Alison B. Lowndes, profile picture
Uploaded byAlison B. Lowndes
612 views

AI in the Financial Services Industry

The document discusses the integration of GPU technology in various sectors such as automotive, financial services, healthcare, and customer service, emphasizing its role in enhancing deep learning and machine learning workflows. It details NVIDIA's CUDA-X AI ecosystem, highlighting tools for model training, inference, and optimization, along with various applications in fields like sentiment analysis and fraud detection. Key metrics showcase the significant speed and efficiency improvements provided by GPU acceleration over traditional CPU methods.

Embed presentation

Downloaded 32 times
1 ALISON B LOWNDES AI DevRel | EMEA @alisonblowndes March 2019
2 INSERT M82 GALAXY SIMS
www.FrontierDevelopmentLab.org
4
5
6 The day job AUTOMOTIVE Auto sensors reporting location, problems COMMUNICATIONS Location-based advertising CONSUMER PACKAGED GOODS Sentiment analysis of what’s hot, problems $ FINANCIAL SERVICES Risk & portfolio analysis New products EDUCATION & RESEARCH Experiment sensor analysis HIGH TECHNOLOGY / INDUSTRIAL MFG. Mfg. quality Warranty analysis LIFE SCIENCES MEDIA/ENTERTAINMENT Viewers / advertising effectiveness ON-LINE SERVICES / SOCIAL MEDIA People & career matching HEALTH CARE Patient sensors, monitoring, EHRs OIL & GAS Drilling exploration sensor analysis RETAIL Consumer sentiment TRAVEL & TRANSPORTATION Sensor analysis for optimal traffic flows UTILITIES Smart Meter analysis for network capacity, LAW ENFORCEMENT & DEFENSE Threat analysis - social media monitoring, photo analysis
7 SELECTING THE RIGHT GPU SOLUTION
8 NVIDIA CUDA-X AI ECOSYSTEM FRAMEWORKS CLOUD DEPLOYMENT Workstation CloudServer DA GRAPH DL TRAINML DL INFERENCE Amazon SageMaker Serving Amazon SageMaker Neo Google Cloud ML CUDA-X AI CUDA Azure Machine Learning
9 CUDA X ECOSYSTEM PRogrammable Acceleration across multiple Domains with one Architecture (PRADA) Specialized PerformanceEase of use FrameworksApplications Libraries Directives and Standard Languages Extended Standard Languages CUDA-C++ CUDA Fortran GPU Users Domain Specialists Problem Specialists New Algorithm Developers and Optimization Experts
10 THE NEW NGC GPU-optimized Software Hub. Simplifying DL, ML and HPC Workflows NGC 50+ Containers DL, ML, HPC 50+ Pre-trained Models NLP, Classification, Object Detection & more Industry Workflows Medical Imaging, Intelligent Video Analytics 10+ Model Training Scripts NLP, Image Classification, Object Detection & more Innovate Faster Deploy Anywhere Simplify Deployments
11 TESLA V100 TENSOR CORE GPU World’s Most Advanced Data Center GPU 5,120 CUDA cores 640 NEW Tensor cores 7.8 FP64 TFLOPS | 15.7 FP32 TFLOPS 125 Tensor TFLOPS 32 GB HBM2 @ 900GB/s | 300GB/s NVLink
12 THE PATH FORWARD CPU + Accelerator Simulation + AIFull-stack Optimization FP64 + Multi-Precision + 5.3 7.810.6 15.7 21.2 125 P100 V100
NETWORK COMPLEXITY IS EXPLODING
14 PRICING DERIVATIVES COMPARING CPU TO GPU European Options SPX Monte Carlo Simulation Louis Scott — March 2018 American Options SPY Finite Difference Method Finite Difference Method Hardware Compute Time Speed Up Factor Seconds Days to Option Expiration 105 CPU - i7 5630 146 GPU- V100 0.268 544 x faster Monte Carlo 2m simulations 4 options Hardware Compute Time Speed Up Factor Seconds Days to Option Expiration 105 CPU - i7 5630 47 GPU- V100 0.18 267 x faster http://on-demand.gputechconf.com/gtc/2018/video/S8123/
15 NVIDIA RESEARCH
16 HOW
17
18 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
NATURAL LANGUAGE PROCESSING FOR SIGNAL GENERATION ON NEWS DATA
WORD EMBEDDINGS GLoVe – Global Vectors for Word Representation, utilizes the word-to-word co- occurrence statistics from a given corpus.
22 Algorithmic Trading using Deep Autoencoder based Statistical Arbitrage NVIDIA Deep Learning Institute
23 Moving Average – Mean Reversion
24 Autoencoder
25
26 Snorkel https://github.com/HazyResearch/snorkel Automation of labelling data A system for rapidly creating, modeling, and managing training data, For domains in which large labelled training sets are not available or easy to obtain. Learning, essentially, which labelling functions are more accurate than others—and then using this to train a DNN A general framework for many weak supervision techniques.
27 AUTO ML: AI CREATING AI http://automl.chalearn.org/
Definitions
29 “the machine equivalent of experience”
30 WHAT IS A TENSOR? And why do they flow?
31 WHAT IS A TENSOR? And why do they flow? Scalar is a list of numbers with 0 indices (length 1) a Vector is a list of numbers with 1 index of length k a[k] Matrix is a list of numbers with 2 indices of length r,c a[r,c] A Tensor is a list of numbers with n indices of length n1, n2, …, nm a[n1..nm] an n-dimensional array
32 Origin of Neural Networks Input Output
33 A Simple Neuron Input Output Neuron x1 w2x2 y w1x1 x2
34 Neural Network Basics 2nd step: sum 1st step: Activations * Weights 3rd step: activate
35 Combining Neurons x1 x2 x3 x4 x5 —Additional neurons can be added to create a layer —Multiple layers can also be added, resulting in input, hidden, and output layers —Expanding the neural network size creates additional predictive power —In feed forward neural networks, neurons are fully connected to surrounding layers y Input Layer Output Layer Hidden Layers
36 Deep Neural Networks (DNNs) x1 x2 x3 x4 x5 Input Layer Output LayerMany Hidden Layers y
37 Image “Volvo XC90” Image source: “Unsupervised Learning of Hierarchical Representations with Convolutional Deep Belief Networks” ICML 2009 & Comm. ACM 2011. Honglak Lee, Roger Grosse, Rajesh Ranganath, and Andrew Ng. CONVOLUTIONAL NEURAL NETWORKS
38Yann Le Cun (FaceBook) ISSCC, Feb 2019
39 ROBUST ML https//madry-lab.ml
40 http://distill.pub/2017/momentum/
41 Li et al, University of Maryland & US Naval Academy https://arxiv.org/pdf/1712.09913.pdf
42 AUTOENCODERS UNSUPERVISED feature learning Sparse Representation Training Data Reconstruction Encoder Decoder Minimize Reconstruction Error - InputLayer HiddenLayer HiddenLayer HiddenLayer BottleneckLayer HiddenLayer HiddenLayer HiddenLayer OutputLayer
43 Long short-term memory (LSTM) Hochreiter (1991) analysed vanishing gradient “LSTM falls out of this almost naturally” Gates control importance of the corresponding activations Training via backprop unfolded in time LSTM: input gate output gate Long time dependencies are preserved until input gate is closed (-) and forget gate is open (O) forget gate Fig from Vinyals et al, Google April 2015 NIC Generator Fig from Graves, Schmidhuber et al, Supervised Sequence Labelling with RNNs
44 ARCHITECTURES Larger image: http://www.asimovinstitute.org/neural-network-zoo/
45 Types of ML/DL
46 TRAINING VS INFERENCE
47 POET
48 Tabish Rashid et al, Oxford/DeepMind QMIX: https://arxiv.org/pdf/1803.11485.pdf
49 DEEPMIND ALPHA* 1.0 => 2.0 => 0.0 * ..a deeply structured hybrid (Gary Marcus, Jan 2018)
50
51
52
