Extracting Formulae in Many-Valued Logic from Deep Neural Networks
Machine Learning
2025-03-07 v2 Artificial Intelligence
Logic in Computer Science
Abstract
We propose a new perspective on deep ReLU networks, namely as circuit counterparts of Lukasiewicz infinite-valued logic -- a many-valued (MV) generalization of Boolean logic. An algorithm for extracting formulae in MV logic from deep ReLU networks is presented. As the algorithm applies to networks with general, in particular also real-valued, weights, it can be used to extract logical formulae from deep ReLU networks trained on data.
Cite
@article{arxiv.2401.12113,
title = {Extracting Formulae in Many-Valued Logic from Deep Neural Networks},
author = {Yani Zhang and Helmut Bölcskei},
journal= {arXiv preprint arXiv:2401.12113},
year = {2025}
}
Comments
Signicant extension of the previous version