Research map / Federated learning and privacy / Personalized and heterogeneous FL
Parameterized Knowledge Transfer for Personalized Federated Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 personalized soft-prediction aggregation with a learned knowledge coefficient matrix rather than the strongest listed heterogeneous baseline | KT-pFL · heterogeneous models across EMNIST, Fashion_MNIST, and CIFAR-10, both Non-IID settings | test accuracy (%): 84.65 vs 83.00; 76.30 vs 75.72; 75.48 vs 72.82; 75.60 vs 73.45; 43.82 vs 42.83; 41.04 vs 39.90 | Helped |
| use KT-pFL instead of the strongest listed personalized baseline, pFedHN | KT-pFL · homogeneous CNN, EMNIST, Fashion_MNIST, and CIFAR-10 with 20 or 100 clients | test accuracy (%): 94.40 vs 94.27; 92.50 vs 92.21; 93.93 vs 92.22; 90.37 vs 89.14; 66.96 vs 65.31; 58.29 vs 57.38 | Helped |
| increase local epochs from 5 to 10 (ablation) | KT-pFL · EMNIST | test accuracy (%): 90.15 -> 92.76 | Helped |
| increase local epochs from 10 to 20 (ablation) | KT-pFL · EMNIST | test accuracy (%): 92.76 -> 90.15 | Hurt |
| increase distillation steps from 1 to 5 (ablation) | KT-pFL · EMNIST | test accuracy (%): 91.76 -> 91.54 | Hurt |
| increase distillation steps from 1 to 2 (ablation) | KT-pFL · Fashion_MNIST | test accuracy (%): 89.14 -> 89.90 | Helped |
| vary distillation steps from 1 to 5 (ablation) | KT-pFL · CIFAR-10 | test accuracy (%): 59.24 -> 58.91 | No clear effect |
| increase regularization ρ from 0.1 to 0.7 (ablation) | KT-pFL · EMNIST | test accuracy (%): 90.96 -> 91.76 | Helped |
| increase regularization ρ from 0.1 to 0.7 (ablation) | KT-pFL · Fashion_MNIST and CIFAR-10 | test accuracy (%): 89.49 -> 89.14; 59.08 -> 58.40 | Mixed |
| replace MNIST public data with EMNIST public data (ablation) | KT-pFL · Fashion_MNIST | test accuracy (%): 89.14 (± 0.15) -> 88.76 (± 0.21) | No clear effect |
| increase public data size from 10 to 3000 (ablation) | KT-pFL · EMNIST and Fashion_MNIST | test accuracy (%): 56.19 -> 91.66; 50.80 -> 89.50 | Helped |
| increase public data size from 3000 to 5000 (ablation) | KT-pFL · EMNIST and Fashion_MNIST | test accuracy (%): 91.66 -> 91.32; 89.50 -> 89.14 | Hurt |
Closest papers by meaning
- Knowledge-Aware Parameter Coaching for Personalized Federated Learning (AAAI 2024 · similarity 0.91)
- Understanding the Statistical Accuracy-Communication Trade-off in Personalized Federated Learning with Minimax Guarantees (ICML 2025 · similarity 0.90)
- FedAS: Bridging Inconsistency in Personalized Federated Learning (CVPR 2024 · similarity 0.90)
- Harnessing Heterogeneous Statistical Strength for Personalized Federated Learning via Hierarchical Bayesian Inference (ICML 2025 · similarity 0.89)
- PerAda: Parameter-Efficient Federated Learning Personalization with Generalization Guarantees (CVPR 2024 · similarity 0.89)
- pFedClub: Controllable Heterogeneous Model Aggregation for Personalized Federated Learning (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.