Research map / Federated learning and privacy / Privacy attacks and secure learning
Fed-FA: Theoretically Modeling Client Data Divergence for Federated Language Backdoor Defense
Zhiyuan Zhang, Deli Chen, Hao Zhou, Fandong Meng, Jie Zhou, Xu Sun
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 |
|---|---|---|---|
| use f-divergence indicator instead of Euclidean indicator (ablation) | Fed-FA · Average over models, datasets, attacks | ASR (%): 58.50 -> 19.51 | Improved |
| use f-divergence indicator instead of Euclidean indicator, correlation with divergence (ablation) | Fed-FA · dataset mixing ratio trials | correlation: 0.9045 -> 0.9847 | Improved |
| add embedding Hessian reassignment (ablation) | Fed-FA · Average over models, datasets, attacks | ASR (%): 39.78 -> 19.51 | Improved |
| use synthesized dataset instead of labeled dataset for Hessian scales (ablation) | Fed-FA · Average over models, datasets, attacks | ASR (%): 20.06 -> 19.51 | No clear effect |
| use proportional reassignment instead of inverse reassignment (ablation) | Fed-FA · Average over models, datasets, attacks | ASR (%): 82.27 -> 19.51 | Improved |
| reassign Hessians only on embeddings instead of layer-wise (ablation) | Fed-FA · Average over models, datasets, attacks | ASR (%): 52.75 -> 19.51 | Improved |
| reassign Hessians only on embeddings instead of entire model (ablation) | Fed-FA · Average over models, datasets, attacks | ASR (%): 42.03 -> 19.51 | Improved |
| Fed-FA versus Dim-Krum aggregation | Fed-FA · 3 models average | ASR (%): 31.56 -> 19.51 | Improved |
| Fed-FA versus Dim-Krum on AgNews | Fed-FA · AgNews | ASR (%): 4.00 -> 3.43 | Improved |
| Fed-FA versus Dim-Krum on EP attack | Fed-FA · EP attack | ASR (%): 14.09 -> 12.13 | No clear effect |
| Fed-FA versus Dim-Krum on BadSent attack | Fed-FA · BadSent attack | ASR (%): 48.42 -> 28.30 | Improved |
| Fed-FA versus FedAvg clean accuracy | Fed-FA · 3 models average | clean ACC (%): 85.28 -> 85.70 | No clear effect |
Closest papers by meaning
- Defending against Backdoors in Federated Learning with Robust Learning Rate (AAAI 2021 · similarity 0.89)
- MARS: A Malignity-Aware Backdoor Defense in Federated Learning (NeurIPS 2025 · similarity 0.87)
- Resisting Backdoor Attacks in Federated Learning via Bidirectional Elections and Individual Perspective (AAAI 2024 · similarity 0.87)
- Parameter Disparities Dissection for Backdoor Defense in Heterogeneous Federated Learning (NeurIPS 2024 · similarity 0.87)
- Label-Free Backdoor Attacks in Vertical Federated Learning (AAAI 2025 · similarity 0.87)
- Infighting in the Dark: Multi-Label Backdoor Attack in Federated Learning (CVPR 2025 · 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.
