Accelerated Rates between Stochastic and Adversarial Online Convex Optimization
Abstract
Stochastic and adversarial data are two widely studied settings in online learning. But many optimization tasks are neither i.i.d. nor fully adversarial, which makes it of fundamental interest to get a better theoretical understanding of the world between these extremes. In this work we establish novel regret bounds for online convex optimization in a setting that interpolates between stochastic i.i.d. and fully adversarial losses. By exploiting smoothness of the expected losses, these bounds replace a dependence on the maximum gradient length by the variance of the gradients, which was previously known only for linear losses. In addition, they weaken the i.i.d. assumption by allowing, for example, adversarially poisoned rounds, which were previously considered in the related expert and bandit settings. In the fully i.i.d. case, our regret bounds match the rates one would expect from results in stochastic acceleration, and we also recover the optimal stochastically accelerated rates via online-to-batch conversion. In the fully adversarial case our bounds gracefully deteriorate to match the minimax regret. We further provide lower bounds showing that our regret upper bounds are tight for all intermediate regimes in terms of the stochastic variance and the adversarial variation of the loss gradients.
Cite
@article{arxiv.2303.03272,
title = {Accelerated Rates between Stochastic and Adversarial Online Convex Optimization},
author = {Sarah Sachs and Hedi Hadiji and Tim van Erven and Cristobal Guzman},
journal= {arXiv preprint arXiv:2303.03272},
year = {2025}
}
Comments
There is an unfixable mistake in the proof of Lemma 15, as kindly brought to our attention by Peng Zhao. The mistake is not present in the earlier NeurIPS 2022 conference version of the paper, which does not include this result