Research map / Adversarial robustness and security / Backdoor and poisoning
Defending Backdoor Attacks on Vision Transformer via Patch Processing
Khoa D. Doan, Yingjie Lao, Peng Yang, Ping Li
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 |
|---|---|---|---|
| Apply PatchDrop to inputs of a BadNets-poisoned ViT and use prediction-change frequency for detection | ViT · BadNets, CIFAR10, GTSRB, and TinyImageNet | TPR: 90.08; 94.89; 95.80 | Improved |
| Apply PatchShuffle to inputs of a ReFool-poisoned ViT and use prediction-change frequency for detection | ViT · ReFool, CIFAR10, GTSRB, and TinyImageNet | TPR: 90.00; 84.10; 90.90 | Improved |
| Filter detected BadNets samples and retrain the ViT | ViT · CIFAR10, GTSRB, and TinyImageNet | ASR: 98.47 -> 10.01; 96.46 -> 0.48; 98.02 -> 0.51 | Improved |
| Filter detected BadNets samples and retrain the ViT | ViT · CIFAR10, GTSRB, and TinyImageNet | clean accuracy: 98.93 -> 98.94; 98.68 -> 98.38; 86.46 -> 85.57 | Mixed |
| Use PatchDrop on clean ViT inputs | ViT · CIFAR10 and GTSRB | clean accuracy | No clear effect |
| Use PatchDrop on patch-triggered poisoned ViT inputs | ViT · Patch-based attacks | ASR | Improved |
| Use PatchShuffle on clean ViT inputs | ViT · Blending-based attacks | clean accuracy | Hurt |
| Use PatchShuffle on blending-trigger poisoned ViT inputs | ViT · Blending-based attacks | ASR | No clear effect |
| Use PatchDrop for blending-based backdoors | ViT · Blending-based attacks | ASR | No clear effect |
| Use Patch processing detection on ViT rather than CNN models | Patch Processing defense · BadNets and ReFool | detection performance | Mixed |
Closest papers by meaning
- Attention-Imperceptible Backdoor Attacks on Vision Transformers (AAAI 2025 · similarity 0.92)
- Boosting the Transferability of Adversarial Attack on Vision Transformer with Adaptive Token Tuning (NeurIPS 2024 · similarity 0.88)
- Attacking Transformers with Feature Diversity Adversarial Perturbation (AAAI 2024 · similarity 0.88)
- Towards Understanding and Improving Adversarial Robustness of Vision Transformers (CVPR 2024 · similarity 0.87)
- Harnessing the Computation Redundancy in ViTs to Boost Adversarial Transferability (NeurIPS 2025 · similarity 0.87)
- Not All Prompts Are Secure: A Switchable Backdoor Attack Against Pre-trained Vision Transfomers (CVPR 2024 · similarity 0.87)
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.
