Reinforcement Learning from Partial Observation: Linear Function Approximation with Provable Sample Efficiency

arXiv:2204.09787v3 Announce Type: replace
Abstract: We study reinforcement learning for partially observed Markov decision processes (POMDPs) with infinite observation and state spaces, which remains less investigated theoretically. To this end, we make the first attempt at bridging partial observability and function approximation for a class of POMDPs with a linear structure. In detail, we propose a reinforcement learning algorithm (Optimistic Exploration via Adversarial Integral Equation or OP-TENET) that attains an $\epsilon$-optimal policy within $O(1/\epsilon^2)$ episodes. In particular, the sample complexity …

abstract approximation arxiv class cs.lg decision efficiency function linear markov math.oc observability observation processes reinforcement reinforcement learning sample spaces state stat.ml study type