Research map / Graph learning / Graph algorithms and clustering
Rethinking the Gold Standard: Why Discrete Curvature Fails to Fully Capture Over-squashing in GNNs?
Jialong Chen, Bowen Deng, Zibin Zheng, Chuan Chen
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 |
|---|---|---|---|
| use Augmented Forman-3 instead of Ollivier Ricci curvature | GCN · 21 datasets average | MOSR10: .271 -> .067 | Improved |
| use Balanced Forman instead of Ollivier Ricci curvature | GCN · 21 datasets average | MOSR10: .271 -> .078 | Improved |
| add degree weighting f=1/(1+x) to AF3 (WAF3) | GCN · 21 datasets average | MOSR10: .067 -> .036 | Improved |
| add degree weighting f=1/(1+x) to AF3 (WAF3) | GAT · 21 datasets average | MOSR10: .148 -> .111 | Improved |
| add degree weighting f=1/(1+x) to AF3 (WAF3) | GraphSAGE · 21 datasets average | MOSR25: .161 -> .124 | Improved |
| replace GCN with GAT or GraphSAGE as GNN | Ollivier Ricci · 21 datasets average | MOSR10: .271 -> .336 -> .386 | Hurt |
| raise q from 10 to 25 | Ollivier Ricci GCN · 21 datasets average | MOSR: .271 -> .307 | Hurt |
| use MinHash approximation with H=100 instead of exact WAF3 | WAF3 · Erdos-Renyi 5M edges | computation time: 3142s -> 23.5s, 133.7x speed up | Improved |
| raise MinHash hashes from H=100 to H=1000 (ablation) | WAF3 approximation · 21 datasets | Kendall tau-b: 95% -> over 98% | Improved |
| raise MinHash hashes from H=1000 to H=10000 (ablation) | WAF3 approximation · 21 datasets | Kendall tau-b: less than 1% improvement | No clear effect |
| train GNN for more epochs before measuring MOSR (ablation) | GNNs · Citeseer, Cora, Pubmed | MOSR: slowly increases then stabilises | Hurt |
| use f=(1+d)^-1 instead of (1+d)^-2 or (1+d)^-1/2 (ablation) | WAF3 · MOSR across datasets | MOSR: slightly smaller on average | Improved |
Closest papers by meaning
- The Effectiveness of Curvature-Based Rewiring and the Role of Hyperparameters in GNNs Revisited (ICLR 2025 · similarity 0.89)
- Discrete Curvature Graph Information Bottleneck (AAAI 2025 · similarity 0.88)
- Get a GRIP, this will be a long TRIP: A Quantifiable Long-Range Framework for Verifying Over-squashing (NeurIPS 2026 · similarity 0.88)
- Delaunay Graph: Addressing Over-Squashing and Over-Smoothing Using Delaunay Triangulation (ICML 2024 · similarity 0.88)
- CurvGAD: Leveraging Curvature for Enhanced Graph Anomaly Detection (ICML 2025 · similarity 0.87)
- Spectro-Riemannian Graph Neural Networks (ICLR 2025 · 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.
