English

Simultaneously Learning Stochastic and Adversarial Episodic MDPs with Known Transition

Machine Learning 2020-11-03 v2 Machine Learning

Abstract

This work studies the problem of learning episodic Markov Decision Processes with known transition and bandit feedback. We develop the first algorithm with a ``best-of-both-worlds'' guarantee: it achieves O(logT)\mathcal{O}(log T) regret when the losses are stochastic, and simultaneously enjoys worst-case robustness with O~(T)\tilde{\mathcal{O}}(\sqrt{T}) regret even when the losses are adversarial, where TT is the number of episodes. More generally, it achieves O~(C)\tilde{\mathcal{O}}(\sqrt{C}) regret in an intermediate setting where the losses are corrupted by a total amount of CC. Our algorithm is based on the Follow-the-Regularized-Leader method from Zimin and Neu (2013), with a novel hybrid regularizer inspired by recent works of Zimmert et al. (2019a, 2019b) for the special case of multi-armed bandits. Crucially, our regularizer admits a non-diagonal Hessian with a highly complicated inverse. Analyzing such a regularizer and deriving a particular self-bounding regret guarantee is our key technical contribution and might be of independent interest.

Keywords

Cite

@article{arxiv.2006.05606,
  title  = {Simultaneously Learning Stochastic and Adversarial Episodic MDPs with Known Transition},
  author = {Tiancheng Jin and Haipeng Luo},
  journal= {arXiv preprint arXiv:2006.05606},
  year   = {2020}
}

Comments

Update the camera-ready version for NeurIPS 2020

R2 v1 2026-06-23T16:11:48.146Z