Protecting against evaluation overfitting in empirical reinforcement learning | IEEE Conference Publication | IEEE Xplore