Research map / Adversarial robustness and security / Backdoor and poisoning
Beating Backdoor Attack at Its Own Game
Min Liu, Alberto Sangiovanni-Vincentelli, Xiangyu Yue
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 |
|---|---|---|---|
| inject non-adversarial backdoor (NAB) instead of NAD/ABL erasing | backdoor defense · CIFAR-10 ResNet-18 average | clean accuracy (%): NAD 89.00, ABL 90.04 -> NAB 92.44 | Improved |
| use NAB instead of ABL on tiny-ImageNet | backdoor defense · tiny-ImageNet | attack success rate (%): ABL 0.23 vs NAB 0.87 | Hurt |
| use ResNet-50 instead of ResNet-18 for NAB (ablation) | NAB · CIFAR-10 average | attack success rate (%): 0.52 -> 0.19 | Improved |
| add SSL pretrained initialization to NAB (ablation) | NAB · CIFAR-10 | clean accuracy (%): 93.63 -> 94.35 | Improved |
| add test data filtering to NAB (ablation) | NAB · CIFAR-10 ResNet-18 | defense success rate (%): 94.37 -> 99.39 | Improved |
| set detection rate mu much lower than poisoning rate (ablation) | NAB · CIFAR-10 BadNets | attack success rate (%): 0.61 -> 78.72 (mu 0.05 vs 0.01 at lambda 0.2; ASR at mu=0.01, lambda=0.2: 78.72) | Hurt |
| set detection rate mu larger than poisoning rate (ablation) | NAB · CIFAR-10 BadNets | clean accuracy (%): 94.45 -> 91.23 (lambda 0.01, mu 0.01 -> 0.10) | Hurt |
| increase pseudo label accuracy via nearest-center instead of verified data labels (ablation) | NAB · CIFAR-10 | pseudo label quality | Improved |
| apply NAB to all-to-all attack instead of all-to-one | NAB · CIFAR-10 | attack success rate (%): ResNet-18 BadNets 93.37 -> 2.66 | Mixed |
Closest papers by meaning
- Adversarial-Inspired Backdoor Defense via Bridging Backdoor and Adversarial Attacks (AAAI 2025 · similarity 0.93)
- Wicked Oddities: Selectively Poisoning for Effective Clean-Label Backdoor Attacks (ICLR 2025 · similarity 0.90)
- Backdoor Defense via Test-Time Detecting and Repairing (CVPR 2024 · similarity 0.90)
- Mitigating Backdoor Attack by Injecting Proactive Defensive Backdoor (NeurIPS 2024 · similarity 0.90)
- Inspecting Prediction Confidence for Detecting Black-Box Backdoor Attacks (AAAI 2024 · similarity 0.90)
- Unveiling and Mitigating Backdoor Vulnerabilities based on Unlearning Weight Changes and Backdoor Activeness (NeurIPS 2024 · similarity 0.90)
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.
