A Semi-parametric Statistical Approach to Model-free Policy Evaluation

Reinforcement learning (RL) methods based on least-squares temporal difference (LSTD) have been developed recently and have shown good practical performance. However, the quality of their estimation h