Research map / Federated learning and privacy / Personalized and heterogeneous FL
Spectral Co-Distillation for Personalized Federated Learning
Zihan Chen, Howard Hao Yang, Tony Quek, Kai Fong Ernest Chong
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 |
|---|---|---|---|
| Remove spectral distillation from generic model training (SCD-GM) (ablation) | PFL+ (Ours) · CIFAR-10 alpha=0.1 | generic model accuracy: 75.52 $\pm$ 0.11 -> 73.51 $\pm$ 0.17 | Improved |
| Remove SCD-GM (ablation) | PFL+ (Ours) · CIFAR-10 alpha=0.1 | personalized model accuracy: 82.69 $\pm$ 0.16 -> 81.03 $\pm$ 0.20 | Improved |
| Remove SCD-GM (ablation) | PFL+ (Ours) · CIFAR-10 alpha=1 | generic model accuracy: 86.37 $\pm$ 0.15 -> 85.35 $\pm$ 0.11 | Improved |
| Remove SCD-PM (personalized model trained locally without distillation) (ablation) | PFL+ (Ours) · CIFAR-10 alpha=0.1 | personalized model accuracy: 82.69 $\pm$ 0.16 -> 70.51 $\pm$ 1.21 | Improved |
| Remove SCD-PM (ablation) | PFL+ (Ours) · CIFAR-10 alpha=0.1 | generic model accuracy: 75.52 $\pm$ 0.11 -> 68.96 $\pm$ 0.47 | Improved |
| Remove SCD-PM (ablation) | PFL+ (Ours) · CIFAR-10 alpha=1 | personalized model accuracy: 92.25 $\pm$ 0.21 -> 79.65 $\pm$ 0.83 | Improved |
| Remove both SCD-GM and SCD-PM (equals FedAvg for GM) (ablation) | PFL+ (Ours) · CIFAR-10 alpha=0.1 | personalized model accuracy: 82.69 $\pm$ 0.16 -> 70.51 $\pm$ 1.21 | Improved |
| Add wait-free training protocol (5 local epochs, target 40%) | Ditto · CIFAR-10 alpha=0.1 | communication speedup: 2.87$\times$ | Improved |
| Add wait-free training protocol (3 local epochs, target 80%) | Ditto · CIFAR-10 alpha=0.1 | communication speedup: 1.38 $\times$ | Improved |
| Add wait-free training protocol (5 local epochs, target 80%) | FedRoD · CIFAR-10 alpha=0.1 | communication speedup: 2.19$\times$ | Improved |
| Add wait-free training protocol (3 local epochs, 40%) | PFL+ (Ours) · CIFAR-10 alpha=0.1 | communication speedup: 1.82 $\times$ | Improved |
| Use spectral co-distillation instead of FedRoD | FedRoD · CIFAR-100 alpha=1 | personalized model accuracy: 62.51 $\pm$ 0.15 -> 63.15 $\pm$ 0.16 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FedBABU — FedBABU: Toward Enhanced Representation for Federated Image Classification (ICLR 2022) · better on 6 of 6 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 0.88)
- pFedClub: Controllable Heterogeneous Model Aggregation for Personalized Federated Learning (NeurIPS 2024 · similarity 0.88)
- One-shot Federated Learning via Synthetic Distiller-Distillate Communication (NeurIPS 2024 · similarity 0.88)
- Efficient Federated Learning via Clients-to-Server Knowledge Distillation (Student Abstract) (AAAI 2025 · similarity 0.88)
- Understanding the Statistical Accuracy-Communication Trade-off in Personalized Federated Learning with Minimax Guarantees (ICML 2025 · similarity 0.88)
- FedAS: Bridging Inconsistency in Personalized Federated Learning (CVPR 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.
