Policy Improvement Directions for Reinforcement Learning in Reproducing Kernel Hilbert Spaces | IEEE Conference Publication | IEEE Xplore