Research map / Adversarial robustness and security / Backdoor and poisoning
Backdoor Defense via Enhanced Splitting and Trap Isolation
Hongrui Yu, Lu Qi, Wanyu Lin, Jian Chen, Hailong Sun, Chengbin Sun
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 trap isolation stage after enhanced splitting (ablation) | ESTI · CIFAR-10 (ES ablation, avg over 8 attacks) | ASR (%): 0.9 -> 0.2 | Improved |
| add trap isolation stage after enhanced splitting (ACC) (ablation) | ESTI · CIFAR-10 (ES ablation, avg over 8 attacks) | ACC (%): 92.7 -> 93.7 | Improved |
| use enhanced splitting clean pool instead of all clean data (ablation) | Enhanced splitting · CIFAR-10 (ES ablation, avg over 8 attacks) | ASR (%): 0.7 -> 0.9 | No clear effect |
| reduce reserved clean seed samples from 1% to 0.1% (ablation) | ESTI · CIFAR-10 (BadNets, seed proportion) | ACC (%): 93.7 -> 93.4 | No clear effect |
| increase poisoning ratio from 5% to 20% for BadNets (ablation) | ESTI · CIFAR-10 (BadNets, poisoning ratio) | ASR (%): 0.2 -> 0.4 | No clear effect |
| vary target label across 0-9 on GTSRB (ablation) | ESTI · GTSRB (BadNets, target labels) | ASR (%): 0.0 -> 0.1 | No clear effect |
| ESTI versus PDB on average ASR | PDB · CIFAR-10 (avg over 8 attacks) | ASR (%): 0.5 -> 0.2 | Improved |
| ESTI versus ABL on clean-data ACC | ABL · CIFAR-10 (no poisoning) | ACC (%): 64.3 -> 94.1 | Improved |
| ESTI versus ASD on training time | ASD · CIFAR-10 training time | Training time (s): 15660 -> 9862 | Improved |
| ESTI versus ASD on Tiny-ImageNet | ASD · Tiny-ImageNet (avg over 4 attacks) | ASR (%): 99.4 -> 0.1 | Improved |
Closest papers by meaning
- Uncovering, Explaining, and Mitigating the Superficial Safety of Backdoor Defense (NeurIPS 2024 · similarity 0.90)
- Bi-perspective Splitting Defense: Achieving Clean-Seed-Free Backdoor Security (ICML 2025 · similarity 0.89)
- BaDExpert: Extracting Backdoor Functionality for Accurate Backdoor Input Detection (ICLR 2024 · similarity 0.89)
- Fusing Pruned and Backdoored Models: Optimal Transport-based Data-free Backdoor Mitigation (AAAI 2025 · similarity 0.89)
- Unveiling and Mitigating Backdoor Vulnerabilities based on Unlearning Weight Changes and Backdoor Activeness (NeurIPS 2024 · similarity 0.89)
- Backdoor Defense via Test-Time Detecting and Repairing (CVPR 2024 · 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.
