Research map / Federated learning and privacy / Personalized and heterogeneous FL
FedMerge: Federated Model Merging for Personalization
Shutong Chen, Tianyi Zhou, Guodong Long, Jing Jiang, Chengqi Zhang
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 dynamic (learned) merging weights instead of fixed 1/15 (ablation) | FedMerge (15 global models) · CIFAR-100 Personal Non-IID | accuracy: 40.94 -> 51.18 | Improved |
| use dynamic merging weights instead of fixed 5/15 (ablation) | FedMerge (15 global models) · CIFAR-100 Personal Non-IID | accuracy: 44.70 -> 51.18 | Improved |
| use dynamic merging weights instead of fixed 10/15 (ablation) | FedMerge (15 global models) · CIFAR-100 Personal Non-IID | accuracy: 45.48 -> 51.18 | Improved |
| use dynamic merging weights instead of fixed 1/15 (ablation) | FedMerge (15 global models) · CIFAR-100 Cluster Non-IID | accuracy: 32.55 -> 53.27 | Improved |
| use dynamic merging weights instead of fixed 5/15 (ablation) | FedMerge (15 global models) · CIFAR-100 Cluster Non-IID | accuracy: 32.52 -> 53.27 | Improved |
| use dynamic merging weights instead of fixed 10/15 (ablation) | FedMerge (15 global models) · CIFAR-100 Cluster Non-IID | accuracy: 34.06 -> 53.27 | Improved |
| replace FedAvg with FedMerge (15 global models) | FedAvg · CIFAR-100 Cluster Non-IID | accuracy: 30.29 -> 53.27 | Improved |
| scale up single-model architecture to larger ResNets | single-model FL (FedAvg, pFedMoE) · CIFAR-100 Personal/Cluster Non-IID | accuracy | Hurt |
| increase number of ResNet-9 global models for multi-model methods | FedMerge, FedEM, IFCA · CIFAR-100 non-IID | accuracy | Improved |
| scale up ResNet architecture | single-model FL · PACS non-IID | accuracy | Improved |
| replace pFedMoE (two ResNet-18 per client) with FedMerge (one ResNet-9 per client) | pFedMoE · CIFAR-100 Personal Non-IID | accuracy: 54.34 -> 54.38 | No clear effect |
| replace FedEM with FedMerge | FedEM · Flan federated (average of 8 tasks) | ROUGE-1: 73.77 -> 74.86 | Improved |
Closest papers by meaning
- Personalized Federated Learning with First Order Model Optimization (ICLR 2021 · similarity 0.92)
- FedImpro: Measuring and Improving Client Update in Federated Learning (ICLR 2024 · similarity 0.89)
- FedAS: Bridging Inconsistency in Personalized Federated Learning (CVPR 2024 · similarity 0.89)
- pFedClub: Controllable Heterogeneous Model Aggregation for Personalized Federated Learning (NeurIPS 2024 · similarity 0.88)
- FedSMU: Communication-Efficient and Generalization-Enhanced Federated Learning through Symbolic Model Updates (ICML 2025 · similarity 0.88)
- Fine-Tuning Personalization in Federated Learning to Mitigate Adversarial Clients (NeurIPS 2024 · 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.
