Upload
others
View
39
Download
0
Embed Size (px)
Citation preview
AI AND DATA SCIENCE FOR THE ENTERPRISE
Jennifer St. John-Foster – NVIDIA Business Manager, Financial Services
2
3
4
AI & DATA SCIENCEARE CHANGING THE WORLD
RoboticsManufacturing, construction, navigation
HealthcareCancer detection, drug discovery, genomics
Internet ServicesImage classification, speech recognition, NLP
FinanceTrading strategy, fraud detection
Media & EntertainmentDigital content creation, game development
Autonomous VehiclesPedestrian & traffic sign detection, lane tracking
5
DATA SCIENCE –A NEW PILLAR OF DISCOVERY
PREDICTIVE
MODELFEATURES
PREDICTIONDATA
ML
DL
CVNLU
AI
DATA
ANALYTICSINFERENCE
6
DATA SCIENCE –A NEW PILLAR OF DISCOVERY
PREDICTIVE
MODELFEATURES
PREDICTIONDATA
ETL, Pandas,
Spark, Graph
TensorFlow, PyTorch
MXNet, Scikit-Learn, XGBoost
TensorFlow Serving
ONNX, SageMaker NEO
CSV, PARQ, HDFS
ML
DL
CVNLU
AI
DATA
ANALYTICSINFERENCE
7
DATA SCIENCE –A NEW PILLAR OF DISCOVERY
PREDICTIVE
MODELFEATURES
PREDICTIONDATA
cuIO
cuDF
cuGraph
cuDNN
cuML
TensorRT
TRTIS
ML
DL
CVNLU
AI
DATA
ANALYTICSINFERENCE
8
SMALL CHANGES, BIG SPEED-UPApplication Code
+
GPU CPU5% of Code
Compute-Intensive Functions
Rest of SequentialCPU Code
9
ML SERVICES
NVIDIA CUDA-X AI ECOSYSTEM
FRAMEWORKS DEPLOYMENT
DA GRAPH DL TRAINML DL INFERENCE
Serving
Amazon
SageMaker Neo
CUDA-X AI
CUDA
AmazonSageMaker
Azure Machine Learning
GoogleCloud ML
10
CONTAINERS: SIMPLIFYING WORKFLOWS
Simplifies Deployments
- Eliminates complex, time-consuming builds and installs
Get started in minutes
- Simply Pull & Run the app
- https://ngc.nvidia.com
Portable
- Deploy across various environments, from test to production with minimal changes
11
DESIGNING INFRASTRUCTURE THAT SCALESInsights gained from deep learning data centers
Rack Design Networking Storage Facilities Software
• DL drives
close to
operational
limits
• Similarities
to HPC best
practices
• IB or
Ethernet
based fabric
• 100Gbps
inter-
connect
• High-
bandwidth,
ultra-low
latency
• Datasets
range from
10k’s to
millions
objects
• terabyte
levels of
storage and
up
• High IOPS,
low latency
• assume
higher watts
per-rack
• Higher
FLOPS/watt
= DC less
floorspace
required
• Scale
requires
“cluster-
aware”
software
Example:
• Autonomous vehicle = 1TB / hr
• Training sets up to 500 PB
• RN50: 113 days to train
• Objective: 7 days
• 6 simultaneous developers
= 97 node cluster
12
DATACENTER BECOMES A COMPUTE ENGINE
13
PREDICT CUSTOMER INTENT TO PURCHASE
There’s an increasing need to accurately
predict customers’ intent to buy into
extensive product portfolios to help with
companies’ bottom line.
Cisco uses Driverless AI powered by NVIDIA
GPUs to provide pre-built ready to use
algorithms and models, reducing the
processing time from 1 month to
2 days with much larger data sets.
This resulted in more comprehensive
view of customer behavior.
14
With >100,000 different products in its 4,700 U.S. stores,
the Walmart Labs data science team predicts demand for
500 million item-by-store combinations every week.
By performing forecasting with the open-source RAPIDS
data processing and machine learning libraries built
on CUDA-X AI on NVIDIA GPUs, Walmart speeds
up feature engineering 100x and trains machine
learning algorithms 20x faster, resulting in faster
delivery of products, real-time reaction to
shopper trends, and inventory cost
savings at scale.
IMPROVINGDEMAND FORECASTS
15
AI HELPS DOCTORSDIAGNOSEBREAST CANCEREvery 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%.
16
REAL-TIME FRAUD DETECTIONRecently, PayPal was looking to deploy a new fraud
detection system. The team working on it set a high
bar: this system had to operate worldwide 24/7,
and work in real-time to protect customer
transactions from potential fraud. In spec’ing
the system, it became evident that CPU-only
servers couldn’t meet these requirements.
Using NVIDIA T4 GPUs, PayPal delivered a
new level of service, using GPU inference
to improve real-time fraud detection by
10% while lowering server capacity
by nearly 8x.
17