23
8

Near-Optimal Goal-Oriented Reinforcement Learning in Non-Stationary Environments

Abstract

We initiate the study of dynamic regret minimization for goal-oriented reinforcement learning modeled by a non-stationary stochastic shortest path problem with changing cost and transition functions. We start by establishing a lower bound Ω((BSAT(Δc+B2ΔP))1/3K2/3)\Omega((B_{\star} SAT_{\star}(\Delta_c + B_{\star}^2\Delta_P))^{1/3}K^{2/3}), where BB_{\star} is the maximum expected cost of the optimal policy of any episode starting from any state, TT_{\star} is the maximum hitting time of the optimal policy of any episode starting from the initial state, SASA is the number of state-action pairs, Δc\Delta_c and ΔP\Delta_P are the amount of changes of the cost and transition functions respectively, and KK is the number of episodes. The different roles of Δc\Delta_c and ΔP\Delta_P in this lower bound inspire us to design algorithms that estimate costs and transitions separately. Specifically, assuming the knowledge of Δc\Delta_c and ΔP\Delta_P, we develop a simple but sub-optimal algorithm and another more involved minimax optimal algorithm (up to logarithmic terms). These algorithms combine the ideas of finite-horizon approximation [Chen et al., 2022a], special Bernstein-style bonuses of the MVP algorithm [Zhang et al., 2020], adaptive confidence widening [Wei and Luo, 2021], as well as some new techniques such as properly penalizing long-horizon policies. Finally, when Δc\Delta_c and ΔP\Delta_P are unknown, we develop a variant of the MASTER algorithm [Wei and Luo, 2021] and integrate the aforementioned ideas into it to achieve O~(min{BSALK,(B2S2AT(Δc+BΔP))1/3K2/3})\widetilde{O}(\min\{B_{\star} S\sqrt{ALK}, (B_{\star}^2S^2AT_{\star}(\Delta_c+B_{\star}\Delta_P))^{1/3}K^{2/3}\}) regret, where LL is the unknown number of changes of the environment.

View on arXiv
Comments on this paper