Research map / Graph learning / Graph self-supervised learning
When Do Graph Foundation Models Transfer? A Data-Centric Theory
Jiajun Zhu, Ying Chen, Peihao Wang, Yixuan He, Pan Li, Aditya Akella, Zhangyang Wang
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 |
|---|---|---|---|
| shift training budget toward larger graphs (larger lambda) (ablation) | DeepSets, synthetic graphon classification · synthetic graphon task | test error | Mixed |
| shift training budget toward larger graphs (larger lambda) (ablation) | DeepSets · synthetic graphon task | train-test token discrepancy | Improved |
| replace DeepSets with GIN backbone (ablation) | size-shift study · synthetic graphon task | test error | No clear effect |
| add graph merging augmentation (1% larger synthetic graphs) | DeepSets · synthetic graphon task at lambda 0 or 1 | test error | Improved |
| add graph merging augmentation | DeepSets · synthetic graphon task, intermediate lambda | test error | No clear effect |
| add 1% graph-merging augmentation | classifier · COLLAB | test error: 0.4384 -> 0.4069 | Improved |
| add 3% graph-merging augmentation | classifier · IMDB-BINARY | test error: 0.4631 -> 0.4508 | Improved |
| add 4% graph-merging augmentation | classifier · IMDB-BINARY | test error: 0.4631 -> 0.5369 | Hurt |
| add 1% graph-merging augmentation | classifier · REDDIT-BINARY | test error: 0.4108 -> 0.4367 | Hurt |
| add 3% graph-merging augmentation | classifier · REDDIT-BINARY | test error: 0.4108 -> 0.4084 | Improved |
| perturb test graphon (levels 0-5) for out-of-graphon samples (ablation) | DeepSets lambda=0.2 · synthetic graphon task | out-of-graphon error | Hurt |
| perturb test graphon, in-graphon samples (ablation) | DeepSets lambda=0.2 · synthetic graphon task | in-graphon error | No clear effect |
Closest papers by meaning
- How Much Can Transfer? BRIDGE: Bounded Multi-Domain Graph Foundation Model with Generalization Guarantees (ICML 2025 · similarity 0.88)
- GFT: Graph Foundation Model with Transferable Tree Vocabulary (NeurIPS 2024 · similarity 0.88)
- EvoluNet: Advancing Dynamic Non-IID Transfer Learning on Graphs (ICML 2024 · similarity 0.87)
- Multi-Domain Graph Foundation Models: Robust Knowledge Transfer via Topology Alignment (ICML 2025 · similarity 0.87)
- Position: Graph Foundation Models Are Already Here (ICML 2024 · similarity 0.87)
- Contextual Structure Knowledge Transfer for Graph Neural Networks (AAAI 2025 · similarity 0.87)
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.
