English

Toward $L_\infty$-recovery of Nonlinear Functions: A Polynomial Sample Complexity Bound for Gaussian Random Fields

Machine Learning 2023-05-02 v1

Abstract

Many machine learning applications require learning a function with a small worst-case error over the entire input domain, that is, the LL_\infty-error, whereas most existing theoretical works only guarantee recovery in average errors such as the L2L_2-error. LL_\infty-recovery from polynomial samples is even impossible for seemingly simple function classes such as constant-norm infinite-width two-layer neural nets. This paper makes some initial steps beyond the impossibility results by leveraging the randomness in the ground-truth functions. We prove a polynomial sample complexity bound for random ground-truth functions drawn from Gaussian random fields. Our key technical novelty is to prove that the degree-kk spherical harmonics components of a function from Gaussian random field cannot be spiky in that their LL_\infty/L2L_2 ratios are upperbounded by O(dlnk)O(d \sqrt{\ln k}) with high probability. In contrast, the worst-case LL_\infty/L2L_2 ratio for degree-kk spherical harmonics is on the order of Ω(min{dk/2,kd/2})\Omega(\min\{d^{k/2},k^{d/2}\}).

Keywords

Cite

@article{arxiv.2305.00322,
  title  = {Toward $L_\infty$-recovery of Nonlinear Functions: A Polynomial Sample Complexity Bound for Gaussian Random Fields},
  author = {Kefan Dong and Tengyu Ma},
  journal= {arXiv preprint arXiv:2305.00322},
  year   = {2023}
}

Comments

39 pages

R2 v1 2026-06-28T10:21:39.095Z