GNN-VPA: A Variance-Preserving Aggregation Strategy for Graph Neural Networks
This addresses a bottleneck in GNN design for domains like physics and drug discovery, offering an incremental improvement in aggregation strategies.
The paper tackled the problem of maintaining expressivity in graph neural networks (GNNs) by proposing a variance-preserving aggregation function (VPA), which improved predictive performance and learning dynamics for popular architectures.
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.