Research map / Adversarial robustness and security / Backdoor and poisoning
Backdoor Attacks on the DNN Interpretation System
Shihong Fang, Anna Choromanska
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 a trigger-conditioned targeted saliency loss | VGG19 Grad-CAM attack · Birds validation, poisoned images | FSR: 99.5 | Improved |
| Add a trigger-conditioned non-targeted saliency loss | ResNet50 Grad-CAM attack · Birds validation, poisoned images | FSR: 99.7 | Improved |
| Target SimpleGrad saliency maps | ResNet50 · Birds validation, poisoned images | FSR: 43.5 | Mixed |
| Add a trigger-conditioned targeted saliency loss | ResNet50 VBP attack · Birds validation, poisoned images | FSR: 99.9 | Improved |
| Add a trigger-conditioned attack objective | DenseNet121 on X-ray · X-ray validation, poisoned images | AUROC: 0.818 -> 0.820 | No clear effect |
| Attack all three interpretation systems jointly | ResNet50 · Birds validation, poisoned images | Top-1 accuracy: 81.7 -> 75.5 | Hurt |
| Attack all three interpretation systems jointly | DenseNet121 · X-ray validation, poisoned images | AUROC: 0.818 -> 0.820 | No clear effect |
| Invert the trigger condition so correct maps require the key | Birds models · Birds validation | FSR: 99.7 | Mixed |
| Apply TV input denoising | DenseNet121 with VBP attack · X-ray validation, poisoned images | FSR: 99.2 -> 15.7 | Hurt |
| Use activation clustering to detect the backdoor | Attacked models · Birds and X-ray validation | misclustering rate | No clear effect |
| Use fine-pruning to remove the backdoor | Attacked models · Birds and X-ray validation | FSR | No clear effect |
| Use Neural Cleanse to detect the backdoor | Attacked models · Birds and X-ray validation | attack detection | No clear effect |
Closest papers by meaning
- Neural Attention Distillation: Erasing Backdoor Triggers from Deep Neural Networks (ICLR 2021 · similarity 0.88)
- Architectural Backdoors in Neural Networks (CVPR 2023 · similarity 0.88)
- BaDExpert: Extracting Backdoor Functionality for Accurate Backdoor Input Detection (ICLR 2024 · similarity 0.87)
- Attention-Imperceptible Backdoor Attacks on Vision Transformers (AAAI 2025 · similarity 0.87)
- A Dual Stealthy Backdoor: From Both Spatial and Frequency Perspectives (AAAI 2024 · similarity 0.86)
- Adversarial Feature Map Pruning for Backdoor (ICLR 2024 · 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.
