Verifying Equivalence Properties of Neural Networks with ReLU Activation Functions

Download PDF
Open Webpage

Marko Kleine Büning, Philipp Kern, Carsten Sinz

Published: 2020, Last Modified: 15 May 2026CP 2020EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Neural networks have become popular methods for tackling various machine learning tasks and are increasingly applied in safety-critical systems. This necessitates verified statements about their behavior and properties. One of these properties is the equivalence of two neural networks, which is important, e.g., when neural networks shall be reduced to smaller ones that fit space and memory constraints of embedded or mobile systems.