53 COMPUTATIONAL SCALE REQUIRED 3 million labeled images 1 DGX-1 trains 300k labeled images on 1 DNN in 1 day 10 DNNs required for self-driving 10 parallel experiments at all times 100 DGX-1 per car
54 INNOVATIONS & KEY METRICS DISTILLED NVIDIA SATURNV REFERENCE Design Best Practices Lessons learned from building the worlds largest AI Infrastructure Reference Architectures Partner Reference Architectures, Scale up racks Product Innovation & Quality Rapid exploration & resolution of customer issues
55
Illuminate Deep Networks achler@OptimizingMind.com Original Image Most Certain Uncertain Layer 273: Visualizing expected vs actual outputs Uncertain Inputs of two filters which are most uncertain
57 USE CASES
58 DATA SCIENCE IN FINANCE Alpha Stock Identification Analyze Consumers’ Behavior Anti-fraud API Service Insurance Campaign And Conversion Analysis Credit Card Application Approval Customer service chatbots/routing Claim Fraud Detection Evaluate Create Worthiness Fraud And Credit Risk Analysis Fraud Detection Hedge Fund Management Risk evaluation
59 Financial investment forecasting involves processing vast amounts of data to derive predictions that can outperform the market. SpaceKnow is using GPU to extract global macro and micro-economic activity that helps build high-performance portfolios AI-POWERED INVESTMENT ALPHA
60 EQUIFAX Equifax now has NeuroDecision Technology “NeuroDecision Technology (NDT) is the first regulatory-compliant machine learning credit scoring system reviewed by regulators and credit scoring experts. This technology develops a neural network model that improves performance and accuracy, which gives customers the ability to make more informed business decisions when assessing risk.” https://investor.equifax.com/news-and-events/news/2018/03-26-2018-143044126 “The executive noted that the neural net has improved its ability to make predictive models by as much as 15 percent.” https://www.pymnts.com/innovation/2017/equifax-uses-deep-neural-machine-learning-to-improve-credit-scoring/ Neural nets can be 15% better for prediction
61 FRAUD DETECTION Incumbent firms looking at deep structured history of customers to do supervised learning on fraud / no fraud inline with transactions; some use of autoencoders and other networks to do latent space clustering to identify fraud after the fact. Mixed use of raw transactions over time (RNN) and transaction summary vectors (RNN and CNN) to train models. CNN or fully connected has advantages in-line w.r.t. transcation latency. DL shown to be able to dramatically reduce false positives in transactional fraud! Also use cases around Speech to Text transcription for insider trading etc. Commercial applications challenged today by industry jargon, accents, non-English or Mandarin but these challenges can be overcome. Discussion
62 CREDIT SCORING Likelihood of default. Incumbent firms mostly trying to mine their existing data to more accurately predict repayment/prepayment/default behavior. Can leverage DL to find structure that ML models can then be retrained to exploit in an explainable manner. Challenger / startups looking to use DL to combine multiple data sources to develop models that produce either scores correlated with existing scores or their own “behavioral score”. Frequent challenges around data bias introduced by past credit criteria and practices! Explainability is a primary concern of many firms, especially under GDPR. Techniques including LIME, latent space clustering, nearest training set example, and other emerging network visualization research is being followed closely. Discussion
63 STAC A2™ BENCHMARK STAC A2 Benchmark Developed by banks Macro and micro, performance and accuracy Pricing and Greeks for American exercise basket option, correlated Heston dynamics, Longstaff Schwartz Monte Carlo Independently Audited Results GPU Solution The first system to handle the baseline problem size in "real time" (less than one second) Please see http://www.stacresearch.com/a2 for more details of the STAC Benchmark Also see https://devblogs.nvidia.com/parallelforall/american-option-pricing-monte-carlo-simulation/ for more details on Longstaff-Schwartz Monte Carlo on GPUs
64 DEEP LEARNING FOR CUSTOMER SERVICE OPERATIONS AI-assisted and fully autonomous customer interactions. Integrates with leading customer service software and communications channels Human AgentsHuman Customer Agent Console Deep Neural Networks AI Model is trained on historical chat logs and customer service transcripts. DigitalGenius TensorFlow and Pytorch Supervised Unsupervised
65 GPU-ACCELERATED BERT State-of-the Art Natural Language Processing BERT AVAILABLE ON NGC SUPER-HUMAN QUESTION ANSWERING 280X FASTER TRAINING REAL-TIME INFERENCE Question Answering Translation Dialog Sentiment Analysis Summarizing 86.6 89.3 90.1 91.8 85 87 89 91 93 RM Reader BERT CPU Server DGX-2 CPU Server T4 Training 52 Hours 13 Mins 230 ms 18 ms Inference QANet nlnet Human Source: Question answering accuracy on SQUAD 1.1 for non-ensemble models
66 GPU ACCELERATED MACHINE LEARNING FOR BOND PRICE PREDICTION 100 data elements per trade: Trade size / historical Coupon rate / time to maturity Bond rating Trade type buy/sell Reporting delays Current yield / yield to maturity Launch as many CUDA threads as there are data elements leverage 5120 Cores on V100 to run multiple Kernels in parallel http://on-demand-gtc.gputechconf.com/gtcnew/on-demand- gtc.php?searchByKeyword=s8655&searchItems=session_id&sessionTopic=&sessionEvent=&sessionYear=&sessionFormat=&submit=&select= NEARLY 10X SPEED UP OVER CPU IMPLEMENTATION Bond trading price 20 21 22 23 24 25 p SpeedupoverCPU 0 2 4 6 8 10 Unoptimized CUDA Optimized CUDA
67 APPLYING DEEP LEARNING TO FINANCIAL MARKETS WITH NEWS DATA Recording: http://on-demand.gputechconf.com/gtc/2017/video/s7696-andrew-tan-applying-deep-learning-to-financial-market-signal-identification-with-news- data.mp4 PDF: http://on-demand.gputechconf.com/gtc/2017/presentation/s7696_Andrew-Tan_ FinancialMarketSignalIdentification.pdf
68 GPU ACCELERATED COMPUTING HPC AI ML VISUALIZATION CLARA NVIDIA AI RAPIDS ISAAC DRIVE METROPOLIS WORKSTATIONS SERVERS CLOUD CUDA & GPU COMPUTING ARCHITECTURE NVIDIA GPU CLOUD TESLA DGXTEGRA
69 APPS & FRAMEWORKS CUDA-X NVIDIA LIBRARIES NVIDIA DATA CENTER PLATFORM Single Platform Drives Utilization and Productivity VIRTUAL GPU CUDA & CORE LIBRARIES - cuBLAS | NCCL DEEP LEARNING cuDNN HPC cuFFTOpenACC +550 Applications Amber NAMD CUSTOMER USE CASES VIRTUAL GRAPHICS Speech Translate Recommender SCIENTIFIC APPLICATIONS Molecular Simulations Weather Forecasting Seismic Mapping CONSUMER INTERNET & INDUSTRY APPLICATIONS ManufacturingHealthcare Finance GPUs & SYSTEMS SYSTEM OEM CLOUDTESLA GPU NVIDIA HGXNVIDIA DGX FAMILY MACHINE LEARNING cuMLcuDF cuGRAPH cuDNN CUTLASS TensorRTvDWS vPC Creative & Technical Knowledge Workers vAPPS +600 Applications DX/OGL
70 TRADITIONAL DATA SCIENCE CLUSTER Workload Profile: • 192GB mortgage data set • 16 years, 68 quarters • 34.7 Million single family mortgage loans • 1.85 Billion performance records • XGBoost training set: 50 features 300 Servers | $3M | 180 kW
