Policy optimization

Policy optimization is the family of reinforcement learning methods that directly adjust a model's action-selection function (its "policy") to maximize expected reward, rather than learning a value estimate first and deriving actions from it. If you know supervised…