Research map / Adversarial robustness and security / Backdoor and poisoning
Stealthy Backdoor Attack in Federated Learning via Adaptive Layer-wise Gradient Alignment
Qingqian Yang, Peishen Yan, Xiaoyu Wu, Jiaru Zhang, Tao Song, Yang Hua, Hao Wang, Liangliang Wang, 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 |
|---|---|---|---|
| add layer-wise gradient alignment instead of no alignment (ablation) | BadNets-style attack · CIFAR-10 VGG19, FLAME | backdoor accuracy (%): 7.05 -> 97.46 | Improved |
| add layer-wise gradient alignment instead of no alignment (ablation) | BadNets-style attack · CIFAR-100 ResNet18, Multi-Metrics | backdoor accuracy (%): 0.37 -> 97.05 | Improved |
| use layer-wise instead of model-wise alignment (ablation) | LGA · CIFAR-10 VGG19, FLAME | backdoor accuracy (%): 45.79 -> 97.46 | Improved |
| use layer-wise instead of model-wise alignment (ablation) | LGA · CIFAR-100 ResNet18, no defense | backdoor accuracy (%): 73.29 -> 96.70 | Improved |
| use layer-wise instead of model-wise alignment (ablation) | LGA · CIFAR-10 ResNet18, FLAME | backdoor accuracy (%): 85.32 -> 98.16 | Improved |
| add layer-wise alignment to alignment-free attack (no defense) (ablation) | BadNets · CIFAR-10 VGG19, no defense | backdoor accuracy (%): 97.70 -> 96.16 | Hurt |
| add alignment to trigger-optimization attack IBA | IBA · FLAME | backdoor accuracy (%): 2.71 -> 98.04 | Improved |
| add alignment to trigger-optimization attack Cerp | Cerp · FLAME | backdoor accuracy (%): 29.84 -> 99.64 | Improved |
| add alignment to A3FL | A3FL · Multi-Metrics | backdoor accuracy (%): 86.29 -> 99.89 | Improved |
| use LGA instead of LP attack | LP attack · CIFAR100 ResNet18 no defense | best backdoor accuracy (%): 72.24 -> 98.51 | Improved |
| use LGA instead of LP attack on malicious acceptance | LP attack · CIFAR10 ResNet18, MultiKrum | MAR: 0.72 -> 0.68 | Hurt |
| increase attack interval F to 5 | LGA vs LP attack · CIFAR-100 | backdoor accuracy | Improved |
Closest papers by meaning
- Detecting Backdoor Attacks in Federated Learning via Direction Alignment Inspection (CVPR 2025 · similarity 0.91)
- Beyond Traditional Threats: A Persistent Backdoor Attack on Federated Learning (AAAI 2024 · similarity 0.91)
- Backdoor Federated Learning by Poisoning Backdoor-Critical Layers (ICLR 2024 · similarity 0.90)
- Resisting Backdoor Attacks in Federated Learning via Bidirectional Elections and Individual Perspective (AAAI 2024 · similarity 0.90)
- SADBA: Self-Adaptive Distributed Backdoor Attack Against Federated Learning (AAAI 2025 · similarity 0.90)
- Infighting in the Dark: Multi-Label Backdoor Attack in Federated Learning (CVPR 2025 · similarity 0.89)
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.
