Saved in:
Bibliographic Details
Main Authors: Huang, Tianlong, Li, Zhiyuan
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2511.23443
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911292661432320
author Huang, Tianlong
Li, Zhiyuan
author_facet Huang, Tianlong
Li, Zhiyuan
contents This paper studies the role of activation functions in learning modular addition with two-layer neural networks. We first establish a sharp expressivity gap: sine MLPs admit width-$2$ exact realizations for any fixed length $m$ and, with bias, width-$2$ exact realizations uniformly over all lengths. In contrast, the width of ReLU networks must scale linearly with $m$ to interpolate, and they cannot simultaneously fit two lengths with different residues modulo $p$. We then provide a novel Natarajan-dimension generalization bound for sine networks, yielding nearly optimal sample complexity $\widetilde{\mathcal{O}}(p)$ for ERM over constant-width sine networks. We also derive width-independent, margin-based generalization for sine networks in the overparametrized regime and validate it. Empirically, sine networks generalize consistently better than ReLU networks across regimes and exhibit strong length extrapolation.
format Preprint
id arxiv_https___arxiv_org_abs_2511_23443
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Provable Benefits of Sinusoidal Activation for Modular Addition
Huang, Tianlong
Li, Zhiyuan
Machine Learning
This paper studies the role of activation functions in learning modular addition with two-layer neural networks. We first establish a sharp expressivity gap: sine MLPs admit width-$2$ exact realizations for any fixed length $m$ and, with bias, width-$2$ exact realizations uniformly over all lengths. In contrast, the width of ReLU networks must scale linearly with $m$ to interpolate, and they cannot simultaneously fit two lengths with different residues modulo $p$. We then provide a novel Natarajan-dimension generalization bound for sine networks, yielding nearly optimal sample complexity $\widetilde{\mathcal{O}}(p)$ for ERM over constant-width sine networks. We also derive width-independent, margin-based generalization for sine networks in the overparametrized regime and validate it. Empirically, sine networks generalize consistently better than ReLU networks across regimes and exhibit strong length extrapolation.
title Provable Benefits of Sinusoidal Activation for Modular Addition
topic Machine Learning
url https://arxiv.org/abs/2511.23443