ALIZATION FOR HIGH-DIMENSIONAL MODEL-BASED REINFORCEMENT LEARNING