Algebraic geometry of rational neural networks

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Grosdos, Alexandros, Robeva, Elina, Zubkov, Maksym
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908538284015616
author Grosdos, Alexandros
Robeva, Elina
Zubkov, Maksym
author_facet Grosdos, Alexandros
Robeva, Elina
Zubkov, Maksym
contents We study the expressivity of rational neural networks (RationalNets) through the lens of algebraic geometry. We consider rational functions that arise from a given RationalNet to be tuples of fractions of homogeneous polynomials of fixed degrees. For a given architecture, the neuromanifold is the set of all such expressible tuples. For RationalNets with one hidden layer and fixed activation function $1/x$, we characterize the dimension of the neuromanifold and provide defining equations for some architectures. We also propose algorithms that determine whether a given rational function belongs to the neuromanifold. For deep binary RationalNets, i.e., RationalNets all of whose layers except potentially for the last one are binary, we classify when the Zarisky closure of the neuromanifold equals the whole ambient space, and give bounds on its dimensions.
format Preprint
id arxiv_https___arxiv_org_abs_2509_11088
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Algebraic geometry of rational neural networks
Grosdos, Alexandros
Robeva, Elina
Zubkov, Maksym
Algebraic Geometry
68T07, 14M12, 41A20, 62R01
We study the expressivity of rational neural networks (RationalNets) through the lens of algebraic geometry. We consider rational functions that arise from a given RationalNet to be tuples of fractions of homogeneous polynomials of fixed degrees. For a given architecture, the neuromanifold is the set of all such expressible tuples. For RationalNets with one hidden layer and fixed activation function $1/x$, we characterize the dimension of the neuromanifold and provide defining equations for some architectures. We also propose algorithms that determine whether a given rational function belongs to the neuromanifold. For deep binary RationalNets, i.e., RationalNets all of whose layers except potentially for the last one are binary, we classify when the Zarisky closure of the neuromanifold equals the whole ambient space, and give bounds on its dimensions.
title Algebraic geometry of rational neural networks
topic Algebraic Geometry
68T07, 14M12, 41A20, 62R01
url https://arxiv.org/abs/2509.11088