PulseAugur
LIVE 12:25:40
research · [1 source] ·
0
research

Shallow neural networks learn low-degree spherical polynomials with feature learning

Researchers have developed a novel approach using shallow neural networks with learnable channel attention to efficiently learn low-degree spherical polynomials. This method significantly improves sample complexity compared to existing methods, requiring only $n \text{ \textasymp } \Theta(d^{\ell_0}/\eps)$ samples. The process involves two stages: first, a channel selection algorithm identifies the degree of the target function, and second, standard gradient descent trains the network using the selected channels. This work marks a significant advancement in achieving minimax optimal risk bounds for finite-width neural networks capable of feature learning. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Introduces a novel method for learning spherical polynomials with improved sample complexity, potentially impacting fields requiring spherical data analysis.

RANK_REASON This is a research paper published on arXiv detailing a new theoretical result in machine learning.

Read on arXiv stat.ML →

COVERAGE [1]

  1. arXiv stat.ML TIER_1 · Yingzhen Yang ·

    Shallow Neural Networks Learn Low-Degree Spherical Polynomials with Feature Learning by Learnable Channel Attention

    arXiv:2512.20562v2 Announce Type: replace Abstract: We study the problem of learning a low-degree spherical polynomial of degree $\ell_0 = \Theta(1) \ge 1$ defined on the unit sphere in $\RR^d$ by training an over-parameterized two-layer neural network (NN) with channel attention…