Easy Monotonic Policy Iteration

Joshua Achiam

A key problem in reinforcement learning for control with general function approximators (such as deep neural networks and other nonlinear functions) is that, for many algorithms em…