Policy Optimization

A method in reinforcement learning that directly optimizes the policy by performing local updates at each state to improve performance. Compared with occupancy measure methods, it is computationally simpler but its regret bound is typically larger by a factor of H.