Research map / Federated learning and privacy / Personalized and heterogeneous FL
Robust Heterogeneous Federated Learning under Data Corruption
Xiuwen Fang, Mang Ye, Xiyuan Yang
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 federated communication (HFL) under data corruption (ablation) | local models · CIFAR-10-C xi=0.5, corrupted test | avg accuracy: 58.01 -> 56.08 | Hurt |
| add HFL on clean data (ablation) | local models · CIFAR-10-C xi=0, clean test | avg accuracy: 77.98 -> 78.61 | Improved |
| add PubAug re-weighted communication (ablation) | HFL · CIFAR-10-C xi=0, corrupted test | avg accuracy: 63.45 -> 64.88 | Improved |
| add PubAug when all clients fully corrupted (ablation) | HFL · CIFAR-10-C xi=1, corrupted test | avg accuracy: 57.71 -> 56.45 | No clear effect |
| add corruption-robust augmentation (Aug) without JSD (ablation) | HFL + PubAug · CIFAR-10-C xi=0, clean test | avg accuracy: 78.79 -> 72.10 | Hurt |
| add JSD consistency loss on top of Aug (ablation) | AugHFL · CIFAR-10-C xi=0.5, corrupted test | avg accuracy: 57.44 -> 68.52 | Improved |
| add JSD on top of Aug (ablation) | AugHFL · CIFAR-10-C xi=0.5, clean test | avg accuracy: 66.19 -> 73.17 | Improved |
| use AugHFL instead of prior heterogeneous FL methods when training data heavily corrupted | FedMD/FedDF/RHFL/FCCL · CIFAR-10-C xi=1, corrupted test | avg accuracy: 57.91 -> 70.03 vs FCCL | Improved |
| use AugHFL in fully clean setting | AugHFL vs FCCL · CIFAR-10-C xi=0, clean test | avg accuracy: 80.39 -> 77.86 | Hurt |
| increase private-data corruption rate from 0 to 1 | AugHFL · CIFAR-10-C | avg accuracy: 4.2% on clean, 1.18% on corrupted | Hurt |
Closest papers by meaning
- Scalable and Trustworthy Learning in Heterogeneous Networks (AAAI 2025 · similarity 0.87)
- Fake It Till Make It: Federated Learning with Consensus-Oriented Generation (ICLR 2024 · similarity 0.86)
- Tackling Intertwined Data and Device Heterogeneities in Federated Learning with Unlimited Staleness (AAAI 2025 · similarity 0.86)
- FedGPS: Statistical Rectification Against Data Heterogeneity in Federated Learning (NeurIPS 2025 · similarity 0.85)
- Federated Learning with Sample-level Client Drift Mitigation (AAAI 2025 · similarity 0.85)
- Noise-Aware Algorithm for Heterogeneous Differentially Private Federated Learning (ICML 2024 · similarity 0.85)
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.
