Research map / Label-efficient and robust learning / Test-time adaptation
GFMate: Empowering Graph Foundation Models with Test-time Prompt Tuning
Yan Jiang, Ruihong Qiu, Zi Huang
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 |
|---|---|---|---|
| add test-time graph complementary learning (use unlabelled test nodes) on top of few-shot prompt tuning (ablation) | GFMate · Cornell, 1-shot | accuracy (%): 72.59 -> 79.67 | Improved |
| add test-time graph complementary learning (ablation) | GFMate · Cora, 1-shot | accuracy (%): 54.57 -> 59.68 | Improved |
| use complementary labels instead of pseudo-labels from the last layer for test nodes (ablation) | GFMate · Cora, 1-shot | accuracy (%): 18.29 -> 59.68 | Improved |
| add pseudo-labelled test nodes to few-shot tuning (ablation) | GFMate few-shot tuning · Citeseer, 1-shot | accuracy (%): 52.92 -> 13.79 | Hurt |
| increase the fraction of available unlabelled test nodes from 0% to full (ablation) | GFMate · Cora, 1-shot | accuracy (%): 54.57 -> 59.68 | Improved |
| remove centroid prompt, layer prompt, or TGCL (ablation) | GFMate · one-shot and three-shot node classification | accuracy | Improved |
| plug GFMate into LP+FT pre-trained GCN | LP+FT · Cornell, 1-shot | accuracy (%): 23.88 -> 79.67 | Improved |
| plug GFMate into SAMGPT | SAMGPT · Cornell, 1-shot | accuracy (%): 59.34 -> 76.34 | Improved |
| swap GCN backbone for SAGE, GAT or H2GCN (ablation) | GFMate · Citeseer / Cora, 1-shot | accuracy (%): 56.25 -> 57.39 (SAGE, Citeseer) | No clear effect |
| reduce task to binary classification (ablation) | GFMate vs SAMGPT · Cora | accuracy gain over SAMGPT: 6.85 -> 10.58 | Improved |
| pre-train GFM only on bioinformatics-domain graphs instead of all other domains (ablation) | GFMate · Citeseer, 1-shot | accuracy (%): 56.25 -> 48.09 | Hurt |
| shuffle 50% of test node features (feature shift) (ablation) | GFMate · Texas, 1-shot | accuracy (%): 76.63 -> 71.24 | Hurt |
Closest papers by meaning
- Edge Prompt Tuning for Graph Neural Networks (ICLR 2025 · similarity 0.88)
- Towards Effective Federated Graph Foundation Model via Mitigating Knowledge Entanglement (NeurIPS 2025 · similarity 0.88)
- GFT: Graph Foundation Model with Transferable Tree Vocabulary (NeurIPS 2024 · similarity 0.87)
- Position: Graph Foundation Models Are Already Here (ICML 2024 · similarity 0.87)
- GraphTOP: Graph Topology-Oriented Prompting for Graph Neural Networks (NeurIPS 2025 · similarity 0.87)
- HGPrompt: Bridging Homogeneous and Heterogeneous Graphs for Few-Shot Prompt Learning (AAAI 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.
