Algebraic Expressivity Certificates for Shallow Polynomial Neural Networks
Abstract
We study exact representability by bias-free shallow polynomial neural networks using algebraic geometry. Over ℂ, a width-r network with activation z↦ zd computes a sum of r d-th powers of linear forms, whose Zariski closure is a Veronese secant variety. Ideal elimination therefore yields polynomial certificates of nonrepresentability. We implement this construction as a generic architecture-to-certificate pipeline. For quadratics, we recover the exact symmetric determinantal description and explain its dimension through orthogonal symmetry. In higher degree, the implementation recovers classical catalecticant and secant equations and maps the practical reach of direct elimination across a finite architecture sweep. We also derive the exact population loss floor for a rank-two quadratic network on the sphere, illustrating how an algebraic obstruction induces irreducible approximation error.
Related
- Why Shallow Neural Networks Learn Sinusoidal Components First: An Orthogonal Polynomial Dynamics Framework
- Optimizing Lyapunov Certificates via Stability-Preserving Quadratization for Polynomial Systems
- Constructive Bounds for the Topological Expressivity of Neural Networks
- Algebraic Structure and Orthogonality Properties of Rational Polynomial Phase Rings on Discrete Lattices
- Fully Deterministic Shallow Neural Networks with Localized Activations. Part II: Condition Number Estimates
BibTeX
@misc{akbari2026algebraic,
title = {{Algebraic Expressivity Certificates for Shallow Polynomial Neural Networks}},
author = {Sepehr Akbari and Shahrzad Jamshidi},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.28500},
eprint = {2609.28500},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.28500},
}
RIS
TY - UNPB
TI - Algebraic Expressivity Certificates for Shallow Polynomial Neural Networks
AU - Sepehr Akbari
AU - Shahrzad Jamshidi
PY - 2026
DA - 2026/09/11
PB - arXiv
DO - 10.48550/arxiv.2609.28500
UR - https://arxiv.org/abs/2609.28500
AB - We study exact representability by bias-free shallow polynomial neural networks using algebraic geometry. Over ℂ, a width-r network with activation z↦ z^d computes a sum of r d-th powers of linear forms, whose Zariski closure is a Veronese secant variety. Ideal elimination therefore yields polynomial certificates of nonrepresentability. We implement this construction as a generic architecture-to-certificate pipeline. For quadratics, we recover the exact symmetric determinantal description and explain its dimension through orthogonal symmetry. In higher degree, the implementation recovers classical catalecticant and secant equations and maps the practical reach of direct elimination across a finite architecture sweep. We also derive the exact population loss floor for a rank-two quadratic network on the sphere, illustrating how an algebraic obstruction induces irreducible approximation error.
ER -