High Confidence Generalization for Reinforcement Learning

    James E. Kostas 1 Yash Chandak 1 Scott M. Jordan 1 Georgios Theocharous 2 Philip S. Thomas 1

             Abstract                 formance on MDPs drawn from the distribution, including
                                  MDPs not in the training set.
   We present several classes of reinforcement learn-
   ing algorithms that safely generalize to Markov        HCGAs first train using a standard RL algorithm and then
   decision proc ...