Learning the Reward Function for a Misspecified Model

E. Talvitie

Abstract: In model-based reinforcement learning it is typical to decouple the problems of learning the dynamics model and learning the reward function. However, when the dynamics model is flawed, it may gene...