Research map / Graph learning / Graph self-supervised learning
Learning Graph Foundation Models on Riemannian Graph-of-Graphs
Haokun Liu, Zezhong Ding, Xike Xie
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 R-GFM instead of the strongest baseline | GFM · Citeseer node classification | relative improvement: 18.9% improvement on Citeseer | Improved |
| add 50% edge drop robustness (R-GFM vs strongest baseline) | R-GFM · edge drop p=0.5 | accuracy: 46.91% (drop of 5.76) vs around 41.2% | Improved |
| mask 50% of node attributes (R-GFM vs strongest baseline) | R-GFM · node masking r=0.5 | accuracy: 46.84% vs 41.78% | Improved |
| add GoG construction (vs Stage-1 only, w/o GoG pooling, chain-edge GoG) (ablation) | R-GFM · Photo, Cora, Chameleon, Texas 1-shot | accuracy | Improved |
| use Riemannian encoder instead of Euclidean on GoG (ablation) | R-GFM · Photo, Cora, Chameleon, Texas 1-shot | accuracy | Improved |
| use mixture of manifolds instead of hyperbolic only or hyperspherical only (ablation) | R-GFM · Photo, Cora, Chameleon, Texas 1-shot | accuracy | Improved |
| increase hop number K | GCOPE extended to K-hop · cross-domain 1-shot node classification | accuracy | No clear effect |
| increase hop number K | R-GFM · cross-domain 1-shot node classification | accuracy | Improved |
| use subgraph sampling instead of full-graph training | R-GFM vs SAMGPT and MDGFM · large LM-feature graphs | memory / accuracy: OOM for SAMGPT and MDGFM | Improved |
| set GoG edge retention ratio beta near 0.6 (ablation) | R-GFM · four datasets 1-shot | accuracy: peaks around beta 0.6 | Improved |
| use dynamic top-m gating instead of fixed m in 1..5 (ablation) | R-GFM · four datasets 1-shot | accuracy | Improved |
| use dynamic candidate expert set instead of manual expert number (ablation) | R-GFM · 10-dataset setting | accuracy | Improved |
Closest papers by meaning
- Position: Graph Foundation Models Are Already Here (ICML 2024 · similarity 0.88)
- Distilling Graph Geometry: Knowledge Gap from GNNs to MLPs (NeurIPS 2026 · similarity 0.87)
- Toward a Unified Geometry Understanding : Riemannian Diffusion Framework for Graph Generation and Prediction (NeurIPS 2025 · similarity 0.87)
- GFT: Graph Foundation Model with Transferable Tree Vocabulary (NeurIPS 2024 · similarity 0.87)
- DisRFM: Polar Riemannian Flow Matching for Structure-Preserving Graph Domain Adaptation (NeurIPS 2026 · similarity 0.87)
- Towards Graph Foundation Models: Learning Generalities Across Graphs via Task-Trees (ICML 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.
