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

TL;DR

This paper demonstrates that over-parameterized two-layer neural networks with channel attention can learn low-degree spherical polynomials efficiently, achieving minimax optimal risk bounds with significantly reduced sample complexity.

Contribution

It introduces a novel neural network training method with channel attention that achieves minimax optimal learning rates for low-degree spherical polynomials, with provable channel selection.

Findings

Sample complexity is $n hicksim d^{ ext{degree}}/ ext{error}$, significantly better than previous bounds.

The trained network achieves the minimax optimal nonparametric regression risk rate.

A two-stage training process includes a provable channel selection algorithm.

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 in this paper. Our main result is the significantly improved sample complexity for learning such low-degree polynomials. We show that, for any regression risk $\eps \in (0,1)$, a carefully designed two-layer NN with channel attention and finite width trained by the vanilla gradient descent (GD) requires the lowest sample complexity of $n \asymp \Theta(d^{\ell_0}/\eps)$ with high probability, in contrast with the representative sample complexity $\Theta\pth{d^{\ell_0} \max\set{\eps^{-2},\log d}}$, where $n$ is the training data size. Moreover, such sample complexity is not improvable since the trained network renders a sharp rate of the nonparametric regression risk of the order $\Theta(d^{\ell_0}/{n})$ with high probability. On the other hand, the minimax optimal rate for the regression risk with a kernel of rank $\Theta(d^{\ell_0})$ is $\Theta(d^{\ell_0}/{n})$, so that the rate of the nonparametric regression risk of the network trained by GD is minimax optimal. Training the two-layer NN with channel attention proceeds in two stages: (1) a provable learnable channel selection algorithm, as a learnable harmonic-degree selection process, identifies the ground truth channel number in the target function, $\ell_0$, from $L \ge \ell_0$ channels in the first-layer activation; (2) the second layer is trained by standard GD using the selected channels. To the best of our knowledge, this is the first time a minimax optimal risk bound is obtained by training an over-parameterized but finite-width neural network with feature learning capability to learn low-degree spherical polynomials.