Generalizable Episodic Memory for Deep Reinforcement Learning

          Hao Hu 1 Jianing Ye 2 Guangxiang Zhu 1 Zhizhou Ren 3 Chongjie Zhang 1

            Abstract
                                              Discrete Episodic Memory
   Episodic memory-based methods can rapidly
   latch onto past successful strategies by a non-         Key   Value
   parametric memory and improve sample effi-
   ciency of traditional reinforcement learning. How-
   ever, little effort is put into the continuou ...