Research map / Federated learning and privacy / Personalized and heterogeneous FL
Trustworthy Federated Label Distribution Learning under Annotation Quality Disparity
Junxiang Wu, Zhiqiang Kou, Hongwei Zeng, Wenke Huang, Biao Liu, Hanlin Gu, Yuheng Jia, Di Jiang, Yang Liu, Xin Geng
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 client-side global-semantic-anchor rectifier to FedAvg (ablation) | FedAvg · FER-LDL | KL divergence: 0.342 -> 0.264 | Improved |
| add client-side anchor rectifier to FedAvg (ablation) | FedAvg · PIPAL-LDL | KL divergence: 0.252 -> 0.154 | Improved |
| add server-side reliability-aware aggregation on top of anchor rectifier (ablation) | FedQual (+A) · PIPAL-LDL | KL divergence: 0.154 -> 0.116 | Improved |
| add server-side reliability-aware aggregation on top of anchor rectifier (ablation) | FedQual (+A) · FER-LDL | KL divergence: 0.264 -> 0.253 | Improved |
| add anchor rectifier only (without aggregation change) on PIPAL-LDL (ablation) | FedAvg · PIPAL-LDL | Clark distance: 1.398 -> 1.399 | No clear effect |
| use full FedQual instead of FedAvg (ablation) | FedAvg · PIPAL-LDL | Cosine similarity: 0.929 -> 0.958 | Improved |
| replace sample-size weighting with quality-weighted aggregation alone (FedQAgg) | FedAvg · PIPAL-LDL | KL divergence: 0.2521 -> 0.4398 | Hurt |
| replace sample-size weighting with quality-weighted aggregation alone (FedQAgg) | FedAvg · FER-LDL | KL divergence: 0.3420 -> 0.2756 | Improved |
| replace sample-size weighting with quality-weighted aggregation alone (FedQAgg) | FedAvg · FI-LDL | KL divergence: 0.4233 -> 0.5451 | Hurt |
| add FedProx proximal term to FedAvg | FedAvg · FI-LDL | KL divergence: 0.4233 -> 0.5119 | Hurt |
| add MOON contrastive alignment to FedAvg | FedAvg · KADID-LDL | KL divergence: 0.1248 -> 0.4262 | Hurt |
| increase fraction of noisy clients to 75% (ablation) | FedQual · four Fed-LDL benchmarks | all six LDL metrics | No clear effect |
Closest papers by meaning
- (FL)$^2$: Overcoming Few Labels in Federated Semi-Supervised Learning (NeurIPS 2024 · similarity 0.89)
- FedFixer: Mitigating Heterogeneous Label Noise in Federated Learning (AAAI 2024 · similarity 0.89)
- Progressive Distribution Matching for Federated Semi-Supervised Learning (AAAI 2025 · similarity 0.87)
- FedClean: A General Robust Label Noise Correction for Federated Learning (ICML 2025 · similarity 0.87)
- Exploring Vacant Classes in Label-Skewed Federated Learning (AAAI 2025 · similarity 0.87)
- FedDiv: Collaborative Noise Filtering for Federated Learning with Noisy Labels (AAAI 2024 · similarity 0.87)
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.
