Deep Learning from the inside
summary
An explanation of how deep-learning inference works in practice, focusing on computation graphs and their relationship to models, why efficient inference is hard, and optimization techniques. The talk gives concrete examples using the nGraph library and the OpenVINO framework to demonstrate optimizations for model inference.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 5:26Models as math functions
- 10:00Graphical model representation
- 18:36Why inference is hard
- 21:05Naive optimization limits
- 23:33Kernels and hardware optimization
- 27:04Hardware-independent optimizations
- 31:56Graph optimization examples
- 36:10Model quantization
- 41:46Summary and resources
related talks
16mPreview: Graph NNs
26mDeep Learning in OpenCV
42mSunil Kumar Vuppala of Philips Research at Machine Learning Developers Summit 2019
51mDeep Learning - What is it and What It Can Do For You
03mInterview
39mA Fireside Chat with Turing Award Winner Geoffrey Hinton, Pioneer of Deep Learning
34mPanel Discussion: Training and Inference at Planet Scale
27mDemystifying Deep Neural Networks