Efficient Performance Bounds for Primal-Dual Reinforcement
                Learning from Demonstrations

                  Angeliki Kamoutsi 1 Goran Banjac 1 John Lygeros 1

             Abstract                In the standard RL setting a cost signal is given to instruct
    We consider large-scale Markov decision pro-        agents how to complete the desired task. However, often-
    cesses with an unknown cost function and ad-        times encoding preferences using demonstrations provided
  ...