Research map / Adversarial robustness and security / Backdoor and poisoning
Invisible Backdoor Attack With Sample-Specific Triggers
Yuezun Li, Yiming Li, Baoyuan Wu, Longkang Li, Ran He, Siwei Lyu
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 trigger generated from a different test image (inconsistent trigger) (ablation) | SSBA · ImageNet | ASR (%): 99.5 -> 23.3 | Hurt |
| use trigger generated from a different test image (inconsistent trigger) (ablation) | SSBA · MS-Celeb-1M | ASR (%): 100 -> 98.1 | No clear effect |
| train encoder on another dataset (MS-Celeb-1M for ImageNet) (ablation) | SSBA · ImageNet | ASR (%): 99.5 -> 99.4 | No clear effect |
| train encoder on ImageNet for MS-Celeb-1M classifier (ablation) | SSBA · MS-Celeb-1M | ASR (%): 100 -> 99.5 | No clear effect |
| change target label to 1, 2 or 3 (ablation) | SSBA · ImageNet | BA/ASR (%): 85.4/99.4, 85.6/99.3, 85.6/99.5 | No clear effect |
| use sample-specific triggers instead of BadNets patch | SSBA vs BadNets · ImageNet | PSNR: 25.635 -> 27.195 | Improved |
| use sample-specific triggers instead of Blended Attack | SSBA vs Blended Attack · ImageNet | ASR (%): 95.8 -> 99.5 | Improved |
| use sample-specific triggers instead of Blended Attack | SSBA vs Blended Attack · ImageNet | PSNR: 45.809 -> 27.195 | Hurt |
| increase the poisoning rate gamma (ablation) | SSBA · ImageNet and MS-Celeb-1M | ASR (%): ASR > 95% with only 2% poisoned samples | Improved |
| prune 20% of neurons (fine-pruning defense) | BadNets and Blended Attack vs SSBA · ImageNet and MS-Celeb-1M | ASR (%): Blended Attack ASR below 10% | Improved |
| replace ImageNet test images by COCO or random noise inputs (ablation) | SSBA · ImageNet | ASR (%): 99.5 -> 100 | No clear effect |
Closest papers by meaning
- Hidden Trigger Backdoor Attacks (AAAI 2020 · similarity 0.93)
- CLEAR: Clean-Up Sample-Targeted Backdoor in Neural Networks (ICCV 2021 · similarity 0.91)
- Wicked Oddities: Selectively Poisoning for Effective Clean-Label Backdoor Attacks (ICLR 2025 · similarity 0.90)
- Adversarial-Inspired Backdoor Defense via Bridging Backdoor and Adversarial Attacks (AAAI 2025 · similarity 0.89)
- A Set of Generalized Components to Achieve Effective Poison-only Clean-label Backdoor Attacks with Collaborative Sample Selection and Triggers (NeurIPS 2025 · similarity 0.89)
- Demystifying Poisoning Backdoor Attacks from a Statistical Perspective (ICLR 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.
