Logistic Q-Learning

Joan Bas-Serrano, Sebastian Curi, Andreas Krause +1

We propose a new reinforcement learning algorithm derived from a regularized linear-programming formulation of optimal control in MDPs. The method is closely related to the classic…