Research map / Graph learning / GNN architectures
Distance-Restricted Folklore Weisfeiler-Leman GNNs with Provable Cycle Counting Power
Junru Zhou, Jiarui Feng, Xiyuan Wang, Muhan Zhang
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 |
|---|---|---|---|
| increase restriction distance d from 2 to 3 (ablation) | d-DRFWL(2) GNN · Synthetic 7-cycle counting | normalized MAE: 0.0362 -> 0.0176 | Improved |
| increase d from 2 to 3 (ablation) | d-DRFWL(2) GNN · Synthetic 6-cycle counting | normalized MAE: 0.0087 -> 0.0099 | No clear effect |
| increase d from 2 to 3 (ablation) | d-DRFWL(2) GNN · Synthetic 4-cycle counting | normalized MAE: 0.0015 -> 0.0020 | No clear effect |
| reduce d from 2 to 1 (ablation) | d-DRFWL(2) GNN · Synthetic cycle counting | normalized MAE | Hurt |
| use 2-DRFWL(2) GNN instead of NGNN (subgraph GNN) | 2-DRFWL(2) GNN · QM9 efficiency | GPU memory (GB): 13.72 -> 2.31 | Improved |
| use 2-DRFWL(2) GNN instead of I2-GNN | 2-DRFWL(2) GNN · QM9 efficiency | preprocessing time (s): 5287 -> 430 | Improved |
| use 2-DRFWL(2) GNN instead of I2-GNN | 2-DRFWL(2) GNN · QM9 efficiency | training time (s/epoch): 209.9 -> 141.9 | Improved |
| use 2-DRFWL(2) GNN instead of NGNN | 2-DRFWL(2) GNN · QM9 efficiency | training time (s/epoch): 107.8 -> 141.9 | No clear effect |
| use 2-DRFWL(2) GNN instead of I2-GNN | 2-DRFWL(2) GNN · ogbg-molhiv efficiency | GPU memory (GB): 11.07 -> 4.44 | Improved |
| use 2-DRFWL(2) GNN instead of I2-GNN | 2-DRFWL(2) GNN · QM9 U_0 | MAE: 0.211 -> 0.156 | Improved |
| use 2-DRFWL(2) GNN instead of I2-GNN | 2-DRFWL(2) GNN · QM9 ZPVE | MAE: 0.00014 -> 0.00017 | Hurt |
| use 2-DRFWL(2) GNN instead of I2-GNN | 2-DRFWL(2) GNN · QM9 mu | MAE: 0.428 -> 0.346 | Improved |
Closest papers by meaning
- Boosting the Cycle Counting Power of Graph Neural Networks with I$^2$-GNNs (ICLR 2023 · similarity 0.94)
- On the Scalability of GNNs for Molecular Graphs (NeurIPS 2024 · similarity 0.88)
- Homomorphism Counts for Graph Neural Networks: All About That Basis (ICML 2024 · similarity 0.87)
- Chaining 2-FWL GNNs for Combinatorial Graph Alignment (NeurIPS 2026 · similarity 0.87)
- Improved Expressivity of Hypergraph Neural Networks through High-Dimensional Generalized Weisfeiler-Leman Algorithms (ICML 2025 · similarity 0.87)
- A Flexible, Equivariant Framework for Subgraph GNNs via Graph Products and Graph Coarsening (NeurIPS 2024 · similarity 0.87)
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.
