Future Information Minimization as PAC Bayes regularization in Reinforcement Learning

Interactions between an organism and its environment are commonly treated in the framework of Markov Decision Processes (MDP). While standard MDP is aimed at maximizing expected future rewards (value)