Reward Propagation
        Using Graph Convolutional Networks

        Martin Klissarov                  Doina Precup
      Mila, McGill University          Mila, McGill University and DeepMind
   martin.klissarov@mail.mcgill.ca            dprecup@cs.mcgill.ca

                       Abstract
     Potential-based reward shaping provides an approach for designing good reward
     functions, with the purpose of speeding up learning. However, automatically find-
     ing potential functions for com ...