Client-Side Deep Learning
?
1 like?8,338 views
The document discusses client-side deep learning and introduces MPSCNN, a library that allows running convolutional neural networks on iOS devices using Metal Performance Shaders. MPSCNN can import trained models from frameworks like TensorFlow and run them to perform tasks like object detection on images at 60 times per second. Client-side deep learning could enable new mobile applications for areas like self-driving cars, AI assistants, and cancer detection by taking advantage of on-device processing power.
Client-Side Deep Learning
- 1. Client-Side Deep Learning Shuichi Tsutsumi @shu223 try! Swift Tokyo 2017
- 2. CNN APIs in Metal Performance Shaders
- 3. Samples
- 6. New function for Object Detection ?!
- 7. CNN APIs in Metal Performance Shaders
- 8. CNN APIs in Metal Performance Shaders CNN APIs in Metal Performance Shaders MPSCNN
- 9. MPSCNN ? Implement & Run CNN on iOS devices with Metal
- 13. "
- 14. Train RunTrained Params MPSCNN? TensorFlow ? Chainer ? etc… ? Any file format ? The shapes of tensors are defined.
- 15. Image Result 60 times / sec Why Client-Side? xUsers # #
- 16. $
- 17. I don’t explain today about… ? What is CNN? ? What is Deep Learning? - (Deep Learning is a “deep” neural network.)
- 18. Self-Driving Cars AIphaGo Cancer Detection
- 19. Self-Driving Cars AIphaGo Cancer Detection
- 20. Self-Driving Cars AIphaGo Cancer Detection Deep Learning
- 21. Self-Driving Cars AIphaGo Cancer Detection Deep Learning is
- 22. Self-Driving Cars AIphaGo Cancer Detection Deep Learning is “Something great”
- 23. Self-Driving Cars AIphaGo Cancer Detection “Something great” is available on iOS%
- 24. Demo
- 26. Thank you!