English

How Much Over-parameterization Is Sufficient to Learn Deep ReLU Networks?

Machine Learning 2022-01-03 v4 Optimization and Control Machine Learning

Abstract

A recent line of research on deep learning focuses on the extremely over-parameterized setting, and shows that when the network width is larger than a high degree polynomial of the training sample size nn and the inverse of the target error ϵ1\epsilon^{-1}, deep neural networks learned by (stochastic) gradient descent enjoy nice optimization and generalization guarantees. Very recently, it is shown that under certain margin assumptions on the training data, a polylogarithmic width condition suffices for two-layer ReLU networks to converge and generalize (Ji and Telgarsky, 2019). However, whether deep neural networks can be learned with such a mild over-parameterization is still an open question. In this work, we answer this question affirmatively and establish sharper learning guarantees for deep ReLU networks trained by (stochastic) gradient descent. In specific, under certain assumptions made in previous work, our optimization and generalization guarantees hold with network width polylogarithmic in nn and ϵ1\epsilon^{-1}. Our results push the study of over-parameterized deep neural networks towards more practical settings.

Keywords

Cite

@article{arxiv.1911.12360,
  title  = {How Much Over-parameterization Is Sufficient to Learn Deep ReLU Networks?},
  author = {Zixiang Chen and Yuan Cao and Difan Zou and Quanquan Gu},
  journal= {arXiv preprint arXiv:1911.12360},
  year   = {2022}
}

Comments

21 pages, 1 figure, 1 table. In ICLR 2021

R2 v1 2026-06-23T12:29:24.137Z