E(n) Equivariant Graph Neural Networks

               Victor Garcia Satorras 1 Emiel Hoogeboom 1 Max Welling 1

             Abstract
   This paper introduces a new model to learn graph
   neural networks equivariant to rotations, transla-
   tions, reflections and permutations called E(n)-
   Equivariant Graph Neural Networks (EGNNs). In
   contrast with existing methods, our work does not
   require computationally expensive higher-order
   representations in intermediate layers while it
   s ...