Tag
This paper introduces Good Policy Identification (GPI) in reinforcement learning, aiming to find a policy meeting a reward threshold rather than the optimal one, and proposes the BEE-GPI algorithm with near-optimal sample complexity guarantees.