Research map / Graph learning / GNN architectures
Same Graph Cross-Task Transfer in GNNs: Protocols and Predictors
Neelam Akula, Surbhi Kumar, Murat Kantarcioglu, Baris Coskunuzer
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 |
|---|---|---|---|
| transfer NC supervision to LP via multi-view or joint training on homophilic graphs | GCN, GraphSAGE, GPS · CORA, CITESEER, PUBMED (NC->LP) | LP AUC gain: CORA GCN Joint +12.3 AUC points | Improved |
| reuse LP representation by replacing node features (ET Rep) for NC | GCN · CORA, CITESEER, PUBMED (LP->NC) | NC accuracy gain: -12.7, -16.8, -21.4 | Hurt |
| LP->NC transfer via ET Concat/Joint on structure-dominant graphs | GCN · USA, EUROPE, BRAZIL (LP->NC) | NC accuracy gain: USA ET Concat +7.8; BRAZIL Joint +13.7 | Improved |
| LP->NC transfer on homophilic graphs with warm start, MV or Joint | GCN · CORA, CITESEER, PUBMED (LP->NC) | NC accuracy gain: WS +0.6, +0.5, +0.1 | No clear effect |
| NC->LP via ET Rep on heterophilic ROMAN-EMPIRE | GCN, GraphSAGE · ROMAN-EMPIRE (NC->LP) | LP AUC gain: -10.6 (GCN), -12.5 (GraphSAGE) | Hurt |
| NC->LP warm start on USA | GCN · USA (NC->LP) | LP AUC gain: +0.3 (WS) | No clear effect |
| add Joint training for GPS on ROMAN-EMPIRE NC->LP | GPS · ROMAN-EMPIRE (NC->LP) | LP AUC gain: +27.9 over base 68.8 | Improved |
| couple objectives with MV or Joint rather than post hoc reuse on heterophilic graphs | GCN, GraphSAGE, GPS · TEXAS, CORNELL, WISCONSIN, ACTOR, ROMAN-EMPIRE | CoTask Score: TEXAS GPS Joint 38.4 vs ET Rep 19.6 | Improved |
| homophily as predictor of NC->LP gains | GCN · 11 datasets | Pearson r with gain: ET Concat r with Hn 0.904 | Improved |
| homophily as predictor of LP->NC gains | GCN · 11 datasets | Pearson r with gain: ET Rep He -0.114, Hn -0.353 | No clear effect |
| clustering coefficient as predictor of LP->NC gains for ET Concat | GCN · 11 datasets | Pearson r with gain: r = 0.782 | Improved |
| transfer on mixed-regime graphs measured by CTS | GraphSAGE, GPS · USA, EUROPE, BRAZIL | CoTask Score: EUROPE GPS MV -15.4; BRAZIL GPS ET Con 13.9 | Mixed |
Closest papers by meaning
- A Topological Perspective on Demystifying GNN-Based Link Prediction Performance (ICLR 2024 · similarity 0.88)
- GraphBridge: Towards Arbitrary Transfer Learning in GNNs (ICLR 2025 · similarity 0.88)
- GNNs Meet Sequence Models Along the Shortest-Path: an Expressive Method for Link Prediction (NeurIPS 2026 · similarity 0.88)
- Towards Pre-trained Graph Condensation via Optimal Transport (NeurIPS 2025 · similarity 0.87)
- Towards Dynamic Message Passing on Graphs (NeurIPS 2024 · similarity 0.87)
- On the Impact of Feature Heterophily on Link Prediction with Graph Neural Networks (NeurIPS 2024 · 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.
