Research map / Federated learning and privacy / Privacy attacks and secure learning
Eliminate Distance Differences Induced by Backdoor Attacks: Layer-Selective Training and Clipping to Mask Backdoor Models
Xuzeng Li, Tao Zhang, Xiangyun Tang, Jiacheng Wang, Jian Wang, Jiawen Kang, Jiqiang Liu, Zhen Han, Dusit Niyato, Dong In Kim
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-selective training with constraint loss and clipping (LaySelFL) to Badnet | Badnet · CIFAR10, Multi-Krum defense, ResNet18 | ASR (%): 3.0 -> 53.0 | Improved |
| add LaySelFL to A3FL | A3FL · CIFAR10, Multi-Krum defense, ResNet18 | ASR (%): 1.0 -> 99.8 | Improved |
| add LaySelFL to A3FL | A3FL · CIFAR10, Multimetric defense, ResNet18 | ASR (%): 1.4 -> 76.2 | Improved |
| add LaySelFL to Badnet against undefended and similarity/direction defenses | Badnet · CIFAR10, AlignIns/Foolsgold/FedAvg, ResNet18 | ASR (%): 97.7 -> 34.2 (AlignIns) | Hurt |
| add LaySelFL to A3FL under RFA | A3FL · SVHN, RFA defense, ResNet18 | ASR (%): 100 -> 0.7 | Hurt |
| add LaySelFL to attacks overall | Badnet and A3FL · three datasets, five defenses | share of attacks prevented by defenses: 26.6% -> 4% | Improved |
| use EfficientNet instead of ResNet18 | LaySelFL · A3FL, CIFAR10, Average | ASR (%): 99.5 -> 61.3 | Hurt |
| add targeted clipping to sensitive-layer constrained training (ablation) | SLI+CT · CIFAR10, Badnet, Multi-Krum, ResNet18 | ASR (%): 34.9 -> 54.2 | Improved |
| add targeted clipping to sensitive-layer constrained training (ablation) | SLI+CT · SVHN, A3FL, AlignIns, ResNet18 | ASR (%): 39.5 -> 22.1 | Hurt |
| add constrained training to SLI+TC (ablation) | SLI+TC · FMNIST, A3FL, Multi-Krum, ResNet18 | ASR (%): 1.4 -> 100 | Improved |
| add constrained training to SLI+TC (ablation) | SLI+TC · CIFAR10, Badnet, FedAvg, ResNet18 | ASR (%): 97.3 -> 36.4 | Hurt |
| increase number of selected sensitive layers (ablation) | LaySelFL · A3FL, ResNet18 | ASR (%) | Improved |
Closest papers by meaning
- 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)
- Privacy Backdoors: Enhancing Membership Inference through Poisoning Pre-trained Models (NeurIPS 2024 · similarity 0.90)
- Beyond Traditional Threats: A Persistent Backdoor Attack on Federated Learning (AAAI 2024 · similarity 0.90)
- MARS: A Malignity-Aware Backdoor Defense in Federated Learning (NeurIPS 2025 · similarity 0.89)
- 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.
