Research map / Federated learning and privacy / Personalized and heterogeneous FL
FedRA: A Random Allocation Strategy for Federated Tuning to Unleash the Power of Heterogeneous Clients
Shangchao Su, Bin Li, Xiangyang Xue
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 |
|---|---|---|---|
| replace fixed-prefix depth allocation (InclusiveFL) with random layer allocation (FedRA) | heterogeneous federated tuning · DomainNet feature-skew (ViT) | average accuracy (%): 55.97 -> 63.90 | Improved |
| replace fixed-prefix allocation with random allocation | heterogeneous federated tuning · NICO++ feature-skew (MLP-Mixer) | average accuracy (%): 72.28 -> 77.57 | Improved |
| random allocation under 30 clients with label skew | FedRA · NICO++ feature&label-skew (ViT) | average accuracy (%): 83.03 -> 85.44 | Improved |
| use random allocation when all clients are smaller than the server model | FedRA · NICO++ (ViT, clients with 6 layers) | average accuracy (%): 20.58 -> 70.38 | Improved |
| random allocation under dynamic heterogeneity | FedRA · NICO++ dynamic heterogeneity (ViT) | average accuracy (%): 88.73 -> 89.74 | Improved |
| use constrained allocation (each layer to at least one client) instead of previous-round substitution (ablation) | FedRA · NICO++ (ViT, clients [6]*6) | average accuracy (%): 70.38 -> 67.62 | Hurt |
| constrained allocation versus previous-round substitution with 10-layer clients (ablation) | FedRA · NICO++ (ViT, clients [10]*6) | average accuracy (%): 90.53 -> 90.54 | No clear effect |
| change LoRA insertion position (both, l0 only, l1 only) (ablation) | FedRA · NICO++ (ViT) | average accuracy (%): 86.48 / 86.38 / 85.65 | No clear effect |
| apply InclusiveFL/DepthFL versus ceiling on largest clients | InclusiveFL, DepthFL · DomainNet feature-skew (ViT, clipart 12 layers) | accuracy (%): 85.01 -> 81.13 | Hurt |
| FedRA versus InclusiveFL on grass domain quickdraw (hard domain) | FedRA · DomainNet feature-skew (ViT, quickdraw) | accuracy (%): 73.69 (ceiling) vs 25.41 (FedRA) | Mixed |
Later papers that compared with this method
2 later accepted papers report FedRA as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- FedBRICK: Structural Bias Aware Heterogeneous Foundation Model Federated Tuning (AAAI 2026 · better on 9 of 9 shared results)
- EFTViT: Efficient Federated Training of Vision Transformers with Masked Images on Resource-Constrained Clients (ICCV 2025 · better on 13 of 13 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
- FedPop: Federated Population-based Hyperparameter Tuning (AAAI 2025 · similarity 0.88)
- Adaptive LoRA Experts Allocation and Selection for Federated Fine-Tuning (NeurIPS 2025 · similarity 0.88)
- FedSMU: Communication-Efficient and Generalization-Enhanced Federated Learning through Symbolic Model Updates (ICML 2025 · similarity 0.88)
- FedAdaVR: Adaptive Variance Reduction for Robust Federated Learning under Limited Client Participation (NeurIPS 2026 · similarity 0.88)
- Efficient Adaptive Federated Optimization (NeurIPS 2025 · similarity 0.88)
- F^3OCUS - Federated Finetuning of Vision-Language Foundation Models with Optimal Client Layer Updating Strategy via Multi-objective Meta-Heuristics (CVPR 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.
