Research map / Graph learning / GNN architectures
Rethinking Graph Transformers as Graph Signal Denoisers: The Role of Block-Diagonal Priors
Jiaming Zhuo, Ziyi Ma, Kun Fu, Di Jin, Chuan Wang, Zhen Wang, Xiaochun Cao, Huazhu Fu, Liang Yang
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 BDFormer block-diagonal anchor propagation instead of SGFormer | Graph Transformer · Chameleon | accuracy: 43.77 -> 47.90 | Improved |
| use BDFormer instead of SGFormer | Graph Transformer · Squirrel | accuracy: 41.09 -> 44.20 | Improved |
| use BDFormer instead of M3Dphormer | Graph Transformer · Minesweeper | ROC-AUC: 97.83 -> 97.35 | Hurt |
| use BDFormer instead of SGFormer | Graph Transformer · ogbn-proteins | ROC-AUC: 79.53 -> 80.31 | Improved |
| use BDFormer instead of Polynormer | Graph Transformer · ogbn-arxiv | accuracy: 73.46 -> 73.66 | No clear effect |
| use BDFormer instead of Polynormer | Graph Transformer · ogbn-products | accuracy: 83.82 -> 83.98 | Improved |
| use BDFormer instead of SGFormer on long-range benchmark | Graph Transformer · Paris | accuracy: 52.00 -> 59.01 | Improved |
| use BDFormer instead of Polynormer on long-range benchmark | Graph Transformer · Shanghai | accuracy: 71.42 -> 71.69 | Improved |
| prune inter-class edges with learned anchor affinity (topology refinement) | BDFormer · Chameleon / CiteSeer edge ratios | inter/intra-class edge ratio | Improved |
| increase number of anchors k and blocks g beyond k>=64, g>=10 (ablation) | BDFormer · Hyperparameter sweep | accuracy: gains saturate at k >= 64 and g >= 10 | No clear effect |
| vary input graph size from 20K to 100K nodes | BDFormer · ogbn-arxiv | training time and GPU memory: linear growth | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- SGFormer — SGFormer: Simplifying and Empowering Transformers for Large-Graph Representations (NeurIPS 2023) · better on 11 of 11 shared results
- Polynormer — Polynormer: Polynomial-Expressive Graph Transformer in Linear Time (ICLR 2024) · better on 5 of 5 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
- Is Noise Conditioning Necessary? A Unified Theory of Unconditional Graph Diffusion Models (NeurIPS 2025 · similarity 0.89)
- Joint Graph Rewiring and Feature Denoising via Spectral Resonance (ICLR 2025 · similarity 0.87)
- From Theory to Practice: Rethinking Green and Martin Kernels for Unleashing Graph Transformers (ICML 2025 · similarity 0.87)
- Less is More: on the Over-Globalizing Problem in Graph Transformers (ICML 2024 · similarity 0.87)
- Generalizable Insights for Graph Transformers in Theory and Practice (NeurIPS 2025 · similarity 0.86)
- Even Sparser Graph Transformers (NeurIPS 2024 · similarity 0.86)
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.
