English

Hardness of Learning Fixed Parities with Neural Networks

Machine Learning 2025-01-09 v2 Machine Learning

Abstract

Learning parity functions is a canonical problem in learning theory, which although computationally tractable, is not amenable to standard learning algorithms such as gradient-based methods. This hardness is usually explained via statistical query lower bounds [Kearns, 1998]. However, these bounds only imply that for any given algorithm, there is some worst-case parity function that will be hard to learn. Thus, they do not explain why fixed parities - say, the full parity function over all coordinates - are difficult to learn in practice, at least with standard predictors and gradient-based methods [Abbe and Boix-Adsera, 2022]. In this paper, we address this open problem, by showing that for any fixed parity of some minimal size, using it as a target function to train one-hidden-layer ReLU networks with perturbed gradient descent will fail to produce anything meaningful. To establish this, we prove a new result about the decay of the Fourier coefficients of linear threshold (or weighted majority) functions, which may be of independent interest.

Keywords

Cite

@article{arxiv.2501.00817,
  title  = {Hardness of Learning Fixed Parities with Neural Networks},
  author = {Itamar Shoshani and Ohad Shamir},
  journal= {arXiv preprint arXiv:2501.00817},
  year   = {2025}
}

Comments

An updated version was uploaded in order to fix a typo at theorem 2 statement