associates to interact with the shopping public, improving consumer satisfaction and raising revenue through larger shopping baskets. As an Inception partner, Fellow is closely aligned with NVIDIA and is poised to deliver incredible impact on retail business processes. VIEW THE VIDEO SHELF SCANNING AND WAY-FINDING ROBOTS
  • 42.
  • 43.
    43 Retailers are lookingfor new ways to enhance the consumer experience whiling reducing costs. The next generation of vending machines need to be cost- effective and connect consumers with desired products. Malong Technologies’ Smart Cabinet offers a computer vision based smart retail solution that is cost-effective and can support a diverse selection of products. With their unique supervised learning techniques, they have achieved outstanding accuracy in product recognition and are able to train a new SKU with only dozens of images. As an Inception partner, Malong Technologies is working closely with NVIDIA to power the future of retail. VIEW THE VIDEO Cameras on shopping cart On shelf cameras Haier Smart Cabinets AUTONOMOUS SHOPPING WITH SMART CABINETS
  • 44.
  • 45.
    45 THE STORE OFTHE FUTURE Future-Proofed IVA Infrastructure Loss Prevention Stock Out Reduction Store Analytics Autonomous Shopping Security DL-BASED IVA EDGE USE CASES Server (T4s) ServerBackofStore Jetson AGX Xavier / Nano T 4 T 4 T 4 T 4 T 4 T 4 In-Store Cameras Sensors
  • 46.
    46 JETSON NANO Small, low-powerAI Computer 128 CUDA Cores | 4 Core CPU 4 GB Memory 472 GFLOPs 70x45mm 5W | 10W
  • 47.
  • 48.
    49 CUDA • Linux4Tegra• ROS JETSON SOFTWARE NsightDeveloperTools Jetson Computer Deepstream Modules Depth estimation Path planning Object detection Gesture recognition Ecosystem modules Pose estimation … TensorRT cuDNN VisionWorks OpenCV libargus Video API GraphicsComputer Vision Accel. ComputingDeep Learning Drivers Ecosystem JetpackSDK cuBLAS cuFFT Jetson software: developer.nvidia.com/jetson Vulkan OpenGL SensorsMultimedia
  • 49.
    52 JETSON NANO DEVELOPERKIT $99 AI Computer 128 CUDA Cores | 4 Core CPU 472 GFLOPs 5W | 10W Available from nvidia.com and distributors worldwide
  • 50.