The Symmetries of Three-Layer ReLU Networks
Machine Learning
2026-05-19 v1 Discrete Mathematics
Algebraic Geometry
Combinatorics
Abstract
We develop a framework for analyzing parameter symmetries in deep ReLU networks and obtain a complete characterization of the generic parameter fibers for three-layer bottleneck architectures. Our approach provides explicit semi-algebraic descriptions of these fibers and yields a polynomial time algorithm for deciding functional equivalence of two parameters. The symmetries include discrete and continuous transformations arising from layer composition, and depend on whether deeper layers hide or preserve geometric structure from preceding layers. Finally, we show that some of these symmetries induce local conservation laws along gradient flow, while others do not.
Cite
@article{arxiv.2605.18319,
title = {The Symmetries of Three-Layer ReLU Networks},
author = {Johanna Marie Gegenfurtner and Moritz Grillo and Guido Montúfar},
journal= {arXiv preprint arXiv:2605.18319},
year = {2026}
}