Research map / Graph learning / GNN architectures
Graph Coarsening with 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 |
|---|---|---|---|
| Learn coarse-edge weights for existing coarsening methods | Graph coarsening on synthetic and real networks · Combinatorial-Laplacian quadratic loss; BA, ER, GEO, WS, CS, Flickr, Physics, PubMed, Shape | relative improvement percentage: 72.7%; 7.4%; 87.3%; 84.2%; 40.0%; 21.8%; 60.3%; 36.4%; 91.4% | Helped |
| Learn coarse-edge weights with GNN rather than mean-pooling MLP (ablation) | Graph coarsening weight assignment · WS, reduction ratio 0.5 | relative improvement percentage: 58.5% -> 84.2% | Helped |
| Learn coarse-edge weights with GNN rather than mean-pooling MLP (ablation) | Graph coarsening weight assignment · Shape, reduction ratio 0.5 | relative improvement percentage: -1.9% -> 79.4% | Helped |
| Use mean-pooling MLP instead of GNN for weight assignment (ablation) | Graph coarsening weight assignment · WS, reduction ratio 0.5 | relative improvement percentage: 84.2% -> 58.5% | Hurt |
| Use mean-pooling MLP instead of GNN for weight assignment (ablation) | Graph coarsening weight assignment · Shape, reduction ratio 0.5 | relative improvement percentage: 79.4% -> -1.9% | Hurt |
| Train with Rayleigh loss as a proxy for Eigenerror | GOREN · Eigenerror; synthetic and real networks | relative improvement percentage: Examples include 93.3% and -377.9%. | Mixed |
| Use normalized-Laplacian quadratic loss | GOREN · GEO, Local var (edges) | relative improvement percentage: -145.1% | Hurt |
| Use normalized-Laplacian quadratic loss | GOREN · Flickr | relative improvement percentage: -31.9%; -27.6%; -67.2%; -73.8%; -440.1%; -43.9% | Hurt |
Closest papers by meaning
- Featured Graph Coarsening with Similarity Guarantees (ICML 2023 · similarity 0.93)
- Graph Coarsening via Supervised Granular-Ball for Scalable Graph Neural Network Training (AAAI 2025 · similarity 0.89)
- Graph Coarsening with Message-Passing Guarantees (NeurIPS 2024 · similarity 0.89)
- Towards Scalable and Deep Graph Neural Networks via Noise Masking (AAAI 2025 · similarity 0.88)
- How Graph Neural Networks Learn: Lessons from Training Dynamics (ICML 2024 · similarity 0.88)
- Taxonomy of reduction matrices for Graph Coarsening (NeurIPS 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.