We identify a smoothness-generalization dilemma in message passing that limits GNN universality across varying homophily and propose the Inceptive Graph Neural Network (IGNN), a universal framework to address the dilemma.

This paper explore the essence of message passing on heterophilic graphs from the perspective of model and design a method with the guidance of theoretical findings.

We present a principled recipe for building graph foundation models that generalize across arbitrary graphs, features, and label spaces.

A Multi-hop Message Passing method for Homophily-Agnostic Node Representation in GNNs.