InformationDirectedRewardLearning forReinforcementLearning

From such expensive feedback, we aim to learn a model of the reward that allows standard RL algorithms to achieve high expected returnswith as few expert queries as possible.