Borrowing From the Future: An Attempt to Address Double Sampling

Yuhua Zhu, Lexing Ying

For model-free reinforcement learning, one of the main difficulty of stochastic Bellman residual minimization is the double sampling problem, i.e., while only one single sample for…