Worst-Case Regret Bounds for Exploration via
       Randomized Value Functions

                     Daniel Russo
                    Columbia University
                  djr2174@gsb.columbia.edu

                       Abstract
     This paper studies a recent proposal to use randomized value functions to drive
     exploration in reinforcement learning. These randomized value functions are
     generated by injecting random noise into the training data, making the approach
     compatible wit ...