Research map / Adversarial robustness and security / Backdoor and poisoning
Rethink Backdoor Robustness in Vision Transformers
Yichuan Mo, Dongxian Wu, Yifei Wang, Yisen Wang
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 AdamW instead of SGD in the fine-tuning defense for ViTs | FT defense on ViT-B · CIFAR-10, Badnets, ACC | ACC (%): 14.32 -> 95.14 | Improved |
| use AdamW instead of SGD in the fine-tuning defense for ViTs | FT defense on ViT-B · CIFAR-10, Badnets, ASR | ASR (%): 5.02 -> 0.72 | Improved |
| adapt ANP to ViT by pruning channels of linear projections in attention and MLP | ANP on ViT-B · CIFAR-10, Badnets | ASR (%): 100.00 -> 1.34 | Improved |
| adapt AWM to ViT with channel-level pruning of linear projections | AWM on ViT-B · CIFAR-10, Badnets | ACC (%): 85.98 -> 95.02 | Improved |
| add CAT trigger perturbation to Badnets | Badnets on ViT-B vs FT · CIFAR-10 | ASR (%): 0.72 -> 65.01 | Improved |
| add CAT trigger perturbation to SIG | SIG on DeiT-S vs AWM · CIFAR-10 | ASR (%): 3.30 -> 13.81 | Improved |
| add CAT to SIG when no defense is applied | SIG on ViT-B · CIFAR-10, no defense | ASR (%): 90.57 -> 91.19 | Improved |
| add CAT to Badnets on ImageNet | Badnets on ViT-B vs AWM · ImageNet | ASR (%): 24.32 -> 81.98 | Improved |
| add perturbation of triggers (PA) to vanilla attack (ablation) | Badnets vs FP on ViT-B · CIFAR-10 | ASR (%): 13.63% | Improved |
| add random masking of perturbation (RMP) on top of PA (ablation) | CAT, Badnets vs FP on ViT-B · CIFAR-10 | ASR (%): 26.99% | Improved |
| raise attack budget epsilon beyond 16/255 (ablation) | CAT · CIFAR-10, Badnets, FT | ASR (%) | Hurt |
| use both BD and TC losses (gamma 0.6) rather than only one (gamma 0 or 1.0) (ablation) | CAT · CIFAR-10, Badnets, FT | ASR (%) | Improved |
Closest papers by meaning
- Attention-Imperceptible Backdoor Attacks on Vision Transformers (AAAI 2025 · similarity 0.92)
- Not All Prompts Are Secure: A Switchable Backdoor Attack Against Pre-trained Vision Transfomers (CVPR 2024 · similarity 0.89)
- Stealthy Backdoor Attack in Self-Supervised Learning Vision Encoders for Large Vision Language Models (CVPR 2025 · similarity 0.88)
- BadVLA: Towards Backdoor Attacks on Vision-Language-Action Models via Objective-Decoupled Optimization (NeurIPS 2025 · similarity 0.88)
- Revisiting Backdoor Attacks against Large Vision-Language Models from Domain Shift (CVPR 2025 · similarity 0.88)
- Harnessing the Computation Redundancy in ViTs to Boost Adversarial Transferability (NeurIPS 2025 · 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.
