Research map / Federated learning and privacy / Personalized and heterogeneous FL
FlatLand: Personalized Graph Federated Learning via Tailored Lorentz Space
Jiahong Liu, Ram Samarth B B, Xinyu Fu, Menglin Yang, Weixi Zhang, Rex Ying, Irwin King
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 Lorentz backbone instead of Euclidean for parameter decoupling (ablation) | FlatLand · Cora (20 clients) | accuracy: 76.23 -> 82.49 | Improved |
| use Lorentz backbone instead of Euclidean for parameter decoupling (ablation) | FlatLand · CiteSeer (20 clients) | accuracy: 66.29 -> 72.24 | Improved |
| add tailored client-specific curvature (vs constant K=1) (ablation) | FlatLand · Cora (10 clients) | accuracy | Improved |
| add time-like parameter decoupling strategy (ablation) | FlatLand · Cora (10 clients) | accuracy | Improved |
| vary curvature initialization (Ricci, Ollivier, constant, MLP) (ablation) | FlatLand · node classification | accuracy | No clear effect |
| use hyperbolic Local (L) instead of Euclidean Local (E) | Local training · CiteSeer (10 clients) | accuracy: 67.82 -> 72.30 | Improved |
| use hyperbolic Local (L) instead of Euclidean Local (E) | Local training · BIO-CHEM-SN graph classification | accuracy: 67.17 -> 65.31 | Hurt |
| add FlatLand to Lorentz local training | Local (L) · CiteSeer (20 clients) | accuracy: 69.52 -> 72.24 | Improved |
| reduce embedding dimension from 64 to 4 | FlatLand and FedHGCN · CiteSeer (20 clients) | accuracy | No clear effect |
| use 10% client participation | FlatLand vs FedAvg · Cora (50 clients) | accuracy: FedAvg 18.14% vs FlatLand 81.82% | Improved |
| use FlatLand vs FED-PUB | FlatLand · Cora (10 clients) | accuracy: 81.54 -> 80.46 | Hurt |
| replace FedAvg with FlatLand | federated GNN · Cora (10 clients) | accuracy: 69.19 -> 80.46 | Improved |
Closest papers by meaning
- FedSSP: Federated Graph Learning with Spectral Knowledge and Personalized Preference (NeurIPS 2024 · similarity 0.88)
- Subgraph Federated Learning via Spectral Methods (NeurIPS 2025 · similarity 0.87)
- Federated Graph Learning under Domain Shift with Generalizable Prototypes (AAAI 2024 · similarity 0.87)
- Less is More: Federated Graph Learning with Alleviating Topology Heterogeneity from A Causal Perspective (ICML 2025 · similarity 0.87)
- Personalized Subgraph Federated Learning with Differentiable Auxiliary Projections (NeurIPS 2025 · similarity 0.87)
- FedSPA: Generalizable Federated Graph Learning under Homophily Heterogeneity (CVPR 2025 · 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.
