Optimization of Graph Neural Networks:
         Implicit Acceleration by Skip Connections and More Depth

             Keyulu Xu * 1 Mozhi Zhang 2 Stefanie Jegelka 1 Kenji Kawaguchi * 3

             Abstract                theoretical aspects of GNNs to understand their success and
   Graph Neural Networks (GNNs) have been stud-        limitations. Existing works have studied GNNs’ expressive
   ied through the lens of expressive power and gen-      power (Keriven & Peyre, 2019; Maron et al.,  ...