PsiPhi-Learning: Reinforcement Learning with Demonstrations using
      Successor Features and Inverse Temporal Difference Learning

    Angelos Filos 1 Clare Lyle 1 Yarin Gal 1 Sergey Levine 2 Natasha Jaques * 2 3 Gregory Farquhar * 4

              Abstract
   We study reinforcement learning (RL) with no-
   reward demonstrations, a setting in which an RL
   agent has access to additional data from the inter-
   action of other agents with the same environment.
   However, it has no access to  ...