Graph Isomorphism Network

A GNN architecture designed to strengthen the theoretical power of distinguishing graph structures.

The Graph Isomorphism Network is an important architecture developed to achieve highly discriminative representation learning on graphs. It is especially notable because of its theoretical connection to the Weisfeiler-Lehman test. The goal is to build neighborhood aggregation mechanisms that can distinguish graph structures more precisely. It is a strong example of how theory and practical GNN design can meet.