On the Expressive Power of Graph Neural Networks

The study explores the expressive power of Graph Neural Networks (GNNs) and their application in fields like social science, chemistry, and medicine. The authors examine the theoretical properties of GNNs, their ability to distinguish different graph structures, and compute graph properties. The research also covers the Universal Approximation Theorem, Graph Isomorphism Test, and the design choices in GNN architecture.

Publication date: 3 Jan 2024
Project Page: arXiv:2401.01626v1
Paper: https://arxiv.org/pdf/2401.01626