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

arXiv:2403.04747v1 Announce Type: new
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 …

abstract aggregation arxiv cs.ai cs.lg discovery domains drug discovery excel functions gnn gnns graph graph neural networks graphs modeling networks neural networks physics stat.ml strategy type variance