Research map / Graph learning / GNN architectures
SGFormer: Simplifying and Scaling Graph Transformers With Single-Layer Attention and Approximation-Free Linear Complexity
Qitian Wu, Kai Yang, Hengrui Zhang, David Wipf, Junchi Yan
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 single-layer global attention (SGFormer) instead of GCN | SGFormer · Actor | accuracy (%): 30.1 -> 37.9 | Improved |
| use SGFormer instead of NodeFormer | SGFormer · Chameleon | accuracy (%): 34.7 -> 44.9 | Improved |
| use SGFormer instead of DIFFormer on homophilic citation graph | SGFormer · Cora | accuracy (%): 85.9 -> 84.5 | Hurt |
| use SGFormer instead of DIFFormer on PubMed | SGFormer · PubMed | accuracy (%): 81.8 -> 80.3 | Hurt |
| use SGFormer instead of NodeFormer on ogbn-arxiv | SGFormer · ogbn-arxiv | accuracy (%): 59.90 -> 72.63 | Improved |
| use SGFormer instead of DIFFormer on pokec | SGFormer · pokec | accuracy (%): 69.24 -> 73.76 | Improved |
| use SGFormer on gigantic graph | SGFormer · ogbn-papers100M | accuracy (%): 65.11 -> 66.01 | Improved |
| use SGFormer instead of DIFFormer for training speed | SGFormer · Cora | ms per epoch: 49.7 -> 15.0 | Improved |
| use SGFormer instead of DIFFormer for inference speed | SGFormer · Cora | inference ms: 9.6 -> 3.8 | Improved |
| use SGFormer instead of DIFFormer for memory | SGFormer · Amazon2M | GPU memory (GB): 4.5 -> 2.7 | Improved |
| use smaller Graphormer instead of GCN on small labeled set | Graphormer-Smaller · Cora | accuracy (%): 81.6 -> 75.8 | Hurt |
| use SGFormer instead of NodeFormer with 1% labels | SGFormer · Cora (1% training ratio) | accuracy (%): 53.47 -> 65.14 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- NodeFormer — NodeFormer: A Scalable Graph Structure Learning Transformer for Node Classification (NeurIPS 2022) · better on 12 of 12 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Generalizable Insights for Graph Transformers in Theory and Practice (NeurIPS 2025 · similarity 0.90)
- Unifying and Enhancing Graph Transformers via a Hierarchical Mask Framework (NeurIPS 2025 · similarity 0.90)
- Towards Principled Graph Transformers (NeurIPS 2024 · similarity 0.90)
- Even Sparser 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)
- 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.
