Local and global topological complexity measures of ReLU neural network functions
Topology Proceedings · 13.10.2026 · 69 · 67-123
DOI: 10.25071/7etkr663 ↗Kurz gefasst
Fachartikel in der Zeitschrift Topology Proceedings, online erschienen am 13.10.2026. Verfasst von J. Elisenda Grigsby und 2 weiteren Autor:innen.
Abstract (Originalsprache)
We apply a generalized piecewise-linear (PL) version of Morse theory due to Grunert-Kühnel-Rote to define and study new local and global notions of topological complexity for fully-connected feedforward ReLU neural network functions $F: \mathbb{R}^n \rightarrow \mathbb{R}$. Along the way, we show how to construct, for each such $F$, a canonical polytopal complex $\mathcal{K}(F)$ and a deformation retract $\mathbb{R}^n \rightarrow \mathcal{K}(F)$, yielding a convenient compact model for performing calculations. We also give a construction showing that local complexity can be arbitrarily high.
Zitieren
APA
Grigsby, J. E., Lindsey, K., & Madsen, M. (2026). Local and global topological complexity measures of ReLU neural network functions. Topology Proceedings, 69, 67-123. https://doi.org/10.25071/7etkr663
Vancouver
Grigsby JE, Lindsey K, Madsen M. Local and global topological complexity measures of ReLU neural network functions. Topology Proceedings. 2026;69:67-123. doi:10.25071/7etkr663
BibTeX
@article{Grigsby2026,
author = {Grigsby, J. Elisenda and Lindsey, Kathryn and Madsen, Marissa},
title = {Local and global topological complexity measures of ReLU neural network functions},
journal = {Topology Proceedings},
year = {2026},
volume = {69},
pages = {67--123},
doi = {10.25071/7etkr663}
}
EUVIMED