actor-critic methods

A type of reinforcement learning method that maintains separate parameterizations for the policy and the value function.