Controlling Graph Dynamics with
           Reinforcement Learning and Graph Neural Networks

               Eli A. Meirom 1 Haggai Maron 1 Shie Mannor 1 Gal Chechik 1

             Abstract
    We consider the problem of controlling a partially-
    observed dynamic process on a graph by a limited
    number of interventions. This problem naturally
    arises in contexts such as scheduling virus tests to
    curb an epidemic; targeted marketing in order to
    promote a product; and manually ins ...