Reproduce “Graph Kernels: A Survey” 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

Giannis Nikolentzos, Giannis Siglidis, Michalis Vazirgiannis

A comprehensive survey and empirical evaluation compares a wide array of graph kernels with state-of-the-art Graph Neural Networks (GNNs) on graph classification tasks across 17 datasets. The analysis reveals that Weisfeiler-Lehman-based graph kernels frequently outperform GNNs on node-labeled graphs, while GNNs show stronger performance for node-attributed graphs.

Key insights

  • Weisfeiler-Lehman based kernels (e.g., WL-Optimal Assignment, CORE-WL-VH) frequently achieve superior classification accuracy on node-labeled graphs compared to state-of-the-art GNNs.
  • Graph Neural Networks (GIN, DGCNN, GraphSAGE, DiffPool) demonstrate stronger performance for graph classification on datasets with continuous node attributes.
  • Many complex graph kernels are computationally expensive, leading to frequent timeouts or out-of-memory errors on larger datasets, limiting their practical applicability.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.