Differentiable programming for gradient-based machine learning

Here's a list of features mentioned in the manifesto but out of scope for this proposal:

The manifesto also mentioned a number of general features that aren't directly related to differentiable programming. These are out of scope for this proposal as well:

  • Anonymous functions, aka. func _ in the manifesto. This would be a general-purpose feature which I think can be pitched separately to support use cases beyond differentiation (e.g. dynamic method replacement).
  • Compiler-synthesized conformances for AdditiveArithmetic.
  • @memberwise attribute for triggering derived conformances. Explicit derivation is a general feature that I think deserves its own proposal so that other derivable protocols can adopt it.
4 Likes