Reproduce “Graph Neural Networks Do Not Always Oversmooth” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Bastian Epping, Alexandre René, Moritz Helias, Michael T. Schaub
The authors analyze how initial weight variance affects oversmoothing in graph convolutional networks (GCNs). In an infinite-width Gaussian-process (GP) model, they track covariance changes and identify a transition where the identical-node-feature state becomes unstable. On a two-community synthetic graph, GP classification error stays below chance beyond 1,000 layers, although input features are lost and topology carries the prediction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.