Research map / Federated learning and privacy / Personalized and heterogeneous FL
Fine-Tuning Impairs the Balancedness of Foundation Models in Long-tailed Personalized Federated Learning
Shihao Hou, Chikai Shang, Zhiheng Yang, Jiacheng Yang, Xinyi Shang, Junlong Gao, Yiqun Zhang, Yang Lu
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 gradient purification (ablation) | prompt-based FL baseline · CIFAR-100-LT (IF=100) | global accuracy All: 61.98 -> 69.77 | Improved |
| add gradient purification (ablation) | prompt-based FL baseline · CIFAR-100-LT (IF=100) | global Few-class accuracy: 42.32 -> 64.74 | Improved |
| add gradient purification (ablation) | prompt-based FL baseline · CIFAR-100-LT (IF=100) | global Many-class accuracy: 80.23 -> 78.63 | Hurt |
| add gradient purification (ablation) | prompt-based FL baseline · Places-LT | global accuracy All: 38.90 -> 43.88 | Improved |
| use gradient purification instead of LS-KD (ablation) | FedPuReL global training · CIFAR-100-LT (IF=100) | global accuracy All: 66.43 -> 69.77 | Improved |
| use gradient purification instead of WiSE-FT (ablation) | FedPuReL global training · CIFAR-100-LT (IF=100) | global accuracy All: 63.64 -> 69.77 | Improved |
| add residual learning (ablation) | prompt-based FL baseline · CIFAR-100-LT (IF=100) | personalized accuracy All: 70.52 -> 72.25 | Improved |
| add residual learning (ablation) | prompt-based FL baseline · Places-LT | personalized accuracy All: 36.49 -> 39.34 | Improved |
| add gradient purification on top of residual learning (ablation) | FedPuReL · CIFAR-100-LT (IF=100) | personalized accuracy All: 72.25 -> 73.37 | Improved |
| use temperature-aligned KL instead of standard KL (ablation) | FedPuReL gradient purification · CIFAR-100-LT (IF=100) | global accuracy: 67.92 -> 69.77 | Improved |
| use temperature-aligned KL instead of standard KL (ablation) | FedPuReL gradient purification · CIFAR-100-LT (IF=100) | global balancedness: 25.03 -> 27.87 | Improved |
| use temperature-aligned KL instead of standard KL (ablation) | FedPuReL gradient purification · CIFAR-100-LT (IF=100) | personalized accuracy: 71.62 -> 73.37 | Improved |
Closest papers by meaning
- FedLoGe: Joint Local and Generic Federated Learning under Long-tailed Data (ICLR 2024 · similarity 0.91)
- Tackling Feature-Classifier Mismatch in Federated Learning via Prompt-Driven Feature Transformation (NeurIPS 2025 · similarity 0.90)
- Understanding the Statistical Accuracy-Communication Trade-off in Personalized Federated Learning with Minimax Guarantees (ICML 2025 · similarity 0.90)
- Flexible Sharpness-Aware Personalized Federated Learning (AAAI 2025 · similarity 0.90)
- FedAS: Bridging Inconsistency in Personalized Federated Learning (CVPR 2024 · similarity 0.89)
- Fine-Tuning Personalization in Federated Learning to Mitigate Adversarial Clients (NeurIPS 2024 · similarity 0.89)
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.
