Research map / Federated learning and privacy / Personalized and heterogeneous FL
DaFKD: Domain-Aware Federated Knowledge Distillation
Haozhao Wang, Yichen Li, Wenchao Xu, Ruixuan Li, Yufeng Zhan, Zhigang Zeng
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 domain-correlation weighting of local model predictions (ablation) | DaFKD · MNIST (α = 0.05, 0.1), SVHN (α = 0.05, 0.1), FASHION MNIST (α = 0.05, 0.1), EMNIST (α = 0.05, 0.1) | Test accuracy (%): 80.57±1.67 -> 82.33±0.44; 94.14± 0.85 -> 95.56±0.41; 50.33±2.08 -> 51.14±0.16; 72.51±0.98 -> 72.80±0.11; 34.64±1.68 -> 3 | Improved |
| Add parameter sharing between discriminator and classifier (ablation) | DaFKD · MNIST (α = 0.05, 0.1), SVHN (α = 0.05, 0.1), FASHION MNIST (α = 0.05, 0.1), EMNIST (α = 0.05, 0.1) | Test accuracy (%): 84.67±0.92 -> 82.33±0.44; 95.23±0.16 -> 95.56±0.41; 50.78±0.03 -> 51.14±0.16; 74.01±1.08 -> 72.80±0.11; 35.45±0.34 -> 37 | Mixed |
| Use domain-correlation weighting of local predictions (ablation) | DaFKD · MNIST, SVHN, FASHION MNIST, EMNIST (α = 0.1) | Communication rounds to target accuracy: 45.33±2.67 -> 38.33±1.67; 31.00±3.67 -> 18.67±1.33; 37.67±1.67 -> 33.67±3.00; 42.33±0.33 -> 40.67±4.33 | Improved |
| Use DaFKD instead of the strongest listed baseline | ResNet11 · SVHN, α = 0.1 | Top-1 test accuracy (%): 66.78±0.33 -> 72.80±0.11 | Improved |
| Use DaFKD instead of the strongest listed baseline | ResNet11 · FASHION MNIST, α = 0.05; EMNIST, α = 0.05 | Top-1 test accuracy (%): 39.71±0.23 -> 37.85±0.24; 69.73±0.17 -> 67.64±1.86 | Hurt |
Later papers that compared with this method
2 later accepted papers report DaFKD as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- FedCD: Towards Consolidated Distillation for Heterogeneous Federated Learning (AAAI 2026 · better on 3 of 3 shared results)
- Feature Distillation is the Better Choice for Model-Heterogeneous Federated Learning (NeurIPS 2025 · better on 4 of 4 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
- Efficient Federated Learning via Clients-to-Server Knowledge Distillation (Student Abstract) (AAAI 2025 · similarity 0.91)
- HYDRA-FL: Hybrid Knowledge Distillation for Robust and Accurate Federated Learning (NeurIPS 2024 · similarity 0.90)
- Rethinking the Temperature for Federated Heterogeneous Distillation (ICML 2025 · similarity 0.89)
- Feature Distillation is the Better Choice for Model-Heterogeneous Federated Learning (NeurIPS 2025 · similarity 0.89)
- FedGMKD: An Efficient Prototype Federated Learning Framework through Knowledge Distillation and Discrepancy-Aware Aggregation (NeurIPS 2024 · similarity 0.88)
- One-shot Federated Learning via Synthetic Distiller-Distillate Communication (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.
