Download as PDF, PPTX
Uploaded byAlison B. Lowndes
Phi Week 2019
The document discusses several pillars for national AI initiatives, including establishing AI centers of excellence, reskilling the workforce, and investing in key industries to drive growth and solve economic and social challenges. It also outlines different approaches for designing and optimizing AI systems, such as using GANs and GPU-accelerated simulations. Overall, the document promotes the development and application of AI through collaboration between universities, industry, and government.
More Related Content
Phi Week 2019
- 1. 1 ALISON B LOWNDES AIDevRel | EMEA @alisonblowndes Obtaining & accelerating insight from downstream data with GPUs
- 3. Key ”Key decisions andinvestments are needed in the next 12 months” David Jensen, UN
- 4. 4 AI CENTERS OFEXCELLENCE StartingtheAI economyrequiresknowledgeandskills, which requirecomputinginfrastructure,which canbe afforded only ifthereisdemand.Governmentcan kick- startthisvirtuouscycleby providingaccessto large-scale AI computinginfrastructureto its universities,startups, and industry. SKILLINGAND RESKILLING AI will acceleratejob growth,butharnessingitrequires specific skillsandexpertise.Countriesneed to investin trainingandeducation programsthatadd AI skillsto their workforce. INDUSTRY SOLUTIONS By investinginkey industries,countries havea unique opportunity to drivegrowth and solvetheir greatestsocial and economicchallenges. Reskilling Workforce Basic ResearchIndustry-Standard Platform Rich Software Ecosystem State-of-the-Art AI Computing University-Industry Collaboration Vibrant Startup Community Hyper-Productive Manufacturing Safe, Efficient Transportation Affordable, Accessible Healthcare Smart, Safe Cities PILLARS OF NATIONAL AI INIATIVES
- 7.
- 8. 8 Detection Planning Acceleration Assimilation EnhancementParametrization AugmentationPrediction Optimize Disaster Planning
- 9. 9 Detection Planning Acceleration Assimilation EnhancementParametrization AugmentationPrediction Develop Autonomous Agents for Vehicles and Sensors
- 10. 10 Detection Planning Acceleration Assimilation EnhancementParametrization AugmentationPrediction Monitor EnvironmentalChange drought flooding deforestation urbanification melting glaciers sea-level change
- 11.
- 12. 12 CONDITIONAL GANS Generates outputconsistent with the training set Generator (Regressor) Discriminator (Classifier) Generated Target • If the output is under-constrained your output will look fuzzy. • It won’t capture the true peaks and valleys • We can overcome this problem using a conditional GAN • Generates output matching the training set Loss fcn
- 13. 13 NOISE ELIMINATION IN GLOBALCLIMATE OBSERVATIONS HadCRUT4 data is one of the most widely used long term data sets for climate. But it is filled with missing values. By using a noise2noise algorithm, we can fill in the missing values, making a complete, gridded product, far more accurately than the usual approach of linear interpolation. Freie Universität Berlin & NVIDIA
- 14. 14 MINIMISE ENERGY FOOTPRINT MINIMISE TIMETO COMPUTE MAINTAIN ACCURACY & NUMERICAL STABILITY Results with NVLINK/NVSWITCH/DGX-2/V100 [email protected]
- 15. Global Modeling andAssimilation Office gmao.gsf c.nasa.govGMAO National Aeronautics and Space Administration XGBoost for simulating atmospheric chemistry [email protected]
- 17.
- 18. 18 DESIGNING SPACECRAFTS WITH GPUACCELERATED SIMULATIONS Large scale aerodynamic simulations are necessary to help define the shape and performance of aircraft, spacecrafts, automotive vehicles and others. NASA Langley Research Center develops FUN3D computational fluid dynamics software to simulate fluid flow for a broad range of aerodynamics applications. This single application consumes more cycles at NASA’s supercomputers than any other application. GPU acceleration enables over 23X higher performance than CPU servers while running these simulations. The performance on GPUs scales very well to enable efficient computation of the largest and the most complex simulations. NASA has shown that a thousand GPU servers on Summit supercomputer can do the work of over a million CPU cores for a fraction of the energy costs. Images: SLS)Block 1B booster separation flowfield simulated using NASA's FUN3D code. . Credit: Jamie Meeroff, Henry Lee, NASA/Ames
- 20.
- 21.
- 22.
- 23.
