Research map / Graph learning / GNN architectures
Federated Learning on Non-IID Graphs via Structural Knowledge Sharing
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Share only the structure encoder while keeping feature learning local (ablation) | FedStar · BIO-CHEM, BIO-CHEM-SN, BIO-SN-CV | accuracy: 74.54$\pm$2.50 vs 74.08$\pm$2.45; 72.15$\pm$2.43 vs 71.30$\pm$1.89; 69.49$\pm$1.81 vs 68.76$\pm$2.24 | Helped |
| Share all parameters rather than only the structure encoder (ablation) | FedStar · BIO-CHEM, BIO-CHEM-SN, BIO-SN-CV | accuracy: 74.54$\pm$2.50 vs 71.97$\pm$2.14; 72.15$\pm$2.43 vs 69.85$\pm$2.43; 69.49$\pm$1.81 vs 65.78$\pm$4.25 | Hurt |
| Share the feature encoder rather than the structure encoder (ablation) | FedStar · BIO-CHEM, BIO-CHEM-SN, BIO-SN-CV | accuracy: 74.54$\pm$2.50 vs 71.00$\pm$3.51; 72.15$\pm$2.43 vs 68.53$\pm$2.74; 69.49$\pm$1.81 vs 64.14$\pm$2.73 | Hurt |
| Decouple feature and structure channels (ablation) | FedStar · BIO-CHEM, BIO-CHEM-SN, BIO-SN-CV | accuracy: 70.86$\pm$2.25 -> 71.97$\pm$2.14; 69.32$\pm$2.42 -> 69.85$\pm$2.43; 65.23$\pm$2.52 -> 65.78$\pm$4.25 | Helped |
| Use both DSE and RWSE instead of all-ones structure embeddings (ablation) | FedStar · BIO-CHEM, BIO-CHEM-SN, BIO-SN-CV | accuracy: 69.51$\pm$2.25 -> 74.54$\pm$2.50; 69.64$\pm$1.92 -> 72.15$\pm$2.43; 66.05$\pm$2.92 -> 69.49$\pm$1.81 | Helped |
| Add DSE when RWSE is present (ablation) | FedStar · BIO-CHEM, BIO-CHEM-SN, BIO-SN-CV | accuracy: 72.74$\pm$3.44 -> 74.54$\pm$2.50; 70.48$\pm$3.37 -> 72.15$\pm$2.43; 67.23$\pm$2.74 -> 69.49$\pm$1.81 | Helped |
| Add RWSE to DSE (ablation) | FedStar · BIO-CHEM, BIO-CHEM-SN, BIO-SN-CV | accuracy: 74.42$\pm$3.15 -> 74.54$\pm$2.50; 72.05$\pm$2.82 -> 72.15$\pm$2.43; 69.25$\pm$2.41 -> 69.49$\pm$1.81 | Helped |
| Increase local epochs from 1 to 4 (ablation) | FedStar · CHEM | accuracy: 79.79$\pm$2.44 -> 80.58$\pm$2.48 | Helped |
| Increase local epochs from 1 to 4 (ablation) | FedAvg · CHEM | accuracy: 75.26$\pm$2.00 -> 76.36$\pm$3.36 | Helped |
| Increase local epochs from 1 to 4 (ablation) | GCFL · CHEM | accuracy: 76.49$\pm$1.23 -> 76.27$\pm$2.14 | No clear effect |
| Use FedStar instead of GCFL | Federated graph classification · CHEM, BIO-CHEM, BIO-CHEM-SN, BIO-SN-CV | accuracy: 79.79$\pm$2.44 vs 76.49$\pm$1.23; 74.54$\pm$2.50 vs 71.60$\pm$2.20; 72.16$\pm$2.43 vs 70.65$\pm$1.84; 69.49$\pm$1.81 vs | Helped |
Closest papers by meaning
- Federated Graph-Level Clustering Network (AAAI 2025 · similarity 0.92)
- Less is More: Federated Graph Learning with Alleviating Topology Heterogeneity from A Causal Perspective (ICML 2025 · similarity 0.91)
- FedSSP: Federated Graph Learning with Spectral Knowledge and Personalized Preference (NeurIPS 2024 · similarity 0.91)
- Decoupled Subgraph Federated Learning (ICLR 2025 · similarity 0.91)
- Towards Effective Federated Graph Foundation Model via Mitigating Knowledge Entanglement (NeurIPS 2025 · similarity 0.91)
- $S^2$FGL: Spatial Spectral Federated Graph Learning (ICML 2025 · similarity 0.90)
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.