Research map / Graph learning / GNN architectures
How Powerful are K-hop Message Passing Graph Neural Networks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| add peripheral subgraph information (KP-GNN) to K-hop message passing (ablation) | K-GIN · SR25 K=1 | accuracy: 6.67 -> 100 | Helped |
| add peripheral subgraph information (ablation) | K-GIN · CSL K=2 (SPD) | accuracy: 32 -> 52.7 | Helped |
| add peripheral subgraph information (ablation) | K-GIN · EXP K=2 (SPD) | accuracy: 50 -> 100 | Helped |
| increase hops K from 2 to 4 (ablation) | K-GIN · CSL (SPD) | accuracy: 32 -> 92.7 | Helped |
| use shortest path distance kernel instead of graph diffusion (ablation) | K-GIN K=3 · EXP | accuracy: 66.9 -> 100 | Helped |
| use shortest path distance kernel instead of graph diffusion (ablation) | K-GIN K=4 · CSL | accuracy: 62.7 -> 92.7 | Helped |
| use shortest path distance vs graph diffusion kernel in KP-GIN (ablation) | KP-GIN · CSL K=2 | accuracy: 52.7 vs 52.7 | No clear effect |
| add peripheral subgraph info to K-GIN+ for graph connectedness (ablation) | K-GIN+ · connectedness log10(MSE) | log10(MSE): -2.1782 -> -4.4322 | Helped |
| add peripheral subgraph info to K-GIN+ for diameter (ablation) | K-GIN+ · diameter log10(MSE) | log10(MSE): -3.9695 -> -3.9361 | Hurt |
| add peripheral subgraph info to K-GIN+ for 4-cycle counting (ablation) | K-GIN+ · 4-cycle MAE | MAE: 0.2079 -> 0.0395 | Helped |
| KP-GIN' variant instead of KP-GIN+ (ablation) | KP-GIN · QM9 and ZINC | test MAE: 0.111 -> 0.093 (ZINC MAE) | Helped |
Closest papers by meaning
- A Practical, Progressively-Expressive GNN (NeurIPS 2022 · similarity 0.91)
- Extending the Design Space of Graph Neural Networks by Rethinking Folklore Weisfeiler-Lehman (NeurIPS 2023 · similarity 0.91)
- Sign is Not a Remedy: Multiset-to-Multiset Message Passing for Learning on Heterophilic Graphs (ICML 2024 · similarity 0.89)
- Towards Dynamic Message Passing on Graphs (NeurIPS 2024 · similarity 0.89)
- What Expressivity Theory Misses: Message Passing Complexity for GNNs (NeurIPS 2025 · similarity 0.88)
- Pruning for GNNs: Lower Complexity with Comparable Expressiveness (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.