Research map / Federated learning and privacy / Personalized and heterogeneous FL
AnaPFL: When Closed-Form Solutions Meet Generalizationand Personalization in Personalized Federated Learning
Kejia Fan, Jianheng Tang, Zhirui Yang, Feijiang Han, Yajiang Huang, Run He, Jiaxu Li, Songning Lai, Anfeng Liu, Houbing Herbert Song, Yunhuai Liu, HUIPING ZHUANG
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-stream (primary + refinement) analytic personalization to AFL | AFL · CIFAR-100, 50 clients, alpha=0.1 | average accuracy: 56.63% -> 80.64% | Improved |
| add refinement stream for local personalization to AFL | AFL · CIFAR-100, 50 clients, alpha=1.0 | average accuracy: 56.66% -> 62.63% | Improved |
| add refinement stream to AFL on ImageNet-R | AFL · ImageNet-R, 50 clients, alpha=0.1 | average accuracy: 28.87% -> 46.23% | Improved |
| analytic AnaPFL instead of gradient-based FedSelect | FedSelect · ImageNet-R, 50 clients, alpha=0.1 | average accuracy: 34.83% -> 46.23% | Improved |
| increase clients from 50 to 100 | AnaPFL · CIFAR-100, alpha=0.1 | average accuracy: 80.64% -> 79.22% | Hurt |
| lower heterogeneity alpha from 0.1 to 1.0 | AnaPFL · CIFAR-100, 50 clients | average accuracy: 80.64% -> 62.63% | Hurt |
| add local fine-tuning (personalization) to FedAvg under high heterogeneity | FedAvg · CIFAR-100, 50 clients, alpha=0.1 | average accuracy: 43.53% -> 58.81% | Improved |
| add local fine-tuning to FedAvg under low heterogeneity | FedAvg · CIFAR-100, 50 clients, alpha=1.0 | average accuracy: 48.91% -> 37.73% | Hurt |
| use analytic learning for aggregation instead of gradient-based updates in efficiency comparison | FedAvg and FedPCL · CIFAR-100, 50 clients, alpha=0.1 | computation time: 100-300 s vs 45,000-66,000 s | Improved |
| use analytic learning instead of gradient-based updates for communication | FedAvg and FedPCL · CIFAR-100, 50 clients, alpha=0.1 | communication overhead: 200-400 MB vs 20,000-40,000 MB | Improved |
| add personalization to AFL costs slight extra overhead | AFL · CIFAR-100 | efficiency: accuracy gaps of up to 17.36%-24.01% | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- FedALA — FedALA: Adaptive Local Aggregation for Personalized Federated Learning (AAAI 2023) · better on 3 of 3 shared results
- FedSelect — FedSelect: Personalized Federated Learning with Customized Selection of Parameters for Fine-Tuning (CVPR 2024) · better on 2 of 2 shared results
- Ditto — Ditto: Fair and Robust Federated Learning Through Personalization (ICML 2021) · 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
- Decentralized Directed Collaboration for Personalized Federated Learning (CVPR 2024 · similarity 0.89)
- Understanding the Statistical Accuracy-Communication Trade-off in Personalized Federated Learning with Minimax Guarantees (ICML 2025 · similarity 0.88)
- PeFLL: Personalized Federated Learning by Learning to Learn (ICLR 2024 · similarity 0.88)
- FedAS: Bridging Inconsistency in Personalized Federated Learning (CVPR 2024 · similarity 0.88)
- FedSelect: Personalized Federated Learning with Customized Selection of Parameters for Fine-Tuning (CVPR 2024 · similarity 0.88)
- PerAda: Parameter-Efficient Federated Learning Personalization with Generalization Guarantees (CVPR 2024 · 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.