71 GPU-ACCELERATED MACHINE LEARNING CLUSTER 1 DGX-2 | 10 kW 1/8 the Cost | 1/15 the Space 1/18 the Power DGX-2 and RAPIDS for Predictive Analytics 0 2,000 4,000 6,000 8,000 10,000 20 CPU Nodes 30 CPU Nodes 50 CPU Nodes 100 CPU Nodes DGX-2 5x DGX-1 End-to-End
2 PFLOPS | 512GB HBM2 | 10 kW | 350 lbs NVIDIA DGX-2
73©2018 VMware, Inc. OPTIMIZED SOFTWARE FASTER DEPLOYMENTS Eliminates installations. Simply Pull & Run the app Key DL frameworks updated monthly for perf optimization Empowers users to deploy the latest versions with IT support Better Insights and faster time-to-solution NGC – SIMPLIFYING AI & HPC WORKFLOWS ZERO MAINTENANCE HIGHER PRODUCTIVITY EMBEDDING EXPERTISE Deliver greater value, faster
74
75 GET STARTED WITH NGC Deploy containers: ngc.nvidia.com Learn more about NGC offering: nvidia.com/ngc Technical information: developer.nvidia.com Explore the NGC Registry for DL, ML & HPC
76 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
RAPIDS RAPIDS GPU Accelerated End-to-End Data Science RAPIDS is a set of open source libraries for GPU accelerating data preparation and machine learning. OSS website: rapids.ai GPU Memory Data Preparation VisualizationModel Training cuGraph Graph Analytics cuML Machine Learning cuDF Data Preparation
78 DATA SCIENCE WORKFLOW WITH RAPIDS Open Source, End-to-end GPU-accelerated Workflow Built On CUDA DATA DATA PREPARATION GPUs accelerated compute for in-memory data preparation Simplified implementation using familiar data science tools Python drop-in Pandas replacement built on CUDA C++. GPU-accelerated Spark (in development) PREDICTIONS
79 DATA SCIENCE WORKFLOW WITH RAPIDS Open Source, End-to-end GPU-accelerated Workflow Built On CUDA MODEL TRAINING GPU-acceleration of today’s most popular ML algorithms XGBoost, PCA, Kalman, K-means, k-NN, DBScan, tSVD … DATA PREDICTIONS
80 DATA SCIENCE WORKFLOW WITH RAPIDS Open Source, End-to-end GPU-accelerated Workflow Built On CUDA VISUALIZATION Effortless exploration of datasets, billions of records in milliseconds Dynamic interaction with data = faster ML model development Data visualization ecosystem (Graphistry & OmniSci), integrated with RAPIDS DATA PREDICTIONS
81 www.RAPIDS.ai
82 PILLARS OF RAPIDS PERFORMANCE CUDA Architecture NVLink/NVSwitch Integrated Software Massively Parallel Processing High Speed Connecting between GPUs for Distributed Algorithms Fully Integrated Software and Hardware for Instant Productivity NVSwitch 6x NVLink CUDA PYTHON APACHE ARROW on GPU Memory DASK cuDNN RAPIDS cuMLcuDF DL FRAMEWORKS
83 NEW TURING TENSOR CORE MULTI-PRECISION FOR AI INFERENCE & SCALE-OUT TRAINING 65 TFLOPS FP16 | 130 TeraOPS INT8 | 260 TeraOPS INT4
84 TENSOR CORE AUTOMATIC MIXED PRECISION Over 3x Speedup With Just Two Lines of Code TOOLS AND LIBRARIES MAINTAIN NETWORK ACCURACY TRAINING SPEEDUP OVER 3X INFERENCE SPEEDUP OVER 4X 0 20000 40000 60000 80000 100000 PyTorch GNMT TotalTokens/sec FP32 Mixed 3.4X 1xV100 0 2000 4000 6000 8000 TensorRT ResNet50 Images/sec FP32 INT8 Mixed 4.4X 7ms Latency 1xV100 Tensor Core Journey Page Github Profiler Tools
85 Circa 2000 - Torch7 - 4th (using odd numbers only 1,3,5,7) Web-scale learning in speech, image and video applications Maintained by top researchers including Soumith Chintala - Research Engineer @ Facebook All the goodness of Torch7 with an intuitive Python frontend that focuses on rapid prototyping, readable code & support for a wide variety of deep learning models. https://pytorch.org/2018/05/02/road-to-1.0.html
86 Apex - A PyTorch Extension ● Goal: Raise PyTorch customer awareness and increase adoption of NVIDIA Tensor Cores ● Content: Provide an easy to use set of utility functions in PyTorch for mixed-precision optimizations ● Benefit: Few lines of code to achieve improved training speed while maintaining accuracy and stability of single precision (Tensor Cores) ● Target audience: Deep learning researchers and developers of PyTorch with NVIDIA Volta ● Key Features: AMP (Auditor for mixed-precision) and Optimizer Wrapper (Dynamic loss scaling and master parameters) ● Teams Involved: Leading NVIDIA PyTorch team and collaboration with external FB PyTorch team Overview
87 EVEN MORE SOFTWARE
88 DALI Full input pipeline acceleration including data loading and augmentation Drop-in integration with direct plugins to frameworks – MxNet, TensorFlow, PyTorch Portable workflows through multiple input formats and configurable graphs Unblock CPU with GPU-accelerated DL pre-processing library Version 0.1 supports: • Resnet-50 image classification & segmentation training • Input formats – JPEG, LMDB, RecordIO, TFRecord • Python APIs to define, build and run an input pipelineAvailable as open source - https://github.com/NVIDIA/DALI • DALI Samples & Tutorial: https://github.com/NVIDIA/DALI/blob/master/examples/Getting%20started.ipynb • nvJPEG (Webpage, Documentation) - https://developer.nvidia.com/nvjpeg
89 AI INFERENCE NEEDS TO RUN EVERYWHERE Training InferencingDNN Model
90 T4: UNIVERSAL INFERENCE ACCELERATOR
91 GTC-Pre Announce during keynote Customer sign-up page will go live, post-keynote ANNOUNCING NVIDIA T4 ON AMAZON AWS
92 92
93 Learn more here: https://nvidia.com/data-center-inference https://docs.nvidia.com/deeplearning/sdk/inference-release-notes/index.html Get the ready-to-deploy container with monthly updates from the NGC container registry: https://ngc.nvidia.com/catalog/containers/nvidia%2Ftensorrtserver Open source GitHub repository: https://github.com/NVIDIA/tensorrt-inference-server LEARN MORE AND DOWNLOAD TO USE
94d e v e l o p e r . n v i d i a . c o m
Fundamentals Accelerated Computing Game Development & Digital Content Finance NVIDIA DEEP LEARNING INSTITUTE Online self-paced labs and instructor-led workshops on deep learning and accelerated computing Take self-paced labs at www.nvidia.co.uk/dlilabs View upcoming workshops and request a workshop onsite at www.nvidia.co.uk/dli Educators can join the University Ambassador Program to teach DLI courses on campus and access resources. Learn more at www.nvidia.com/dli Intelligent Video Analytics Healthcare Robotics Autonomous Vehicles Virtual Reality
96 NVIDIA INCEPTION PROGRAM Accelerates AI startups with a boost of GPU tools, tech and deep learning expertise Startup Qualifications Driving advances in the field of AI Business plan Incorporated Web presence Technology DL startup kit* Pascal Titan X Deep Learning Institute (DLI) credit Connect with a DL tech expert DGX-1 ISV discount* Software release notification Live webinar and office hours *By application Marketing Inclusion in NVIDIA marketing efforts GPU Technology Conference (GTC) discount Emerging Company Summit (ECS) participation+ Marketing kit One-page story template eBook template Inception web badge and banners Social promotion request form Event opportunities list Promotion at industry events GPU ventures+ +By invitation www.nvidia.com/inception
97 alowndes@nvidia.com

