Research map / Federated learning and privacy / Privacy attacks and secure learning
SkyMask: Attack-agnostic Robust Federated Learning with Fine-grained Learnable Masks
Peishen Yan, Hao Wang, Tao Song, Yang Hua, Ruhui Ma, Ningxin Hu, Mohammad Reza Haghighat, Haibing Guan
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 SkyMask learnable-mask defense instead of FLTrust under fine-grained attacks | FL aggregation · CIFAR-100 (ResNet20), Min-Max | testing accuracy: 0.30 -> 0.44 | Improved |
| use SkyMask under no attack instead of FLTrust | FL aggregation · CIFAR-100 (ResNet20), no attack | testing accuracy: 0.39 -> 0.44 | Improved |
| drop the trusted root model (SkyMask-NR) when under 50% malicious (ablation) | SkyMask · CIFAR-10 (ResNet20), 20% malicious | testing accuracy: 0.77 -> 0.77 (Min-Max) | No clear effect |
| use learnable-mask clustering instead of FLDetector for malicious client detection | malicious client detection · CIFAR-10 (ResNet20), Min-Max | testing accuracy: 0.11 -> 0.77 | Improved |
| use learnable-mask clustering instead of Tolpegin defense | malicious client detection · CIFAR-10 (ResNet20), Fang-Krum | testing accuracy: 0.17 -> 0.77 | Improved |
| use SkyMask at high malicious fractions (above 50%) versus FLTrust | FL aggregation · Fashion-MNIST (CNN), 40-80% malicious | convergence and accuracy | Improved |
| use Krum as defense when there is no attack | FL aggregation · CIFAR-10/100, no attack | testing accuracy: 0.77 -> 0.54 (CIFAR-10) | Hurt |
| scale to 200 and 500 clients with SkyMask | SkyMask · CIFAR-10 (ResNet20) | FPR and FNR: FPR and FNR remain 0% for fine-grained and targeted attacks | Improved |
Closest papers by meaning
- FedInv: Byzantine-Robust Federated Learning by Inversing Local Model Updates (AAAI 2022 · similarity 0.90)
- Resisting Backdoor Attacks in Federated Learning via Bidirectional Elections and Individual Perspective (AAAI 2024 · similarity 0.87)
- Efficient Federated Learning against Byzantine Attacks and Data Heterogeneity via Aggregating Normalized Gradients (NeurIPS 2025 · similarity 0.87)
- Towards the Robustness of Differentially Private Federated Learning (AAAI 2024 · similarity 0.87)
- Dual Defense: Enhancing Privacy and Mitigating Poisoning Attacks in Federated Learning (NeurIPS 2024 · similarity 0.87)
- SPMC: Self-Purifying Federated Backdoor Defense via Margin Contribution (ICML 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.
