Research map / Graph learning / GNN architectures
Graph Neural Networks Are Not Continuous Across Graph Resolutions
Christian Koke, Yuesong Shen, Abhishek Saroha, Marvin Eisenberger, Bastian Rieck, Michael M. Bronstein, Daniel Cremers
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace standard propagation with Laplace-transform (exponential) propagation, MPNN | GCN-style MPNN · QM7 train High infer Low | MAE (kcal/mol): 136.7 -> 17.4 | Improved |
| replace standard propagation with Laplace-transform (resolvent) propagation, MPNN | GCN-style MPNN · QM7 train Low infer High | MAE (kcal/mol): 138.1 -> 17.6 | Improved |
| use spectral Laplace-transform (exponential) propagation instead of ChebNet | ChebNet · QM7 train High infer Low | MAE (kcal/mol): 447.8 -> 15.9 | Improved |
| use Laplace-transform propagation instead of Lanczos multi-scale propagation | Lanczos · QM7 train Low infer High | MAE (kcal/mol): 658.6 -> 16.0 | Improved |
| use Laplace-transform MPNN (exp) instead of GCN for latent embedding | GCN · QM7 embedding difference High to Low | embedding difference: 20.3 -> 0.2 | Improved |
| use Laplace-transform MPNN (exp) instead of GIN for latent embedding | GIN · QM7 embedding difference Low to High | embedding difference: 2875.8 -> 0.2 | Improved |
| test GNNs cross-resolution instead of same-resolution (train High, infer Low) | GCN · QM7 | MAE (kcal/mol): 63.6 -> 136.7 | Hurt |
| test GNNs cross-resolution instead of same-resolution (train High, infer Low) | Lanczos · QM7 | MAE (kcal/mol): 9.9 -> 938.4 | Hurt |
| increase clique size k when expanding nodes | standard GNNs (GCN) · citation networks node classification | accuracy | Hurt |
| use Laplace-transform propagation under increasing clique size k | Laplace-transform GNNs · citation networks node classification | accuracy | Improved |
Closest papers by meaning
- On Vanishing Gradients, Over-Smoothing, and Over-Squashing in GNNs: Bridging Recurrent and Graph Learning (NeurIPS 2025 · similarity 0.90)
- Almost Surely Asymptotically Constant Graph Neural Networks (NeurIPS 2024 · similarity 0.89)
- Exploring Consistency in Graph Representations: from Graph Kernels to Graph Neural Networks (NeurIPS 2024 · similarity 0.88)
- Improving Graph Neural Networks by Learning Continuous Edge Directions (ICLR 2025 · similarity 0.88)
- Graph Neural Networks Are More Than Filters: Revisiting and Benchmarking from A Spectral Perspective (ICLR 2025 · similarity 0.88)
- A Manifold Perspective on the Statistical Generalization of Graph Neural Networks (ICML 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