- 24. 24NVIDIA CONFIDENTIAL. DONOT DISTRIBUTE. LEARNING STRATEGY • Evolutionary algorithms (EA): population-based optimizer for black-box functions, combination and mutation to generatenew architectures • Gradient-based: based on continuousrelaxation of the architecture representation, allowing search of the architecture with gradient descent • Reinforcement learning (RL): an agent (network generator) interacts with an environment (search space) with the goal of maximizingrewards (accuracy on validation data)
- 25. 25 KAOLIN - A Pytorchlibrary for 3D DL - Supports a wide range of 3D data representations - Convenient dataloading/preprocessing/conversions - Large collection of 3D neural nets to choose from - Optimized implementations - Omniverse-Kit integration for easy rendering, interactive visualization, and much more. https://gitlab- master.nvidia.com/Toronto_ DL_Lab/kaolin
- 26. RAPIDS RAPIDS.ai GPU Accelerated End-to-EndData Science RAPIDS is a set of open source libraries for GPU accelerating data preparation and machine learning. GPU Memory Data Preparation VisualizationModel Training cuGraph Graph Analytics cuML Machine Learning cuDF Data Preparation
- 27.
- 28. 28 ISAAC OPEN SDK IsaacRobot Engine – Modular robot framework | Isaac Sim - Virtual roboticslaboratory Isaac Gym – Reinforcement learning simulator | Isaac Robot Apps – Kaya, Carter and Link Available at developer.nvidia.com/isaac-sdk CARTER (Xavier)KAYA (Nano) LINK (Multi Xavier) JETSONNANO ISAAC OPEN TOOLBOX Sensor and Actuator Drivers Core Libraries GEMS Reference DNN Tools CUDA-X Isaac Robot Engine JETSONTX2 JETSONAGX XAVIER Isaac Sim Isaac Gym
- 30. 30 JETSON NANO €99 NVIDIACUDA-X AI Computer CUDA-X acceleration stack | High-resolution sensor support | Runs all CUDA-X AI models
- 31.
- 32. 32 A WORKSTATION FORDATA SCIENTISTS Powered by NVIDIA GPU and CUDA-X AI Dual Quadro RTX 8000 with 96 GB Memory | Pre-installed for CUDA-X Accelerated DataScience — RAPIDS, TensorFlow, PyTorch, Caffe, AnacondaDistribution | 10X Faster 0 2 4 6 8 10 Data Prep Training End to end Acceleration of FSI Data Science Workflow 2x RTX8000 1x RTX8000 CPU
- 33.
- 34. 2 PFLOPS |512GB HBM2 | 10 kW | 350 lbs NVIDIA DGX-2
- 35. 35 NVIDIA GPUs INTHE CLOUD AVAILABLE ON-DEMANDFROM THE TOP CLOUD SERVICE PROVIDERS • Immediate access to NVIDIA GPU infrastructure for data science in the cloud • Wide variety of deployment and management options using containers, Kubernetes, Kubeflow, support for cloud native services, and more
- 37. 37 FIVE ROADS TOGPU COMPUTING GPU Libraries ______________ Drop-in replacement for existing libraries cuBLAS, CUDA Math, cuSPARSE, cuRAND, cuSOLVER, nvGRAPH, cuDNN, cuFFT, Thrust OPEN-ACC ______________ Comment-based directives in C / C++ / Fortran Single source code parallelization for multiple architectures CUDA ______________ Parallel Programming Model for GPUs in C, C++, Fortran, Python, MATLAB Specialized Kernels for general purpose GPU RAPIDS ______________ GPU Acceleration of Traditional Machine Learning Accelerate Scikit-Learn style ML algorithms DEEP LEARNING ______________ GPU accelerated deep learning frameworks TensorFlow, Pytorch Build GPU-accelerated functions directly from data
- 38. 38 PURPOSE-BUILT AI SUPERCOMPUTERS AIWORKSTATION AI DATA CENTER Universal SW for Deep Learning Predictable execution across platforms Pervasive reach NGC DL SOFTWARE STACK The Essential Instrument for AI Research DGX-1 The Personal AI Supercomputer DGX Station The World’s Most Powerful AI System for the Most Complex AI Challenges DGX-2
- 39.
- 40. 40 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
- 41. 41 Available to InstructorsNow! developer.nvidia.com/teaching-kits Robotics Teaching Kit with ‘Jet’ - ServoCity D E E P L E A R N I N G
- 42. 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
- 43. 43 CONNECT Connect with expertsfrom NVIDIA, GE Healthcare, NSF Carnegie Mellon, Google, and other leading organizations LEARN Gain insight and valuable hands-on training through over 100 sessions DISCOVER See how GPU technologies are creating amazing breakthroughs in important fields such as deep learning INNOVATE Explore disruptive innovations that can transform your work Don’t miss the premier AI conference. nvidia.com/en-us/gtc-dc/ Join us at GTC DC | Use VIP code NVALOWNDES for 25% off | Govt. attends free November 4 - 6, 2019 | Washington, D.C.
- 44.