Expected Return

reinforcement-learning

Definition

Expected Return

Let be a parameterised policy and let be the start-state distribution. The expected return is the objective maximised in reinforcement learning:

where is the return from the initial state and is the discount factor.