Fetching the paper…

Action-depedent Control Variates for Policy Optimization via Stein's Identity · Around