More Related Content

PPTX
Ai in healthcare (3)
byNicholas Gormley
 
PPTX
Artificial intelligence in healthcare
byYamini Shah
 
PPTX
Responsible AI
byAnand Rao
 
PPTX
Artificial Intelligence in Health Care
by247 Labs Inc
 
PPTX
AI in manufacturing
byStephanie Locke
 
PPTX
Difference between Artificial Intelligence, Machine Learning, Deep Learning a...
bySanjay Srivastava
 
PPTX
Artificial intelligence and its application
byMohammed Abdel Razek
 
PDF
Artificial Intelligence
byDanial Shaikh, CPA, CA
 
Ai in healthcare (3)
byNicholas Gormley
 
Artificial intelligence in healthcare
byYamini Shah
 
Responsible AI
byAnand Rao
 
Artificial Intelligence in Health Care
by247 Labs Inc
 
AI in manufacturing
byStephanie Locke
 
Difference between Artificial Intelligence, Machine Learning, Deep Learning a...
bySanjay Srivastava
 
Artificial intelligence and its application
byMohammed Abdel Razek
 
Artificial Intelligence
byDanial Shaikh, CPA, CA
 

What's hot

PDF
AI in Healthcare | Future of Smart Hospitals
byRenee Yao
 
PPTX
What is Artificial Intelligence?
byErdogan Dagdelenli
 
PDF
Intro Artificial Intelligence
byHans-Dieter Wehle
 
PPTX
Augmented reality and Virtual reality in Healthcare
byAryaShahi
 
PDF
AI and machine learning
byITU
 
PPTX
AI Solutions in Manufacturing
bySri Ambati
 
PDF
Artificial Intelligence & Robotics in Medicine: what does future hold?
byVaibhav Bagaria
 
PDF
Understanding generative AI models A comprehensive overview.pdf
byStephenAmell4
 
PDF
Explainable AI (XAI)
byManojkumar Parmar
 
PPTX
Artificial intelligence in medicine
bydhanyashree15
 
PDF
Artificial Intelligence in Healthcare Future Outlook.pdf
bySoluLab1231
 
PPTX
What is Weak (Narrow) AI? Here Are 8 Practical Examples
byBernard Marr
 
PDF
Responsible Generative AI
byCMassociates
 
PDF
AI and Healthcare 2022.pdf
byKimberley Barker
 
PDF
Human-Centered Artificial Intelligence: Reliable, Safe & Trustworthy
byJalnaAfridi
 
PPTX
ARTIFICIAL INTELLIGENCE ROLE IN HEALTH CARE Dr.T.V.Rao MD
bySociety for Microbiology and Infection care
 
PPTX
Artificial intelligence : what it is
bywww.myassignmenthelp.net
 
PPTX
Deep Learning Explained: The future of Artificial Intelligence and Smart Netw...
byMelanie Swan
 
PDF
Responsible AI
byNeo4j
 
PPTX
Role of artificial intelligence in health care
byPrachi Gupta
 
AI in Healthcare | Future of Smart Hospitals
byRenee Yao
 
What is Artificial Intelligence?
byErdogan Dagdelenli
 
Intro Artificial Intelligence
byHans-Dieter Wehle
 
Augmented reality and Virtual reality in Healthcare
byAryaShahi
 
AI and machine learning
byITU
 
AI Solutions in Manufacturing
bySri Ambati
 
Artificial Intelligence & Robotics in Medicine: what does future hold?
byVaibhav Bagaria
 
Understanding generative AI models A comprehensive overview.pdf
byStephenAmell4
 
Explainable AI (XAI)
byManojkumar Parmar
 
Artificial intelligence in medicine
bydhanyashree15
 
Artificial Intelligence in Healthcare Future Outlook.pdf
bySoluLab1231
 
What is Weak (Narrow) AI? Here Are 8 Practical Examples
byBernard Marr
 
Responsible Generative AI
byCMassociates
 
AI and Healthcare 2022.pdf
byKimberley Barker
 
Human-Centered Artificial Intelligence: Reliable, Safe & Trustworthy
byJalnaAfridi
 
ARTIFICIAL INTELLIGENCE ROLE IN HEALTH CARE Dr.T.V.Rao MD
bySociety for Microbiology and Infection care
 
Artificial intelligence : what it is
bywww.myassignmenthelp.net
 
Deep Learning Explained: The future of Artificial Intelligence and Smart Netw...
byMelanie Swan
 
Responsible AI
byNeo4j
 
Role of artificial intelligence in health care
byPrachi Gupta
 

Similar to AI in the Financial Services Industry

PDF
AI in the Enterprise: Past, Present & Future - StampedeCon AI Summit 2017
byStampedeCon
 
PDF
Alison Lowndes, Artificial Intelligence DevRel, Nvidia – Fueling the Artifici...
byTechsylvania
 
PDF
Nvidia why every industry should be thinking about AI today
byJustin Hayward
 
PDF
NVIDIA @ Infinite Conference, London
byAlison B. Lowndes
 
PDF
AI is moving from its academic roots to the forefront of business and industry
byDigital Transformation EXPO Event Series
 
PPTX
Deep Learning on Qubole Data Platform
byShivaji Dutta
 
PPTX
Deep Learning Workflows: Training and Inference
byNVIDIA
 
PDF
Fuelling the AI Revolution with Gaming
byC4Media
 
PDF
Enterprise deep learning lessons bodkin o reilly ai sf 2017
byRon Bodkin
 
PPTX
Machine Learning AND Deep Learning for OpenPOWER
byGanesan Narayanasamy
 
PDF
AI and Deep Learning with NVIDIA Technologies
bySandeepKS52
 
PDF
Introduction to Deep Learning (NVIDIA)
byRakuten Group, Inc.
 
PDF
How to program DL & AI applications
byJordi TORRES.ai
 
PDF
Machine Learning & AI - 2022 intro for pre-college students.pdf
byEd Fernandez
 
PDF
Data Science, Machine Learning and Neural Networks
byBICA Labs
 
PDF
AI and Deep Learning
bySubrat Panda, PhD
 
PPTX
abelbrownnvidiarakuten2016-170208065814 (1).pptx
bygopikahari7
 
PDF
Deep Learning Demystified
byAffine Analytics
 
PDF
Introduction to the Artificial Intelligence and Computer Vision revolution
byDarian Frajberg
 
PDF
Neural Networks and Deep Learning
byAsim Jalis
 
AI in the Enterprise: Past, Present & Future - StampedeCon AI Summit 2017
byStampedeCon
 
Alison Lowndes, Artificial Intelligence DevRel, Nvidia – Fueling the Artifici...
byTechsylvania
 
Nvidia why every industry should be thinking about AI today
byJustin Hayward
 
NVIDIA @ Infinite Conference, London
byAlison B. Lowndes
 
AI is moving from its academic roots to the forefront of business and industry
byDigital Transformation EXPO Event Series
 
Deep Learning on Qubole Data Platform
byShivaji Dutta
 
Deep Learning Workflows: Training and Inference
byNVIDIA
 
Fuelling the AI Revolution with Gaming
byC4Media
 
Enterprise deep learning lessons bodkin o reilly ai sf 2017
byRon Bodkin
 
Machine Learning AND Deep Learning for OpenPOWER
byGanesan Narayanasamy
 
AI and Deep Learning with NVIDIA Technologies
bySandeepKS52
 
Introduction to Deep Learning (NVIDIA)
byRakuten Group, Inc.
 
