In our latest paper we propose that attention-based Meta-RL can achieve improved generalization capabilities by learning transition dynamics as well as learning dynamics by paying attention to intra- as well as inter-episode experiences.