Towards Interpretable Reinforcement Learning Using
      Attention Augmented Agents

              Alex Mott*  Daniel Zoran*  Mike Chrzanowski
             Daan Wierstra Danilo J. Rezende
                   DeepMind
                  London, UK
     {alexmott,danielzoran,chrzanowski,wierstra,danilor}@google.com

                       Abstract
     Inspired by recent work in attention models for image captioning and question
     answering, we present a soft attention model for the reinforcement ...