Research map / Federated learning and privacy / Privacy attacks and secure learning
Lockdown: Backdoor Defense for Federated Learning with Isolated Subspace Training
Tiansheng Huang, Sihao Hu, Ka-Ho Chow, Fatih Ilhan, Selim Furkan Tekin, Ling Liu
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 gradient-based recovery instead of random recovery (ablation) | Lockdown · CIFAR10 IID | ASR (%): 13.0 -> 1.4 | Improved |
| use gradient-based recovery instead of random recovery (ablation) | Lockdown · CIFAR10 IID | Benign Acc (%): 91.1 -> 90.7 | Hurt |
| use gradient-based recovery instead of random recovery (ablation) | Lockdown · CIFAR10 Non-IID | ASR (%): 17.2 -> 2.2 | Improved |
| use ERK initialization instead of uniform (ablation) | Lockdown · CIFAR10 IID | Benign Acc (%): 81.7 -> 90.1 | Improved |
| use ERK initialization instead of uniform (ablation) | Lockdown · CIFAR10 Non-IID | Benign Acc (%): 75.5 -> 86.1 | Improved |
| use ERK initialization instead of uniform (ablation) | Lockdown · CIFAR10 IID | ASR (%): 5.7 -> 7.1 | No clear effect |
| increase sparsity s from 0 to 0.75 (ablation) | Lockdown · CIFAR10 IID | ASR (%): 68.4 -> 7.1 | Improved |
| increase sparsity s from 0.75 to 0.9 (ablation) | Lockdown · CIFAR10 IID | Benign Acc (%): 90.1 -> 88.3 | Hurt |
| set initial pruning rate alpha0 from 0 to 1e-4 (ablation) | Lockdown · CIFAR10 IID | ASR (%): 49.4 -> 7.1 | Improved |
| raise alpha0 from 1e-4 to 1e-1 (ablation) | Lockdown · CIFAR10 IID | Benign Acc (%): 90.1 -> 87.0 | Hurt |
| enlarge alpha0 to 0.1 against FixMask attack (ablation) | Lockdown · CIFAR10 IID FixMask | ASR (%): 63.5 -> 3.5 | Improved |
| add heterogeneous mask initialization with alpha0=0.1 against FixMask (ablation) | Lockdown · CIFAR10 Non-IID FixMask | ASR (%): 4.6 -> 3.7 | No clear effect |
Later papers that compared with this method
2 later accepted papers report LockDown as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- DoBlock: Blocking Malicious Association Propagation for Backdoor-Robust Federated Learning Under Domain Skew (AAAI 2026 · better on 1 of 1 shared result)
- Detecting Backdoor Attacks in Federated Learning via Direction Alignment Inspection (CVPR 2025 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Backdoor Federated Learning by Poisoning Backdoor-Critical Layers (ICLR 2024 · similarity 0.87)
- SPMC: Self-Purifying Federated Backdoor Defense via Margin Contribution (ICML 2025 · similarity 0.87)
- Parameter Disparities Dissection for Backdoor Defense in Heterogeneous Federated Learning (NeurIPS 2024 · similarity 0.87)
- Beyond Traditional Threats: A Persistent Backdoor Attack on Federated Learning (AAAI 2024 · similarity 0.86)
- SADBA: Self-Adaptive Distributed Backdoor Attack Against Federated Learning (AAAI 2025 · similarity 0.86)
- Detecting Backdoor Attacks in Federated Learning via Direction Alignment Inspection (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.
