Upgrade to Pro — share decks privately, control downloads, hide ads and more …

NVIDIA Deep Learning Platform and latest inform...

NVIDIA Deep Learning Platform and latest information by Takeshi Izaki, Nvidia - TMLS #4 

Deep Learning has been evolved rapidly recent years. In particular it has started being utilized in various fields over the past year. Today’s Deep Learning solution relies almost exclusively on NVIDIA Deep Learning Platform including GPU. In this session I will talk about overview of this platform and some latest information.

More Decks by Tokyo Machine Learning Society

Other Decks in Technology

Transcript

  1. An amazing year in ai AlphaGo Rivals a World Champion

    Microsoft & Google “Superhuman” Image Recognition Microsoft “Super Deep Network” Berkeley’s Brett One network, everything robotics Deep Speech 2 One network, 2 languages A New Computing Model Hits Pop Culture
  2. A new computing model 0% 10% 20% 30% 40% 50%

    60% 70% 80% 90% 100% 2009 2010 2011 2012 2013 2014 2015 2016 Traditional CV Deep Learning ImageNet Deep Learning Object Detection DNN + Data + HPC Traditional Computer Vision Experts + Time Deep Learning Achieves “Superhuman” Results
  3. exabytes of content produced daily 10M Users 40 years of

    video/day 1.7M Broadcasters Users watch 1.5 hours/day 6B Queries/day 10% use speech 270M Items sold/day 43% on mobile devices 8B Video views/day 400% growth in 6 months 300 hours of video/minute 50% on mobile devices
  4. Deep Learning Use Cases Image Classification, Object Detection, Localization Face

    Recognition Speech & Natural Language Processing Medical Imaging & Interpretation Autonomous Machines Recommendation
  5. Why is Deep learning hot now? “Google’s AI engine also

    reflects how the world of computer hardware is changing. (It) depends on machines equipped with GPUs… And it depends on these chips more than the larger tech universe realizes.” DNN GPU BIG DATA
  6. Ex. Typical network Many Matrix Multiply with many training data

    Purpose Face recognition Training Data 10M~100M Images Network Architecture 10 Layers 1B Parameters Learning Algorithm ~30 ExaFLOPS ~30 GPU Days
  7. 8 Machine Learning evolution by NVIDIA TESLA GPU GOOGLE BRAIN

    APPLICATION – DEEP LEARNING BEFORE TESLA AFTER TESLA Cost $5,000K $200K Server# 1,000 server 16 Tesla server Power 600 KW 4 KW Performace 1x 6x
  8. NVIDIA GPU for hyperscale 10X Speed up | 20 images/s/W

    Cloud Services Powered by AI TESLA M40 + TESLA M4
  9. NVIDIA DGX-1 World’s First Deep Learning Super Computer Engineered for

    Deep Learning 170 TF FP16 8 x Tesla P100 NVLink Hybrid Cube Mesh Accelerates major AI frameworks
  10. NVIDIA Deep Learning Platform Computer Vision Speech and Sound Behavior

    Object Detection Voice Recognition Translation Recommendation Engines Sentiment Analysis cuDNN cuBLAS cuSPARSE NCCL cuFFT Mocha.jl Image Classification Deep Learning SDK Framework Application GPU Platform Cloud GPU Tesla P100 Tesla K80/M40/M4 Jetson TX1 Server DGX-1 GIE DRIVEPX2 Deep Learning Math Library Multi-GPU Communication
  11. Manufacturing Ref. MONOist Achieved 90% success rate only after 8hours

    Learning (The same rate as human experts) DEEP LEARNING DAY2016
  12. Finance Algo trading Over +0.5% -0.5%~+0.5% Under -0.5% 1Hour Predict

    time 14:00 Target time 15:00 Up Down In Range DEEP LEARNING DAY2016