Tag
This paper proposes a robust gradient-based algorithm for learning behavior policies that reduce variance in online reinforcement learning policy evaluation, addressing uncertainties in transition functions with theoretical guarantees and numerical validation.