Research map / Graph learning / GNN architectures
Recipe for a General, Powerful, Scalable Graph Transformer
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 global attention to GPS (ablation) | GPS · PCQM4Mv2-Subset | MAE: 0.1213 ± 0.0002 -> 0.1159 ± 0.0004 | Helped |
| Add global attention to GPS (ablation) | GPS · ZINC | MAE: 0.070 ± 0.004 -> 0.070 ± 0.004 | No clear effect |
| Remove the MPNN (ablation) | GPS · CIFAR10 | Accuracy: 72.305 ± 0.344 -> 68.862 ± 1.138 | Hurt |
| Remove the MPNN (ablation) | GPS · MalNet-Tiny | Accuracy: 92.64 ± 0.78 -> 73.90 ± 0.58 | Hurt |
| Remove RWSE (ablation) | GPS · PCQM4Mv2-Subset | MAE: 0.1159 ± 0.0004 -> 0.1355 ± 0.0035 | Hurt |
| Replace RWSE with PEG LapEig (ablation) | GPS · ZINC | MAE: 0.070 ± 0.002 -> 0.936 ± 0.143 | Hurt |
| Use SignNet DeepSets instead of RWSE (ablation) | GPS · PCQM4Mv2-Subset | MAE: 0.1159 ± 0.0004 -> 0.1144 ± 0.0002 | Helped |
| Use PEG LapEig instead of LapPE (ablation) | GPS · PascalVOC-SP | F1 score: 0.3736 ± 0.0158 -> 0.3956 ± 0.0084 | Helped |
| Remove LapPE (ablation) | GPS · PascalVOC-SP | F1 score: 0.3736 ± 0.0158 -> 0.3846 ± 0.0156 | Helped |
| Replace Transformer attention with Performer (ablation) | GPS · MalNet-Tiny | Accuracy: 93.50 ± 0.41 -> 92.64 ± 0.78 | Hurt |
| Use Performer global attention (ablation) | GPS · MalNet-Tiny | Accuracy: 92.23 ± 0.65 -> 92.64 ± 0.78 | Helped |
| Combine local message passing with global attention and graph encodings | GPS · PascalVOC-SP | F1 score: 0.3230 ± 0.0039 -> 0.3748 ± 0.0109 | Helped |
Closest papers by meaning
- Generalizable Insights for Graph Transformers in Theory and Practice (NeurIPS 2025 · similarity 0.93)
- Towards Principled Graph Transformers (NeurIPS 2024 · similarity 0.90)
- What Improves the Generalization of Graph Transformers? A Theoretical Dive into the Self-attention and Positional Encoding (ICML 2024 · similarity 0.90)
- Supra-Laplacian Encoding for Transformer on Dynamic Graphs (NeurIPS 2024 · similarity 0.90)
- A Scalable and Effective Alternative to Graph Transformers (AAAI 2025 · similarity 0.90)
- Less is More: on the Over-Globalizing Problem in Graph Transformers (ICML 2024 · similarity 0.90)
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.