Research map / Federated learning and privacy / Personalized and heterogeneous FL
Decoupling Shared and Personalized Knowledge: A Dual-Branch Federated Learning Framework for Multi-Domain with Non-IID Data
Yiran Pang, Zhen Ni, Xiangnan Zhong
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 |
|---|---|---|---|
| add dual-branch private/shared split with local pre-training (pFedDB) instead of FedAvg | FedAvg · DomainNet (avg) | accuracy (%): 41.5 -> 50.3 | Improved |
| add two-phase training (local pretraining before federation) (ablation) | pFedDB · Office-Caltech-10 (avg) | accuracy (%): 71.1 -> 73.4 | Improved |
| add two-phase training (local pretraining before federation) (ablation) | pFedDB · Chest-X-Ray-4 (avg) | accuracy (%): 81.3 -> 83.4 | Improved |
| use element-wise addition instead of attention fusion (ablation) | pFedDB · Office-Caltech-10 (avg) | accuracy (%): 70.5 -> 73.4 | Improved |
| use element-wise addition instead of weighted sum alpha=0.5 (ablation) | pFedDB · Office-Caltech-10 (avg) | accuracy (%): 72.7 -> 73.4 | Improved |
| set cut depth so shared branch has about 70% of parameters instead of 99% (ablation) | pFedDB · Chest-X-Ray-4 (avg) | accuracy (%): 82.7 -> 83.4 | Improved |
| set cut depth to share 71% of parameters instead of only 4% (ablation) | pFedDB · Office-Caltech-10 (avg) | accuracy (%): 66.8 -> 73.4 | Improved |
| communicate only the shared branch | pFedDB · three backbones | per-round communication: 30-37% fewer bytes per round | Improved |
| freeze shared branch for newly joining client rather than fine-tune all (ablation) | pFedDB · MNIST-M and Sketch new-client adaptation | validation accuracy | Improved |
| compare pFedDB to FedWon on the SVHN client | pFedDB · Digits-Five (Sv) | accuracy (%): 71.3 -> 68.6 | Mixed |
| use TinyViT backbone instead of AlexNet | pFedDB · Office-Caltech-10 (avg) | accuracy (%): Single 73.5 -> 75.9 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FedWon — FedWon: Triumphing Multi-domain Federated Learning Without Normalization (ICLR 2024) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- pFedClub: Controllable Heterogeneous Model Aggregation for Personalized Federated Learning (NeurIPS 2024 · similarity 0.89)
- Decentralized Directed Collaboration for Personalized Federated Learning (CVPR 2024 · similarity 0.89)
- FedAS: Bridging Inconsistency in Personalized Federated Learning (CVPR 2024 · similarity 0.89)
- Overcoming Data and Model heterogeneities in Decentralized Federated Learning via Synthetic Anchors (ICML 2024 · similarity 0.88)
- Low Precision Local Training is Enough for Federated Learning (NeurIPS 2024 · similarity 0.88)
- Understanding the Statistical Accuracy-Communication Trade-off in Personalized Federated Learning with Minimax Guarantees (ICML 2025 · similarity 0.88)
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.
