Research map / Federated learning and privacy / Privacy attacks and secure learning
Fisher Calibration for Backdoor-Robust Heterogeneous Federated Learning
Wenke Huang, Mang Ye, zekun shi, Bo Du, Dacheng Tao
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 Fisher Difference Regularization and Fisher Discrepancy Aggregation together (ablation) | FedAvg under backdoor attack · Cifar-10 (beta=0.5, 30% malicious) | trade-off V: 48.84 -> 75.39 | Improved |
| add Fisher Difference Regularization alone (ablation) | SDFC · MNIST (beta=0.5, 30% malicious) | trade-off V: 49.93 -> 92.96 | Improved |
| add Fisher Discrepancy Aggregation alone (ablation) | SDFC · MNIST (beta=0.5, 30% malicious) | trade-off V: 49.93 -> 50.34 | No clear effect |
| add FDAgg on top of FDReg (ablation) | SDFC · Cifar-10 (beta=0.5, 30% malicious) | trade-off V: 64.42 -> 75.39 | Improved |
| use absolute difference of local and validation Fisher instead of either alone (ablation) | FDReg · Cifar-10 (beta=0.5, 30% malicious) | trade-off V: 58.45 / 48.33 -> 75.39 | Improved |
| use Fisher (squared gradient) importance instead of absolute gradient (ablation) | FDReg · Cifar-10 (beta=0.5, 30% malicious) | trade-off V: 58.36 -> 64.42 | Improved |
| increase regularization strength lambda (ablation) | FDReg · MNIST (beta=0.5, 30%/50% malicious) | trade-off V: 78.96/81.79 at lambda=1 -> 93.45/92.97 at larger lambda | Improved |
| reduce clean validation set from 256 to 16 samples (ablation) | SDFC · Cifar-10 (beta=0.5, 30% malicious) | trade-off V: 75.39 -> 75.60 | No clear effect |
| increase local epochs from 10 to 40 (ablation) | SDFC · MNIST (beta=0.5, 30% malicious) | trade-off V: 93.48 -> 89.14 | Hurt |
| use SDFC instead of strongest client-wise defenses at 50% malicious clients | backdoor-robust FL · MNIST (beta=0.5, 50% malicious) | trade-off V: 50.43 -> 76.85 | Improved |
| use SDFC instead of DnC | backdoor-robust FL · MNIST/Cifar-10 (beta=0.5, 30% malicious) | trade-off V: DnC 94.16 vs SDFC 92.98 (MNIST); 75.58 vs 75.39 (Cifar-10) | No clear effect |
Closest papers by meaning
- Parameter Disparities Dissection for Backdoor Defense in Heterogeneous Federated Learning (NeurIPS 2024 · similarity 0.93)
- Resisting Backdoor Attacks in Federated Learning via Bidirectional Elections and Individual Perspective (AAAI 2024 · similarity 0.88)
- MARS: A Malignity-Aware Backdoor Defense in Federated Learning (NeurIPS 2025 · similarity 0.87)
- SPMC: Self-Purifying Federated Backdoor Defense via Margin Contribution (ICML 2025 · similarity 0.87)
- Theoretically Unmasking Inference Attacks Against LDP-Protected Clients in Federated Vision Models (ICML 2025 · similarity 0.87)
- Infighting in the Dark: Multi-Label Backdoor Attack in Federated Learning (CVPR 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.
