Proj CDeepFuzz Paper Reading: Automatic differentiation in ML: Where we are and where we should be going
Abstract
本文:Myia
Github: https://github.com/mila-iqia/myia
Task: Review automatic differentiation for array programming in machine learning, propose a new graph-based IR to support fully-general AD, a proof-of-concept compiler toolchain Myia
包括:
- operator overloading
- source transformation
- a new graph-based IR to support fully-general AD, supports function calls, higher-order functions and recursion
- 特点:
- 可以使用closure这一点使得PyTorch能够使用source transformation来完成AD而无需tape
- 可以使用函数语言的工具对编译器进行优化
- 允许高阶导数