Reverse Derivative Ascent

A library for training circuit models as in the paper Reverse Derivative Ascent: A Categorical Approach to Learning Boolean Circuits

This library provides three main things:

• A type-safe, Integer-backed implementation of Bitvectors
• Some tools for compositionally building circuit models as in
  our [reverse derivative ascent] paper (including computing their reverse
  derivative)
• A brute-force implementation of the reverse derivative, which computes the
  same results as the compositional one, but slower.

User Guide

For now, see the project act-2020-experiments, which gives a couple example applications: we build a model on the Iris dataset, and on a subset of MNIST.