English

Fast Convergence of Random Reshuffling under Over-Parameterization and the Polyak-\L ojasiewicz Condition

Machine Learning 2023-04-04 v1 Optimization and Control

Abstract

Modern machine learning models are often over-parameterized and as a result they can interpolate the training data. Under such a scenario, we study the convergence properties of a sampling-without-replacement variant of stochastic gradient descent (SGD) known as random reshuffling (RR). Unlike SGD that samples data with replacement at every iteration, RR chooses a random permutation of data at the beginning of each epoch and each iteration chooses the next sample from the permutation. For under-parameterized models, it has been shown RR can converge faster than SGD under certain assumptions. However, previous works do not show that RR outperforms SGD in over-parameterized settings except in some highly-restrictive scenarios. For the class of Polyak-\L ojasiewicz (PL) functions, we show that RR can outperform SGD in over-parameterized settings when either one of the following holds: (i) the number of samples (nn) is less than the product of the condition number (κ\kappa) and the parameter (α\alpha) of a weak growth condition (WGC), or (ii) nn is less than the parameter (ρ\rho) of a strong growth condition (SGC).

Keywords

Cite

@article{arxiv.2304.00459,
  title  = {Fast Convergence of Random Reshuffling under Over-Parameterization and the Polyak-\L ojasiewicz Condition},
  author = {Chen Fan and Christos Thrampoulidis and Mark Schmidt},
  journal= {arXiv preprint arXiv:2304.00459},
  year   = {2023}
}
R2 v1 2026-06-28T09:45:00.447Z