Research map / Federated learning and privacy / Personalized and heterogeneous FL
Not All Clients Are Equal: Collaborative Model Personalization on Heterogeneous Multi-Modal Clients
Minhyuk Seo, Taeheon Kim, Hankook Lee, Jonghyun Choi, Tinne Tuytelaars
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 Co-LoRA (shareable P,Q modules) over vanilla homogeneous averaging (ablation) | FedMosaic · DRAKE-Dynamic Others | Alast: 47.63 -> 51.33 | Improved |
| add RELA relevance-guided aggregation on top of Co-LoRA (ablation) | FedMosaic · DRAKE-Dynamic Self | Alast: 66.99 -> 67.86 | Improved |
| add RELA on top of Co-LoRA (ablation) | FedMosaic · Fed-Scope Self | Alast: 28.47 -> 30.58 | Improved |
| add RELA on top of Co-LoRA (ablation) | FedMosaic · DRAKE-Dynamic Others | Alast: 51.33 -> 51.16 | No clear effect |
| use FedMosaic instead of local-only SFT | FedMosaic · DRAKE-Dynamic Self | Alast: 65.79 -> 67.86 | Improved |
| use standard PFL methods (DITTO, FedDAT, PerAda) instead of local SFT | PFL baselines · DRAKE-Dynamic Self | Alast: 65.79 -> 58.47 (FedDAT) | Hurt |
| use FedMosaic instead of SFT on HFLB (single-image VQA) | FedMosaic · HFLB-Dynamic Self | Alast: 79.99 -> 80.80 | Improved |
| use FedMosaic across cross-family Llama/Qwen clients | FedMosaic · DRAKE-Dynamic cross-family Self | Alast: 68.60 -> 70.67 | Improved |
| use FedMosaic with 52 heterogeneous LLM clients | FedMosaic · Fed-LLM-Large Self | Alast: 18.03 -> 20.87 | Improved |
| use FedMosaic for complex multi-image task clients (3B models) | FedMosaic · DRAKE-Dynamic client 5 Self | Alast: 62.00 -> 68.87 | Improved |
| use FedMosaic for simple single-image VQA clients | FedMosaic · DRAKE-Dynamic client 3 Self | Alast: 66.63 -> 66.71 | No clear effect |
| vary ratio of 1B and 3B clients (ablation) | FedMosaic · DRAKE-Dynamic Self | Alast: +1.45 to +2.68 over SFT | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- FedDAT — FedDAT: An Approach for Foundation Model Finetuning in Multi-Modal Heterogeneous Federated Learning (AAAI 2024) · better on 5 of 5 shared results
- PerAda — PerAda: Parameter-Efficient Federated Learning Personalization with Generalization Guarantees (CVPR 2024) · better on 1 of 1 shared result
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
- FedAS: Bridging Inconsistency in Personalized Federated Learning (CVPR 2024 · similarity 0.88)
- pFedClub: Controllable Heterogeneous Model Aggregation for Personalized Federated Learning (NeurIPS 2024 · similarity 0.88)
- Fine-Tuning Personalization in Federated Learning to Mitigate Adversarial Clients (NeurIPS 2024 · similarity 0.88)
- Federated Learning with Heterogeneous LLMs: Integrating Small Student Client Models with a Large Hungry Model (AAAI 2025 · similarity 0.87)
- A Bayesian Approach for Personalized Federated Learning in Heterogeneous Settings (NeurIPS 2024 · similarity 0.87)
- Understanding the Statistical Accuracy-Communication Trade-off in Personalized Federated Learning with Minimax Guarantees (ICML 2025 · 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.
