Maximum Entropy Model Correction in Reinforcement Learning