How to program DL & AI applications
byJordi TORRES.ai
 
Machine Learning & AI - 2022 intro for pre-college students.pdf
byEd Fernandez
 
Data Science, Machine Learning and Neural Networks
byBICA Labs
 
AI and Deep Learning
bySubrat Panda, PhD
 
abelbrownnvidiarakuten2016-170208065814 (1).pptx
bygopikahari7
 
Deep Learning Demystified
byAffine Analytics
 
Introduction to the Artificial Intelligence and Computer Vision revolution
byDarian Frajberg
 
Neural Networks and Deep Learning
byAsim Jalis
 

More from Alison B. Lowndes

PDF
Bits and pixels. Efficiency for a multi-planet society.
byAlison B. Lowndes
 
PDF
NVIDIA at Breakthrough Discuss for Space Exploration
byAlison B. Lowndes
 
PDF
Bits & Pixels using AI for Good.........
byAlison B. Lowndes
 
PDF
Exploring solutions for humanity's greatest challenges
byAlison B. Lowndes
 
PDF
Future of Skills
byAlison B. Lowndes
 
PDF
NVIDIA @ AI FEST
byAlison B. Lowndes
 
PDF
MAXSS & NVIDIA
byAlison B. Lowndes
 
PDF
Omniverse for the Metaverse
byAlison B. Lowndes
 
PDF
DataArt
byAlison B. Lowndes
 
PDF
From gaming to the metaverse
byAlison B. Lowndes
 
PDF
Breaking RSA & the internet
byAlison B. Lowndes
 
PDF
Decision-ready climate data
byAlison B. Lowndes
 
PDF
GTC 2022 Keynote
byAlison B. Lowndes
 
PDF
Talk on commercialising space data
byAlison B. Lowndes
 
PDF
NVIDIA DataArt IT
byAlison B. Lowndes
 
PDF
NVIDIA Keynote #GTC21
byAlison B. Lowndes
 
PDF
Tales of AI agents saving the human race!
byAlison B. Lowndes
 
PDF
Harnessing the virtual realm
byAlison B. Lowndes
 
PDF
EPSRC CDT Conference
byAlison B. Lowndes
 
PDF
Harnessing the virtual realm for successful real world artificial intelligence
byAlison B. Lowndes
 
Bits and pixels. Efficiency for a multi-planet society.
byAlison B. Lowndes
 
NVIDIA at Breakthrough Discuss for Space Exploration
byAlison B. Lowndes
 
Bits & Pixels using AI for Good.........
byAlison B. Lowndes
 
Exploring solutions for humanity's greatest challenges
byAlison B. Lowndes
 
Future of Skills
byAlison B. Lowndes
 
NVIDIA @ AI FEST
byAlison B. Lowndes
 
MAXSS & NVIDIA
byAlison B. Lowndes
 
Omniverse for the Metaverse
byAlison B. Lowndes
 
DataArt
byAlison B. Lowndes
 
From gaming to the metaverse
byAlison B. Lowndes
 
Breaking RSA & the internet
byAlison B. Lowndes
 
Decision-ready climate data
byAlison B. Lowndes
 
GTC 2022 Keynote
byAlison B. Lowndes
 
Talk on commercialising space data
byAlison B. Lowndes
 
NVIDIA DataArt IT
byAlison B. Lowndes
 
NVIDIA Keynote #GTC21
byAlison B. Lowndes
 
Tales of AI agents saving the human race!
byAlison B. Lowndes
 
Harnessing the virtual realm
byAlison B. Lowndes
 
EPSRC CDT Conference
byAlison B. Lowndes
 
Harnessing the virtual realm for successful real world artificial intelligence
byAlison B. Lowndes
 

Recently uploaded

DOCX
Open a Bank of America Advantage Savings Account Online.docx
bymarketing
 
PDF
“Stop Letting Banks Control You — Buy a Self Bank Account and Take Charge”.pdf
bymarketing
 
PPTX
IFRS & DIGITAL TRANSFORMATION IN ACCOUNTING.pdf
bybharath321164
 
PPTX
Economic Chapter 4 Lecture Presentation.pptx
byYunusEmreKkl
 
PDF
Unit 2 Introduction of Journal and Ledger.pdf
bybharath321164
 
PDF
''Advancing Circular Economy Practices in Nigeria Through MSMEs'' by Ismail S...
byIsmail Shittu
 
PPTX
The End of Cash: China’s CBDC & The New Global Power Play
byTHE ARCAELITH
 
PPTX
Deep Sea Mining of world ocean GR07.pptx
byabuzafor983
 
PPTX
Elasticity_of_Demand_Economics_UIET.pptx
byshubham15jee
 
DOCX
How to Safely Buy Verified Zelle Accounts Online - Fabble.docx
bycucfuc315
 
PDF
Unit 3 Introduction of Subsidiary book.pdf
bybharath321164
 
DOCX
Buy Verified Wise Account- 100% Secure- Instant Delivery.docx
bymarketing
 
PPTX
Keynesian Theory: of Unemployment .pptx
bysowmyaseco
 
DOCX
How to Safely Buy Twitter Accounts A Complete Guide in ....docx
bymarketing
 
DOCX
“Stop Letting Banks Control You — Buy a Self Bank Account and Take Charge”.docx
bymarketing
 
PDF
What We Learned Running the Commercial HackathonSF — TokenizedOilFi Goals & O...
byTatianaSF.com
 
PDF
Unit 1: Introduction to Business Accounting.pdf
bybharath321164
 
PPTX
Chapter-2.pptx.pptxSay Malaysia places a strict quota on goods imported from ...
byseowyen2003
 
PPTX
2024 Kwentong Biente (1).pptx THISIS A SHORT PRESENTATION OF THE INCOME-GENE...
byayadejasmin198014322
 
PDF
“Send Money Anytime, Anywhere – Trusted by Millions”.pdf
bymarketing
 
Open a Bank of America Advantage Savings Account Online.docx
bymarketing
 
“Stop Letting Banks Control You — Buy a Self Bank Account and Take Charge”.pdf
bymarketing
 
IFRS & DIGITAL TRANSFORMATION IN ACCOUNTING.pdf
bybharath321164
 
Economic Chapter 4 Lecture Presentation.pptx
byYunusEmreKkl
 
Unit 2 Introduction of Journal and Ledger.pdf
bybharath321164
 
''Advancing Circular Economy Practices in Nigeria Through MSMEs'' by Ismail S...
byIsmail Shittu
 
The End of Cash: China’s CBDC & The New Global Power Play
byTHE ARCAELITH
 
Deep Sea Mining of world ocean GR07.pptx
byabuzafor983
 
Elasticity_of_Demand_Economics_UIET.pptx
byshubham15jee
 
How to Safely Buy Verified Zelle Accounts Online - Fabble.docx
bycucfuc315
 
Unit 3 Introduction of Subsidiary book.pdf
bybharath321164
 
Buy Verified Wise Account- 100% Secure- Instant Delivery.docx
bymarketing
 
Keynesian Theory: of Unemployment .pptx
bysowmyaseco
 
How to Safely Buy Twitter Accounts A Complete Guide in ....docx
bymarketing
 
“Stop Letting Banks Control You — Buy a Self Bank Account and Take Charge”.docx
bymarketing
 
What We Learned Running the Commercial HackathonSF — TokenizedOilFi Goals & O...
byTatianaSF.com
 
Unit 1: Introduction to Business Accounting.pdf
bybharath321164
 
Chapter-2.pptx.pptxSay Malaysia places a strict quota on goods imported from ...
byseowyen2003
 
