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

O'reilly AI Conference

GDP Labs
October 25, 2017

O'reilly AI Conference

GDP Labs

October 25, 2017
Tweet

More Decks by GDP Labs

Other Decks in Technology

Transcript

  1. What “It is an AI Event that focus on real-world

    implementations.” GDP Labs Confidential
  2. 1. Building Unbiased AI 2. Accelerating AI 3. Active Learning

    and Transfer Learning 4. Object Detection GDP Labs Confidential
  3. 1. Building Unbiased AI 2. Accelerating AI 3. Active Learning

    and Transfer Learning 4. Object Detection GDP Labs Confidential
  4. Diversity Crisis in AI “We’ve already seeing society’s racial and

    gender biases being encoded into software that uses AI when built by such a homogeneous group.” Building Unbiased AI GDP Labs Confidential
  5. Ignoring - Failing to max a product to different groups

    - Failing to attract potential users Building Unbiased AI GDP Labs Confidential
  6. 1. Building Unbiased AI 2. Accelerating AI 3. Active Learning

    and Transfer Learning 4. Object Detection GDP Labs Confidential
  7. SSD model designed and optimized for PASCALVOC/MSCOCO dataset Know Your

    Model Provenance Accelerating AI GDP Labs Confidential
  8. Smart hospital will generate over 3,000 GB PER DAY Self

    driving cars will generate over 4,000 GB PER DAY … EACH The average internet user will generate ~1.5 GB OF TRAFFIC PER DAY A connected plane will generate over 40,000 GB PER DAY A connected factory will generate over 1,000,000 GB PER DAY Invest in Data Manual annotations Training data Accelerating AI GDP Labs Confidential
  9. 1. Building Unbiased AI 2. Accelerating AI 3. Active Learning

    and Transfer Learning 4. Object Detection GDP Labs Confidential
  10. Active and Transfer Learning Active Attain good learning performance without

    demanding too many labeled data GDP Labs Confidential
  11. Human Active AI Classifier Output Human Annotation Confident N ot

    Confident Active Learning Active and Transfer Learning GDP Labs Confidential
  12. 1. Building Unbiased AI 2. Accelerating AI 3. Active Learning

    and Transfer Learning 4. Object Detection GDP Labs Confidential
  13. Traditional Machine Learning to Deep Learning Object Detection Feature Engineering

    challenges: - Human Scaling - Computational Scaling GDP Labs Confidential
  14. Object Detection Feature Learning challenges: - Need large datasets -

    Model become complex Traditional Machine Learning to Deep Learning GDP Labs Confidential
  15. Region Proposed Approach Object Detection - Run through image to

    detect regions - Process regions extensively: - Bounding boxes - Predict class GDP Labs Confidential
  16. Direct Classification Object Detection - BB and class are predicted

    directly with a single network GDP Labs Confidential
  17. Training Vision Model Challenges Object Detection Batch size epochs Top-1

    Accuracy Hardware Cost ($) Time 256 90 73.0% 1 DGX station 129,000 21h 8192 90 72.7% 1 DGX station 129,000 21h 8192 90 72.7% 32 DGX station 4.1 million 1h 32K 90 72.4% 512 KNLs 1.2 million 1h GDP Labs Confidential