Research map / Federated learning and privacy / Differential privacy
Taming Noise-Induced Prototype Degradation for Privacy-Preserving Personalized Federated Fine-Tuning
Yuhua Wang, Qinnan Zhang, Xiaodong Li, Huan Zhang, Yifan Sun, Wangjie Qiu, Hainan Zhang, Yongxin Tong, Zhiming Zheng
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 VPP (variance-adaptive prototype perturbation) to IGPP (ablation) | FedProto · Office-Caltech | average accuracy (AVG): 90.82 -> 93.06 | Improved |
| add DCR (distillation-guided clipping regularization) to IGPP (ablation) | FedProto · Office-Caltech | average accuracy (AVG): 90.82 -> 92.89 | Improved |
| add VPP to IGPP (ablation) | FedProto · PACS | average accuracy (AVG): 90.58 -> 91.89 | Improved |
| add DCR to IGPP (ablation) | FedProto · PACS | average accuracy (AVG): 90.58 -> 91.91 | Improved |
| add DCR on top of VPP (full VPDR) (ablation) | FedProto · PACS | average accuracy (AVG): 91.89 -> 92.71 | Improved |
| add DCR on top of VPP (full VPDR) (ablation) | FedProto · Office-Caltech | average accuracy (AVG): 93.06 -> 93.36 | Improved |
| replace IGPP with VPDR | FedPCL · PACS (domain skew, eps=1) | average accuracy (AVG): 79.73 -> 86.91 | Improved |
| replace IGPP with VPDR | FedProto · Digits (domain skew, eps=1) | average accuracy (AVG): 94.36 -> 96.05 | Improved |
| replace IGPP with VPDR under extreme label skew (alpha=0.1) | FPL · CIFAR-10 label skew | average accuracy (AVG): 28.42 -> 37.18 | Improved |
| replace IGPP with VPDR under extreme label skew (alpha=0.1) | FedProto · CIFAR-10 label skew | average accuracy (AVG): 42.76 -> 44.14 | Improved |
| replace IGPP with VPDR at alpha=1 | FedPLVM · CIFAR-10 label skew | average accuracy (AVG): 87.21 -> 82.76 | Hurt |
| replace IGPP with VPDR at near-balanced labels (alpha=10) | FedProto · CIFAR-10 label skew | average accuracy (AVG): 96.30 -> 96.55 | Improved |
Closest papers by meaning
- FedPKDA: Personalized Federated Learning with Privacy-Preserving Knowledge Dynamic Alignment (AAAI 2026 · similarity 0.89)
- Privacy-Preserving Federated Convex Optimization: Balancing Partial-Participation and Efficiency via Noise Cancellation (ICML 2025 · similarity 0.87)
- Taming Cross-Domain Representation Variance in Federated Prototype Learning with Heterogeneous Data Domains (NeurIPS 2024 · similarity 0.86)
- Noise-Aware Algorithm for Heterogeneous Differentially Private Federated Learning (ICML 2024 · similarity 0.86)
- Differentially Private Federated Low Rank Adaptation Beyond Fixed-Matrix (NeurIPS 2025 · similarity 0.86)
- Clients Collaborate: Flexible Differentially Private Federated Learning with Guaranteed Improvement of Utility-Privacy Trade-off (ICML 2025 · similarity 0.86)
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.
