Fall 2020

Settling the Horizon-Dependence of Sample Complexity in Reinforcement Learning

Monday, Nov. 15, 2021 10:00 am10:30 am PST

Add to Calendar


Lin Yang (UCLA)


Calvin Lab Auditorium

Recently there is a surge of interest in understanding the horizon-dependence of the sample complexity in reinforcement learning (RL). Notably, for an RL environment with horizon length H, previous work has shown that there is a probably approximately correct (PAC) algorithm that learns an O(1)-optimal policy using polylog(H) episodes of environment interactions when the number of states and actions is fixed. It is yet unknown whether the polylog(H) dependence is necessary or not. In this work, we resolve this question by developing an algorithm that achieves the same PAC guarantee while using only O(1) episodes of environment interactions, completely settling the horizon-dependence of the sample complexity in RL. We achieve this bound by (i) establishing a connection between value functions in discounted and finite-horizon Markov decision processes (MDPs) and (ii) a novel perturbation analysis in MDPs. We believe our new techniques are of independent interest and could be applied in related questions in RL.