2024 Kwentong Biente (1).pptx THISIS A SHORT PRESENTATION OF THE INCOME-GENE...
byayadejasmin198014322
 
“Send Money Anytime, Anywhere – Trusted by Millions”.pdf
bymarketing
 

AI in the Financial Services Industry

  • 1.
    1 ALISON B LOWNDES AIDevRel | EMEA @alisonblowndes March 2019
  • 2.
  • 3.
  • 4.
  • 5.
  • 6.
    6 The day job AUTOMOTIVE Autosensors reporting location, problems COMMUNICATIONS Location-based advertising CONSUMER PACKAGED GOODS Sentiment analysis of what’s hot, problems $ FINANCIAL SERVICES Risk & portfolio analysis New products EDUCATION & RESEARCH Experiment sensor analysis HIGH TECHNOLOGY / INDUSTRIAL MFG. Mfg. quality Warranty analysis LIFE SCIENCES MEDIA/ENTERTAINMENT Viewers / advertising effectiveness ON-LINE SERVICES / SOCIAL MEDIA People & career matching HEALTH CARE Patient sensors, monitoring, EHRs OIL & GAS Drilling exploration sensor analysis RETAIL Consumer sentiment TRAVEL & TRANSPORTATION Sensor analysis for optimal traffic flows UTILITIES Smart Meter analysis for network capacity, LAW ENFORCEMENT & DEFENSE Threat analysis - social media monitoring, photo analysis
  • 7.
  • 8.
    8 NVIDIA CUDA-X AIECOSYSTEM FRAMEWORKS CLOUD DEPLOYMENT Workstation CloudServer DA GRAPH DL TRAINML DL INFERENCE Amazon SageMaker Serving Amazon SageMaker Neo Google Cloud ML CUDA-X AI CUDA Azure Machine Learning
  • 9.
    9 CUDA X ECOSYSTEM PRogrammableAcceleration across multiple Domains with one Architecture (PRADA) Specialized PerformanceEase of use FrameworksApplications Libraries Directives and Standard Languages Extended Standard Languages CUDA-C++ CUDA Fortran GPU Users Domain Specialists Problem Specialists New Algorithm Developers and Optimization Experts
  • 10.
    10 THE NEW NGC GPU-optimizedSoftware Hub. Simplifying DL, ML and HPC Workflows NGC 50+ Containers DL, ML, HPC 50+ Pre-trained Models NLP, Classification, Object Detection & more Industry Workflows Medical Imaging, Intelligent Video Analytics 10+ Model Training Scripts NLP, Image Classification, Object Detection & more Innovate Faster Deploy Anywhere Simplify Deployments
  • 11.
    11 TESLA V100 TENSOR COREGPU World’s Most Advanced Data Center GPU 5,120 CUDA cores 640 NEW Tensor cores 7.8 FP64 TFLOPS | 15.7 FP32 TFLOPS 125 Tensor TFLOPS 32 GB HBM2 @ 900GB/s | 300GB/s NVLink
  • 12.
    12 THE PATH FORWARD CPU+ Accelerator Simulation + AIFull-stack Optimization FP64 + Multi-Precision + 5.3 7.810.6 15.7 21.2 125 P100 V100
  • 13.
  • 14.
    14 PRICING DERIVATIVES COMPARING CPUTO GPU European Options SPX Monte Carlo Simulation Louis Scott — March 2018 American Options SPY Finite Difference Method Finite Difference Method Hardware Compute Time Speed Up Factor Seconds Days to Option Expiration 105 CPU - i7 5630 146 GPU- V100 0.268 544 x faster Monte Carlo 2m simulations 4 options Hardware Compute Time Speed Up Factor Seconds Days to Option Expiration 105 CPU - i7 5630 47 GPU- V100 0.18 267 x faster http://on-demand.gputechconf.com/gtc/2018/video/S8123/
  • 15.
  • 16.
  • 17.
  • 18.
    18 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
  • 19.
    NATURAL LANGUAGE PROCESSING FORSIGNAL GENERATION ON NEWS DATA
  • 20.
    WORD EMBEDDINGS GLoVe –Global Vectors for Word Representation, utilizes the word-to-word co- occurrence statistics from a given corpus.
  • 22.
    22 Algorithmic Trading usingDeep Autoencoder based Statistical Arbitrage NVIDIA Deep Learning Institute
  • 23.
    23 Moving Average –Mean Reversion
  • 24.
  • 25.
  • 26.
    26 Snorkel https://github.com/HazyResearch/snorkel Automation of labellingdata A system for rapidly creating, modeling, and managing training data, For domains in which large labelled training sets are not available or easy to obtain. Learning, essentially, which labelling functions are more accurate than others—and then using this to train a DNN A general framework for many weak supervision techniques.
  • 27.
    27 AUTO ML: AICREATING AI http://automl.chalearn.org/
  • 28.
  • 29.
  • 30.
    30 WHAT IS ATENSOR? And why do they flow?
  • 31.
    31 WHAT IS ATENSOR? And why do they flow? Scalar is a list of numbers with 0 indices (length 1) a Vector is a list of numbers with 1 index of length k a[k] Matrix is a list of numbers with 2 indices of length r,c a[r,c] A Tensor is a list of numbers with n indices of length n1, n2, …, nm a[n1..nm] an n-dimensional array
  • 32.
    32 Origin of NeuralNetworks Input Output
  • 33.
    33 A Simple Neuron InputOutput Neuron x1 w2x2 y w1x1 x2
  • 34.
    34 Neural Network Basics 2ndstep: sum 1st step: Activations * Weights 3rd step: activate
  • 35.
    35 Combining Neurons x1 x2 x3 x4 x5 —Additional neuronscan be added to create a layer —Multiple layers can also be added, resulting in input, hidden, and output layers —Expanding the neural network size creates additional predictive power —In feed forward neural networks, neurons are fully connected to surrounding layers y Input Layer Output Layer Hidden Layers
  • 36.
    36 Deep Neural Networks(DNNs) x1 x2 x3 x4 x5 Input Layer Output LayerMany Hidden Layers y
  • 37.
    37 Image “Volvo XC90” Imagesource: “Unsupervised Learning of Hierarchical Representations with Convolutional Deep Belief Networks” ICML 2009 & Comm. ACM 2011. Honglak Lee, Roger Grosse, Rajesh Ranganath, and Andrew Ng. CONVOLUTIONAL NEURAL NETWORKS
  • 38.
    38Yann Le Cun(FaceBook) ISSCC, Feb 2019
  • 39.
  • 40.
  • 41.
    41 Li et al,University of Maryland & US Naval Academy https://arxiv.org/pdf/1712.09913.pdf
  • 42.
    42 AUTOENCODERS UNSUPERVISED feature learning Sparse Representation Training Data Reconstruction EncoderDecoder Minimize Reconstruction Error - InputLayer HiddenLayer HiddenLayer HiddenLayer BottleneckLayer HiddenLayer HiddenLayer HiddenLayer OutputLayer
  • 43.
    43 Long short-term memory(LSTM) Hochreiter (1991) analysed vanishing gradient “LSTM falls out of this almost naturally” Gates control importance of the corresponding activations Training via backprop unfolded in time LSTM: input gate output gate Long time dependencies are preserved until input gate is closed (-) and forget gate is open (O) forget gate Fig from Vinyals et al, Google April 2015 NIC Generator Fig from Graves, Schmidhuber et al, Supervised Sequence Labelling with RNNs
  • 44.
  • 45.
  • 46.
  • 47.
  • 48.
    48 Tabish Rashid etal, Oxford/DeepMind QMIX: https://arxiv.org/pdf/1803.11485.pdf
  • 49.
    49 DEEPMIND ALPHA* 1.0=> 2.0 => 0.0 * ..a deeply structured hybrid (Gary Marcus, Jan 2018)
  • 50.
  • 51.
  • 52.
  • 53.
    53 COMPUTATIONAL SCALE REQUIRED 3million labeled images 1 DGX-1 trains 300k labeled images on 1 DNN in 1 day 10 DNNs required for self-driving 10 parallel experiments at all times 100 DGX-1 per car
  • 54.
    54 INNOVATIONS & KEYMETRICS DISTILLED NVIDIA SATURNV REFERENCE Design Best Practices Lessons learned from building the worlds largest AI Infrastructure Reference Architectures Partner Reference Architectures, Scale up racks Product Innovation & Quality Rapid exploration & resolution of customer issues
