Research map / Adversarial robustness and security / Backdoor and poisoning
Color Backdoor: A Robust Poisoning Attack in Color Space
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 PSO rather than random trigger selection | Color backdoor · CIFAR-10 | ASR: 92.02 -> 97.55 | Helped |
| Use PSO rather than random trigger selection | Color backdoor · CIFAR-100 | ASR: 83.54 -> 96.27 | Helped |
| Use PSO rather than random trigger selection | Color backdoor · GTSRB | ASR: 91.33 -> 99.70 | Helped |
| Use PSO rather than random trigger selection | Color backdoor · ImageNet | ASR: 87.09 -> 98.16 | Helped |
| Use PSO rather than grid-search for trigger selection | Color backdoor · CIFAR-10 | trigger-search time: 5.33 h -> 1.79 h | Helped |
| Use PSO rather than Genetic Algorithm for trigger selection | Color backdoor · CIFAR-100 | trigger-search time: 6.30 h -> 3.71 h | Helped |
| Add a naturalness restriction during trigger search (ablation) | Color backdoor · triggered images | visual naturalness | Helped |
| Increase poisoning rate from 3% to 5% (ablation) | Color backdoor · CIFAR-10 | ASR: 93.77 -> 97.55 | Helped |
| Increase poisoning rate from 5% to 10% (ablation) | Color backdoor · CIFAR-10 | ACC: 89.77 -> 87.61 | Hurt |
| Apply color backdoor rather than Filter under JPEG compression | CIFAR-10 backdoor attack · CIFAR-10, JPEG compression | ASR: 82.95 -> 81.78 | No clear effect |
| Apply color backdoor rather than BadNet under JPEG compression | CIFAR-10 backdoor attack · CIFAR-10, JPEG compression | ASR: 81.15 -> 81.78 | No clear effect |
| Train with color space augmentation (ablation) | Color backdoor · CIFAR-10 | ASR: 97.55 -> 86.01 | Hurt |
Closest papers by meaning
- A Dual Stealthy Backdoor: From Both Spatial and Frequency Perspectives (AAAI 2024 · similarity 0.87)
- Conditional Backdoor Attack via JPEG Compression (AAAI 2024 · similarity 0.87)
- A Set of Generalized Components to Achieve Effective Poison-only Clean-label Backdoor Attacks with Collaborative Sample Selection and Triggers (NeurIPS 2025 · similarity 0.86)
- A Closer Look at Backdoor Attacks on CLIP (ICML 2025 · similarity 0.85)
- Attention-Imperceptible Backdoor Attacks on Vision Transformers (AAAI 2025 · similarity 0.84)
- Data Poisoning based Backdoor Attacks to Contrastive Learning (CVPR 2024 · similarity 0.84)
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.