Research map / Adversarial robustness and security / Black-box and transfer attacks
On Improving Adversarial Transferability of Vision Transformers
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 |
|---|---|---|---|
| Add self-ensemble with refined tokens to MIM | DeiT-B · ImageNet val 5k; target Res152 | fool rate (%): 44.66 -> 70.18 | Helped |
| Add self-ensemble without token refinement to MIM (ablation) | DeiT-B · ImageNet val 5k; target Res152 | fool rate (%): 44.66 -> 54.30 | Helped |
| Refine tokens in the self-ensemble (ablation) | DeiT-B MIM^E · ImageNet val 5k; target Res152 | fool rate (%): 54.30 -> 70.18 | Helped |
| Add self-ensemble with refined tokens to MIM | DeiT-B · ImageNet val 5k; target ViT-S | fool rate (%): 84.74 -> 94.10 | Helped |
| Add self-ensemble without token refinement to MIM (ablation) | DeiT-B · ImageNet val 5k; target ViT-S | fool rate (%): 84.74 -> 94.46 | Helped |
| Refine tokens in the self-ensemble (ablation) | DeiT-B MIM^E · ImageNet val 5k; target ViT-S | fool rate (%): 94.46 -> 94.10 | Hurt |
| Add self-ensemble with refined tokens to PGD | DeiT-B · ImageNet val 5k; Swin Transformer | fool rate (%): 29.3 -> 46.8 | Helped |
| Add self-ensemble without token refinement to PGD (ablation) | DeiT-B · ImageNet val 5k; Swin Transformer | fool rate (%): 29.3 -> 35.7 | Helped |
| Refine tokens in the self-ensemble (ablation) | DeiT-B PGD^E · ImageNet val 5k; Swin Transformer | fool rate (%): 35.7 -> 46.8 | Helped |
| Add self-ensemble with refined tokens to MIM | DeiT-B · COCO val; classification-to-detection transfer on DETR | mAP at [0.5:0.95] IOU: 32.6 -> 30.6 | Helped |
| Add self-ensemble with refined tokens to MIM | DeiT-B · ImageNet val 5k; classification-to-segmentation transfer on DINO | Jaccard index: 32.6 -> 30.6 | Helped |
Closest papers by meaning
- Harnessing the Computation Redundancy in ViTs to Boost Adversarial Transferability (NeurIPS 2025 · similarity 0.94)
- Improving the Adversarial Transferability of Vision Transformers with Virtual Dense Connection (AAAI 2024 · similarity 0.94)
- Boosting the Transferability of Adversarial Attack on Vision Transformer with Adaptive Token Tuning (NeurIPS 2024 · similarity 0.93)
- Attacking Transformers with Feature Diversity Adversarial Perturbation (AAAI 2024 · similarity 0.91)
- Improving Adversarial Transferability on Vision Transformers via Forward Propagation Refinement (CVPR 2025 · similarity 0.90)
- Random Entangled Tokens for Adversarially Robust Vision Transformer (CVPR 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.