Statistical Guarantees for Lifelong Reinforcement Learning using PAC-Bayes Theory
Abstract
Lifelong reinforcement learning (RL) has been developed as a paradigm for extending single-task RL to more realistic, dynamic settings. In lifelong RL, the "life" of an RL agent is modeled as a stream of tasks drawn from a task distribution. We propose EPIC (Empirical PAC-Bayes that Improves Continuously), a novel algorithm designed for lifelong RL using PAC-Bayes theory. EPIC learns a shared policy distribution, referred to as the world policy, which enables rapid adaptation to new tasks while retaining valuable knowledge from previous experiences. Our theoretical analysis establishes a relationship between the algorithm's generalization performance and the number of prior tasks preserved in memory. We also derive the sample complexity of EPIC in terms of RL regret. Extensive experiments on a variety of environments demonstrate that EPIC significantly outperforms existing methods in lifelong RL, offering both theoretical guarantees and practical efficacy through the use of the world policy.
Cite
@article{arxiv.2411.00401,
title = {Statistical Guarantees for Lifelong Reinforcement Learning using PAC-Bayes Theory},
author = {Zhi Zhang and Chris Chow and Yasi Zhang and Yanchao Sun and Haochen Zhang and Eric Hanchen Jiang and Han Liu and Furong Huang and Yuchen Cui and Oscar Hernan Madrid Padilla},
journal= {arXiv preprint arXiv:2411.00401},
year = {2025}
}
Comments
9 pages, 4 figures, accepted at AISTATS 2025 (PMLR Vol 258), paper ID 9417