Reverse Derivative Ascent - A Categorical Approach to Learning Boolean Circuits

paper

Reverse Derivative Ascent: A Categorical Approach to Learning Boolean Circuits — Paul Wilson & Fabio Zanasi (2021). arXiv:2101.10488 (v1, PDF); EPTCS 333, 2021, pp. 247-260.

Defines reverse derivative ascent, a categorical analogue of gradient-based learning at the level of reverse differential categories, and applies it to learn the parameters of Boolean circuits directly, with experiments on benchmark datasets.

Sources: the paper, arXiv:2101.10488v1, checked against the arXiv listing. Index: Papers.

Key definitions and results

• Reverse derivative ascent over ${\mathbf{Poly}}_{{\mathbb{Z}}_2}$

Concept notes

Reverse Derivative Category, Gradient-Based Learning with Parametric Lenses