Id: arxiv/2611.10700 Url: https://arxiv.org/abs/2611.10700 Tags: cs.LG, math.NA Score: 0 Title: Spectral Bias Reduction in Local Gated ReLU Architectures Domain: cs.LG Source: arxiv Authors: Victorita Dolean Reviews: 0 Subject: cs.LG Category: research Comments: 0 Disputed: false Division: cs Mentions: 0 Discussed: 0 Published: 2026-11-07 Page: https://delphi.science/r/arxiv/2611.10700 Neural networks trained by gradient-based methods learn the low-frequency content of a target function faster than its high-frequency content. This spectral bias is well documented, but it is much less clear which architectural choices mitigate it and which cannot. We study the question for shallow ReLU models linear in their trainable parameters, where each mode is learned at a rate given by a Gram-matrix eigenvalue, and for a general symmetric bilinear form $a(\cdot,\cdot)$, so that least-squares regression ($a=$ the $L^2$ inner product) and variational or physics-informed losses ($a=$ the…