GNN-VPA: A Variance-Preserving Aggregation Strategy for Graph Neural Networks

TL;DR

This paper introduces VPA, a variance-preserving aggregation function for GNNs, enhancing their expressivity and learning dynamics, leading to improved predictive performance and potential for normalizer-free GNNs.

Contribution

The paper proposes a novel variance-preserving aggregation strategy for GNNs that improves expressivity and training dynamics based on signal propagation theory.

Findings

VPA improves GNN predictive accuracy.

VPA enhances training stability and dynamics.

Potential for normalizer-free GNNs.

Abstract

Graph neural networks (GNNs), and especially message-passing neural networks, excel in various domains such as physics, drug discovery, and molecular modeling. The expressivity of GNNs with respect to their ability to discriminate non-isomorphic graphs critically depends on the functions employed for message aggregation and graph-level readout. By applying signal propagation theory, we propose a variance-preserving aggregation function (VPA) that maintains expressivity, but yields improved forward and backward dynamics. Experiments demonstrate that VPA leads to increased predictive performance for popular GNN architectures as well as improved learning dynamics. Our results could pave the way towards normalizer-free or self-normalizing GNNs.