English

Understanding convolution on graphs via energies

Machine Learning 2023-09-07 v5 Machine Learning

Abstract

Graph Neural Networks (GNNs) typically operate by message-passing, where the state of a node is updated based on the information received from its neighbours. Most message-passing models act as graph convolutions, where features are mixed by a shared, linear transformation before being propagated over the edges. On node-classification tasks, graph convolutions have been shown to suffer from two limitations: poor performance on heterophilic graphs, and over-smoothing. It is common belief that both phenomena occur because such models behave as low-pass filters, meaning that the Dirichlet energy of the features decreases along the layers incurring a smoothing effect that ultimately makes features no longer distinguishable. In this work, we rigorously prove that simple graph-convolutional models can actually enhance high frequencies and even lead to an asymptotic behaviour we refer to as over-sharpening, opposite to over-smoothing. We do so by showing that linear graph convolutions with symmetric weights minimize a multi-particle energy that generalizes the Dirichlet energy; in this setting, the weight matrices induce edge-wise attraction (repulsion) through their positive (negative) eigenvalues, thereby controlling whether the features are being smoothed or sharpened. We also extend the analysis to non-linear GNNs, and demonstrate that some existing time-continuous GNNs are instead always dominated by the low frequencies. Finally, we validate our theoretical findings through ablations and real-world experiments.

Keywords

Cite

@article{arxiv.2206.10991,
  title  = {Understanding convolution on graphs via energies},
  author = {Francesco Di Giovanni and James Rowbottom and Benjamin P. Chamberlain and Thomas Markovich and Michael M. Bronstein},
  journal= {arXiv preprint arXiv:2206.10991},
  year   = {2023}
}

Comments

Accepted at TMLR; First two authors equal contribution; 35 pages

R2 v1 2026-06-24T11:59:55.183Z