Review for NeurIPS paper: Instance-based Generalization in Reinforcement Learning