Research map / Federated learning and privacy / Personalized and heterogeneous FL
FedL2P: Federated Learning to Personalize
Royson Lee, Minyoung Kim, Da Li, Xinchi Qiu, Timothy Hospedales, Ferenc Huszár, Nicholas Donald Lane
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 FedL2P's learned client-specific personalization strategy instead of fixed client BN fine-tuning | CIFAR-10, FedAvg pretrained model, α=0.1, e=15 · CIFAR-10, α=0.1 | accuracy: 79.15±0.07 -> 80.28±0.07 | Improved |
| Use FedL2P's learned personalization strategy instead of client BN fine-tuning | CIFAR-10-C, α=0.1 · CIFAR-10-C, α=0.1 | accuracy: 87.25±0.06 -> 89.23±0.15 | Improved |
| Use FedL2P's learned personalization strategy instead of incoming-batch BN statistics | Office-Caltech-10, α=1000 · Office-Caltech-10, α=1000 | accuracy: 81.43±2.16 -> 88.85±0.89 | Improved |
| Use FedL2P instead of client BN fine-tuning on unseen speakers | Speech Commands, unseen clients, e=5 · Speech Commands, unseen clients | accuracy: 62.76±1.83 -> 87.85±0.18 | Improved |
| Add LRNet to BNNet-only personalization (ablation) | CIFAR-10, α=1000, e=15 · CIFAR-10, α=1000 | accuracy: 62.35±0.24 -> 65.13±0.02 | Improved |
| Add BNNet to LRNet with β fixed at 1 (ablation) | CIFAR-10-C, α=1000, e=15 · CIFAR-10-C, α=1000 | accuracy: 60.09±0.02 -> 59.97±0.22 | No clear effect |
| Fix β at 0 instead of learning client BN mixing (ablation) | DomainNet, α=1.0, e=15 · DomainNet, α=1.0 | accuracy: 63.77±0.44 -> diverge | Hurt |
| Replace IFT hypergradients with FOMAML+ (ablation) | FedL2P on DomainNet, α=1000 · DomainNet, α=1000 | accuracy: 54.38±0.45 -> 52.70±0.17 | Hurt |
| Use local client HPO rather than federatedly learned meta-nets | CIFAR-10, α=1000, e=15 · CIFAR-10, α=1000 | accuracy: 58.41±0.38 -> 65.13±0.02 | Hurt |
Later papers that compared with this method
1 later accepted paper reports FedL2P as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- FedP²EFT: Federated Learning to Personalize PEFT for Multilingual LLMs (AAAI 2026 · better on 19 of 19 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
- PeFLL: Personalized Federated Learning by Learning to Learn (ICLR 2024 · similarity 0.90)
- Decentralized Directed Collaboration for Personalized Federated Learning (CVPR 2024 · similarity 0.89)
- FedAS: Bridging Inconsistency in Personalized Federated Learning (CVPR 2024 · similarity 0.88)
- Understanding the Statistical Accuracy-Communication Trade-off in Personalized Federated Learning with Minimax Guarantees (ICML 2025 · similarity 0.88)
- Personalized Federated Learning via Feature Distribution Adaptation (NeurIPS 2024 · 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.
