Research map / Federated learning and privacy / Personalized and heterogeneous FL
DFRD: Data-Free Robustness Distillation for Heterogeneous Federated Learning
Kangyang Luo, Shuai Wang, Yexuan Fu, Xiang Li, Yunshi Lan, Ming Gao
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 |
|---|---|---|---|
| Fine-tune FedAvg with DFRD | FedAvg · FMNIST, ω=1.0 | G.acc: 89.71{ $\pm$0.10} -> 91.65{ $\pm$0.09} | Improved |
| Fine-tune FedAvg with DFRD | FedAvg · SVHN, ω=1.0 | G.acc: 70.63{ $\pm$1.95} -> 77.84{ $\pm$1.39} | Improved |
| Fine-tune FedAvg with DFRD | FedAvg · CIFAR-10, ω=1.0 | G.acc: 67.16{ $\pm$0.98} -> 71.54{ $\pm$0.66} | Improved |
| Fine-tune FedAvg with DFRD | FedAvg · CIFAR-100, ω=1.0 | G.acc: 57.31{ $\pm$0.09} -> 61.49{ $\pm$0.28} | Improved |
| Fine-tune FedRolex with DFRD | FedRolex · SVHN, ρ=5 | G.acc: 34.71{ $\pm$13.68} -> 46.30{ $\pm$10.12} | Improved |
| Fine-tune FedRolex with DFRD | FedRolex · CIFAR-10, ρ=10 | G.acc: 16.57{ $\pm$4.40} -> 25.57{ $\pm$1.37} | Improved |
| Fine-tune FedRolex with DFRD | FedRolex · Tiny-ImageNet, ρ=10 | G.acc: 5.55{ $\pm$0.40} -> 6.80{ $\pm$0.11} | Improved |
| Fine-tune FedRolex with DFRD | FedRolex · FOOD101, ρ=10 | G.acc: 7.54{ $\pm$0.46} -> 10.58{ $\pm$0.29} | Improved |
| Add an EMA copy of the generator (ablation) | DFRD · SVHN, FedRolex+DFRD configuration | G.acc: 26.97{ $\pm$12.28} -> 34.78{ $\pm$9.19} | Improved |
| Add an EMA copy of the generator (ablation) | DFRD · CIFAR-10, FedRolex+DFRD configuration | G.acc: 19.80{ $\pm$2.25} -> 25.57{ $\pm$1.37} | Improved |
| Add the transferability loss (ablation) | DFRD · SVHN, FedRolex+DFRD configuration | G.acc: 29.30{ $\pm$9.25} -> 34.78{ $\pm$9.19} | Improved |
| Add the diversity loss (ablation) | DFRD · SVHN, FedRolex+DFRD configuration | G.acc: 27.68{ $\pm$9.75} -> 34.78{ $\pm$9.19} | Improved |
Later papers that compared with this method
1 later accepted paper reports DFRD as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Data-Free Black-Box Federated Learning via Zeroth-Order Gradient Estimation (AAAI 2025 · better on 8 of 8 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
- HYDRA-FL: Hybrid Knowledge Distillation for Robust and Accurate Federated Learning (NeurIPS 2024 · similarity 0.90)
- Federated Learning via Input-Output Collaborative Distillation (AAAI 2024 · similarity 0.89)
- Data-Free Black-Box Federated Learning via Zeroth-Order Gradient Estimation (AAAI 2025 · similarity 0.89)
- Overcoming Data and Model heterogeneities in Decentralized Federated Learning via Synthetic Anchors (ICML 2024 · similarity 0.89)
- Feature Distillation is the Better Choice for Model-Heterogeneous Federated Learning (NeurIPS 2025 · similarity 0.88)
- Complementary Knowledge Distillation for Robust and Privacy-Preserving Model Serving in Vertical Federated Learning (AAAI 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.