  • 55.
  • 56.
    Illuminate Deep Networks [email protected] Original Image Most Certain Uncertain Layer273: Visualizing expected vs actual outputs Uncertain Inputs of two filters which are most uncertain
  • 57.
  • 58.
    58 DATA SCIENCE IN FINANCE AlphaStock Identification Analyze Consumers’ Behavior Anti-fraud API Service Insurance Campaign And Conversion Analysis Credit Card Application Approval Customer service chatbots/routing Claim Fraud Detection Evaluate Create Worthiness Fraud And Credit Risk Analysis Fraud Detection Hedge Fund Management Risk evaluation
  • 59.
    59 Financial investment forecastinginvolves processing vast amounts of data to derive predictions that can outperform the market. SpaceKnow is using GPU to extract global macro and micro-economic activity that helps build high-performance portfolios AI-POWERED INVESTMENT ALPHA
  • 60.
    60 EQUIFAX Equifax now hasNeuroDecision Technology “NeuroDecision Technology (NDT) is the first regulatory-compliant machine learning credit scoring system reviewed by regulators and credit scoring experts. This technology develops a neural network model that improves performance and accuracy, which gives customers the ability to make more informed business decisions when assessing risk.” https://investor.equifax.com/news-and-events/news/2018/03-26-2018-143044126 “The executive noted that the neural net has improved its ability to make predictive models by as much as 15 percent.” https://www.pymnts.com/innovation/2017/equifax-uses-deep-neural-machine-learning-to-improve-credit-scoring/ Neural nets can be 15% better for prediction
  • 61.
    61 FRAUD DETECTION Incumbent firmslooking at deep structured history of customers to do supervised learning on fraud / no fraud inline with transactions; some use of autoencoders and other networks to do latent space clustering to identify fraud after the fact. Mixed use of raw transactions over time (RNN) and transaction summary vectors (RNN and CNN) to train models. CNN or fully connected has advantages in-line w.r.t. transcation latency. DL shown to be able to dramatically reduce false positives in transactional fraud! Also use cases around Speech to Text transcription for insider trading etc. Commercial applications challenged today by industry jargon, accents, non-English or Mandarin but these challenges can be overcome. Discussion
  • 62.
    62 CREDIT SCORING Likelihood ofdefault. Incumbent firms mostly trying to mine their existing data to more accurately predict repayment/prepayment/default behavior. Can leverage DL to find structure that ML models can then be retrained to exploit in an explainable manner. Challenger / startups looking to use DL to combine multiple data sources to develop models that produce either scores correlated with existing scores or their own “behavioral score”. Frequent challenges around data bias introduced by past credit criteria and practices! Explainability is a primary concern of many firms, especially under GDPR. Techniques including LIME, latent space clustering, nearest training set example, and other emerging network visualization research is being followed closely. Discussion
  • 63.
    63 STAC A2™ BENCHMARK STACA2 Benchmark Developed by banks Macro and micro, performance and accuracy Pricing and Greeks for American exercise basket option, correlated Heston dynamics, Longstaff Schwartz Monte Carlo Independently Audited Results GPU Solution The first system to handle the baseline problem size in "real time" (less than one second) Please see http://www.stacresearch.com/a2 for more details of the STAC Benchmark Also see https://devblogs.nvidia.com/parallelforall/american-option-pricing-monte-carlo-simulation/ for more details on Longstaff-Schwartz Monte Carlo on GPUs
  • 64.
    64 DEEP LEARNING FOR CUSTOMERSERVICE OPERATIONS AI-assisted and fully autonomous customer interactions. Integrates with leading customer service software and communications channels Human AgentsHuman Customer Agent Console Deep Neural Networks AI Model is trained on historical chat logs and customer service transcripts. DigitalGenius TensorFlow and Pytorch Supervised Unsupervised
  • 65.
    65 GPU-ACCELERATED BERT State-of-the ArtNatural Language Processing BERT AVAILABLE ON NGC SUPER-HUMAN QUESTION ANSWERING 280X FASTER TRAINING REAL-TIME INFERENCE Question Answering Translation Dialog Sentiment Analysis Summarizing 86.6 89.3 90.1 91.8 85 87 89 91 93 RM Reader BERT CPU Server DGX-2 CPU Server T4 Training 52 Hours 13 Mins 230 ms 18 ms Inference QANet nlnet Human Source: Question answering accuracy on SQUAD 1.1 for non-ensemble models
  • 66.
    66 GPU ACCELERATED MACHINELEARNING FOR BOND PRICE PREDICTION 100 data elements per trade: Trade size / historical Coupon rate / time to maturity Bond rating Trade type buy/sell Reporting delays Current yield / yield to maturity Launch as many CUDA threads as there are data elements leverage 5120 Cores on V100 to run multiple Kernels in parallel http://on-demand-gtc.gputechconf.com/gtcnew/on-demand- gtc.php?searchByKeyword=s8655&searchItems=session_id&sessionTopic=&sessionEvent=&sessionYear=&sessionFormat=&submit=&select= NEARLY 10X SPEED UP OVER CPU IMPLEMENTATION Bond trading price 20 21 22 23 24 25 p SpeedupoverCPU 0 2 4 6 8 10 Unoptimized CUDA Optimized CUDA
  • 67.
    67 APPLYING DEEP LEARNINGTO FINANCIAL MARKETS WITH NEWS DATA Recording: http://on-demand.gputechconf.com/gtc/2017/video/s7696-andrew-tan-applying-deep-learning-to-financial-market-signal-identification-with-news- data.mp4 PDF: http://on-demand.gputechconf.com/gtc/2017/presentation/s7696_Andrew-Tan_ FinancialMarketSignalIdentification.pdf
  • 68.
    68 GPU ACCELERATED COMPUTING HPCAI ML VISUALIZATION CLARA NVIDIA AI RAPIDS ISAAC DRIVE METROPOLIS WORKSTATIONS SERVERS CLOUD CUDA & GPU COMPUTING ARCHITECTURE NVIDIA GPU CLOUD TESLA DGXTEGRA
  • 69.
    69 APPS & FRAMEWORKS CUDA-X NVIDIA LIBRARIES NVIDIADATA CENTER PLATFORM Single Platform Drives Utilization and Productivity VIRTUAL GPU CUDA & CORE LIBRARIES - cuBLAS | NCCL DEEP LEARNING cuDNN HPC cuFFTOpenACC +550 Applications Amber NAMD CUSTOMER USE CASES VIRTUAL GRAPHICS Speech Translate Recommender SCIENTIFIC APPLICATIONS Molecular Simulations Weather Forecasting Seismic Mapping CONSUMER INTERNET & INDUSTRY APPLICATIONS ManufacturingHealthcare Finance GPUs & SYSTEMS SYSTEM OEM CLOUDTESLA GPU NVIDIA HGXNVIDIA DGX FAMILY MACHINE LEARNING cuMLcuDF cuGRAPH cuDNN CUTLASS TensorRTvDWS vPC Creative & Technical Knowledge Workers vAPPS +600 Applications DX/OGL
  • 70.
    70 TRADITIONAL DATA SCIENCE CLUSTER Workload Profile: •192GB mortgage data set • 16 years, 68 quarters • 34.7 Million single family mortgage loans • 1.85 Billion performance records • XGBoost training set: 50 features 300 Servers | $3M | 180 kW
  • 71.
