$\partial\mathbb{B}$ nets: learning discrete functions by gradient descent
Abstract
nets are differentiable neural networks that learn discrete boolean-valued functions by gradient descent. nets have two semantically equivalent aspects: a differentiable soft-net, with real weights, and a non-differentiable hard-net, with boolean weights. We train the soft-net by backpropagation and then `harden' the learned weights to yield boolean weights that bind with the hard-net. The result is a learned discrete function. `Hardening' involves no loss of accuracy, unlike existing approaches to neural network binarization. Preliminary experiments demonstrate that nets achieve comparable performance on standard machine learning problems yet are compact (due to 1-bit weights) and interpretable (due to the logical nature of the learnt functions).
Cite
@article{arxiv.2305.07315,
title = {$\partial\mathbb{B}$ nets: learning discrete functions by gradient descent},
author = {Ian Wright},
journal= {arXiv preprint arXiv:2305.07315},
year = {2023}
}
Comments
17 pages, 8 figures