Research map / Federated learning and privacy / Personalized and heterogeneous FL
Class-Wise Federated Averaging for Efficient Personalization
Gyuejeong Lee, Daeyoung Choi
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 WDR regularization to cwFedAvg with approximated distribution (ablation) | cwFedAvg · CIFAR-100 (practical, alpha=0.1, cwFedAvg on all layers) | accuracy: 32.25 -> 54.99 | Improved |
| add WDR regularization to cwFedAvg with approximated distribution (ablation) | cwFedAvg · Tiny ImageNet (practical, alpha=0.1, all layers) | accuracy: 20.08 -> 38.94 | Improved |
| use approximated distribution with WDR instead of empirical distribution (ablation) | cwFedAvg · CIFAR-100 (practical, alpha=0.1, cwFedAvg on all layers) | accuracy: 48.27 -> 54.99 | Improved |
| use approximated distribution without WDR instead of FedAvg (ablation) | cwFedAvg · CIFAR-100 (practical, alpha=0.1, cwFedAvg on all layers) | accuracy: 32.44 -> 32.25 | No clear effect |
| apply cwFedAvg only to output layer instead of all layers (ablation) | cwFedAvg · CIFAR-100 (practical, alpha=0.1, layer-selective) | accuracy: 54.99 -> 56.29 | Improved |
| apply cwFedAvg only to output layer instead of all layers (ablation) | cwFedAvg · Tiny ImageNet (practical, alpha=0.1, layer-selective) | accuracy: 38.94 -> 41.38 | Improved |
| apply cwFedAvg only to output layer instead of all layers (ablation) | cwFedAvg · CIFAR-100 (pathological, layer-selective) | accuracy: 65.91 -> 67.50 | Improved |
| use cwFedAvg instead of FedAvg (ablation) | cwFedAvg · CIFAR-100 (pathological) | accuracy: 28.22 -> 67.50 | Improved |
| use cwFedAvg instead of FedAMP on CIFAR-10 practical | cwFedAvg · CIFAR-10 (practical, alpha=0.1) | accuracy: 89.46 -> 88.65 | Hurt |
| increase number of clients from 50 to 100 | cwFedAvg · CIFAR-100 (clients) | accuracy: 55.90 -> 53.54 | Hurt |
| increase alpha toward IID (1.0) | cwFedAvg vs FedAvg · CIFAR-100 (alpha=1.0, 20 clients) | accuracy: 37.50 vs 36.75 | No clear effect |
| use cwFedAvg instead of FedAMP at low heterogeneity alpha=0.5 | cwFedAvg · CIFAR-100 (alpha=0.5, 20 clients) | accuracy: 25.41 -> 40.78 | Improved |
Closest papers by meaning
- Tackling Data Heterogeneity in Federated Learning with Class Prototypes (AAAI 2023 · similarity 0.90)
- Personalized Federated Learning via Feature Distribution Adaptation (NeurIPS 2024 · similarity 0.89)
- Fine-Tuning Personalization in Federated Learning to Mitigate Adversarial Clients (NeurIPS 2024 · similarity 0.88)
- pFedMxF: Personalized Federated Class-Incremental Learning with Mixture of Frequency Aggregation (CVPR 2025 · similarity 0.88)
- Decentralized Directed Collaboration for 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)
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.