    71 GPU-ACCELERATED MACHINE LEARNING CLUSTER 1 DGX-2 |10 kW 1/8 the Cost | 1/15 the Space 1/18 the Power DGX-2 and RAPIDS for Predictive Analytics 0 2,000 4,000 6,000 8,000 10,000 20 CPU Nodes 30 CPU Nodes 50 CPU Nodes 100 CPU Nodes DGX-2 5x DGX-1 End-to-End
  • 72.
    2 PFLOPS |512GB HBM2 | 10 kW | 350 lbs NVIDIA DGX-2
  • 73.
    73©2018 VMware, Inc. OPTIMIZED SOFTWARE FASTER DEPLOYMENTS Eliminatesinstallations. Simply Pull & Run the app Key DL frameworks updated monthly for perf optimization Empowers users to deploy the latest versions with IT support Better Insights and faster time-to-solution NGC – SIMPLIFYING AI & HPC WORKFLOWS ZERO MAINTENANCE HIGHER PRODUCTIVITY EMBEDDING EXPERTISE Deliver greater value, faster
  • 74.
  • 75.
    75 GET STARTED WITHNGC Deploy containers: ngc.nvidia.com Learn more about NGC offering: nvidia.com/ngc Technical information: developer.nvidia.com Explore the NGC Registry for DL, ML & HPC
  • 76.
    76 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
  • 77.
    RAPIDS RAPIDS GPU Accelerated End-to-EndData Science RAPIDS is a set of open source libraries for GPU accelerating data preparation and machine learning. OSS website: rapids.ai GPU Memory Data Preparation VisualizationModel Training cuGraph Graph Analytics cuML Machine Learning cuDF Data Preparation
  • 78.
    78 DATA SCIENCE WORKFLOWWITH RAPIDS Open Source, End-to-end GPU-accelerated Workflow Built On CUDA DATA DATA PREPARATION GPUs accelerated compute for in-memory data preparation Simplified implementation using familiar data science tools Python drop-in Pandas replacement built on CUDA C++. GPU-accelerated Spark (in development) PREDICTIONS
  • 79.
    79 DATA SCIENCE WORKFLOWWITH RAPIDS Open Source, End-to-end GPU-accelerated Workflow Built On CUDA MODEL TRAINING GPU-acceleration of today’s most popular ML algorithms XGBoost, PCA, Kalman, K-means, k-NN, DBScan, tSVD … DATA PREDICTIONS
  • 80.
    80 DATA SCIENCE WORKFLOWWITH RAPIDS Open Source, End-to-end GPU-accelerated Workflow Built On CUDA VISUALIZATION Effortless exploration of datasets, billions of records in milliseconds Dynamic interaction with data = faster ML model development Data visualization ecosystem (Graphistry & OmniSci), integrated with RAPIDS DATA PREDICTIONS
  • 81.
  • 82.
    82 PILLARS OF RAPIDSPERFORMANCE CUDA Architecture NVLink/NVSwitch Integrated Software Massively Parallel Processing High Speed Connecting between GPUs for Distributed Algorithms Fully Integrated Software and Hardware for Instant Productivity NVSwitch 6x NVLink CUDA PYTHON APACHE ARROW on GPU Memory DASK cuDNN RAPIDS cuMLcuDF DL FRAMEWORKS
  • 83.
    83 NEW TURING TENSORCORE MULTI-PRECISION FOR AI INFERENCE & SCALE-OUT TRAINING 65 TFLOPS FP16 | 130 TeraOPS INT8 | 260 TeraOPS INT4
  • 84.
    84 TENSOR CORE AUTOMATICMIXED PRECISION Over 3x Speedup With Just Two Lines of Code TOOLS AND LIBRARIES MAINTAIN NETWORK ACCURACY TRAINING SPEEDUP OVER 3X INFERENCE SPEEDUP OVER 4X 0 20000 40000 60000 80000 100000 PyTorch GNMT TotalTokens/sec FP32 Mixed 3.4X 1xV100 0 2000 4000 6000 8000 TensorRT ResNet50 Images/sec FP32 INT8 Mixed 4.4X 7ms Latency 1xV100 Tensor Core Journey Page Github Profiler Tools
  • 85.
    85 Circa 2000 -Torch7 - 4th (using odd numbers only 1,3,5,7) Web-scale learning in speech, image and video applications Maintained by top researchers including Soumith Chintala - Research Engineer @ Facebook All the goodness of Torch7 with an intuitive Python frontend that focuses on rapid prototyping, readable code & support for a wide variety of deep learning models. https://pytorch.org/2018/05/02/road-to-1.0.html
  • 86.
    86 Apex - APyTorch Extension ● Goal: Raise PyTorch customer awareness and increase adoption of NVIDIA Tensor Cores ● Content: Provide an easy to use set of utility functions in PyTorch for mixed-precision optimizations ● Benefit: Few lines of code to achieve improved training speed while maintaining accuracy and stability of single precision (Tensor Cores) ● Target audience: Deep learning researchers and developers of PyTorch with NVIDIA Volta ● Key Features: AMP (Auditor for mixed-precision) and Optimizer Wrapper (Dynamic loss scaling and master parameters) ● Teams Involved: Leading NVIDIA PyTorch team and collaboration with external FB PyTorch team Overview
  • 87.
  • 88.
    88 DALI Full input pipelineacceleration including data loading and augmentation Drop-in integration with direct plugins to frameworks – MxNet, TensorFlow, PyTorch Portable workflows through multiple input formats and configurable graphs Unblock CPU with GPU-accelerated DL pre-processing library Version 0.1 supports: • Resnet-50 image classification & segmentation training • Input formats – JPEG, LMDB, RecordIO, TFRecord • Python APIs to define, build and run an input pipelineAvailable as open source - https://github.com/NVIDIA/DALI • DALI Samples & Tutorial: https://github.com/NVIDIA/DALI/blob/master/examples/Getting%20started.ipynb • nvJPEG (Webpage, Documentation) - https://developer.nvidia.com/nvjpeg
  • 89.
    89 AI INFERENCE NEEDSTO RUN EVERYWHERE Training InferencingDNN Model
  • 90.
  • 91.
    91 GTC-Pre Announce duringkeynote Customer sign-up page will go live, post-keynote ANNOUNCING NVIDIA T4 ON AMAZON AWS
  • 92.
  • 93.
    93 Learn more here: https://nvidia.com/data-center-inference https://docs.nvidia.com/deeplearning/sdk/inference-release-notes/index.html Getthe ready-to-deploy container with monthly updates from the NGC container registry: https://ngc.nvidia.com/catalog/containers/nvidia%2Ftensorrtserver Open source GitHub repository: https://github.com/NVIDIA/tensorrt-inference-server LEARN MORE AND DOWNLOAD TO USE
  • 94.
    94d e ve l o p e r . n v i d i a . c o m
  • 95.
    Fundamentals Accelerated Computing Game Development& Digital Content Finance NVIDIA DEEP LEARNING INSTITUTE Online self-paced labs and instructor-led workshops on deep learning and accelerated computing Take self-paced labs at www.nvidia.co.uk/dlilabs View upcoming workshops and request a workshop onsite at www.nvidia.co.uk/dli Educators can join the University Ambassador Program to teach DLI courses on campus and access resources. Learn more at www.nvidia.com/dli Intelligent Video Analytics Healthcare Robotics Autonomous Vehicles Virtual Reality
  • 96.
    96 NVIDIA INCEPTION PROGRAM Accelerates AI startupswith a boost of GPU tools, tech and deep learning expertise Startup Qualifications Driving advances in the field of AI Business plan Incorporated Web presence Technology DL startup kit* Pascal Titan X Deep Learning Institute (DLI) credit Connect with a DL tech expert DGX-1 ISV discount* Software release notification Live webinar and office hours *By application Marketing Inclusion in NVIDIA marketing efforts GPU Technology Conference (GTC) discount Emerging Company Summit (ECS) participation+ Marketing kit One-page story template eBook template Inception web badge and banners Social promotion request form Event opportunities list Promotion at industry events GPU ventures+ +By invitation www.nvidia.com/inception
  • 97.