Research map / Adversarial robustness and security / Black-box and transfer attacks
ViT-EnsembleAttack: Augmenting Ensemble Models for Stronger Adversarial Transferability in Vision Transformers
Hanwen Cao, Haobo Lu, Xiaosen Wang, Kun He
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 model augmentation (MHD, ASS, MFM) to the ensemble (ablation) | Ens baseline · Average ASR on ViT targets | attack success rate (%): 70.5 -> 93.4 | Improved |
| add model augmentation (MHD, ASS, MFM) to the ensemble (ablation) | Ens baseline · Average ASR on CNN targets | attack success rate (%): 48.1 -> 78.8 | Improved |
| add automatic reweighting (ablation) | Ens baseline · Average ASR on ViT targets | attack success rate (%): 70.5 -> 78.6 | Improved |
| add automatic reweighting (ablation) | Ens baseline · Average ASR on CNN targets | attack success rate (%): 48.1 -> 52.6 | Improved |
| add step size enlargement (ablation) | Ens baseline · Average ASR on ViT targets | attack success rate (%): 70.5 -> 87.1 | Improved |
| add step size enlargement (ablation) | Ens baseline · Average ASR on CNN targets | attack success rate (%): 48.1 -> 65.5 | Improved |
| add automatic reweighting on top of augmentation and enlargement (ablation) | ViT-EnsembleAttack · Average ASR on ViT targets | attack success rate (%): 98.0 -> 98.4 | Improved |
| add automatic reweighting on top of augmentation and enlargement (ablation) | ViT-EnsembleAttack · Average ASR on CNN targets | attack success rate (%): 88.1 -> 88.3 | Improved |
| add automatic reweighting on top of augmentation (ablation) | ViT-EnsembleAttack · Average ASR on ViT targets | attack success rate (%): 93.4 -> 95.7 | Improved |
| add step size enlargement on top of augmentation (ablation) | ViT-EnsembleAttack · Average ASR on CNN targets | attack success rate (%): 78.8 -> 88.1 | Improved |
| increase step size factor q beyond 3 (to 5 and 10) (ablation) | ViT-EnsembleAttack · Average ASR on ViTs and CNNs | attack success rate (%) | Hurt |
| increase inference loop count beyond 2 (ablation) | ViT-EnsembleAttack · Average ASR on ViTs and CNNs | attack success rate (%) | No clear effect |
Closest papers by meaning
- Rethinking Model Ensemble in Transfer-based Adversarial Attacks (ICLR 2024 · similarity 0.90)
- Attacking Transformers with Feature Diversity Adversarial Perturbation (AAAI 2024 · similarity 0.89)
- Understanding Model Ensemble in Transferable Adversarial Attack (ICML 2025 · similarity 0.89)
- Harnessing the Computation Redundancy in ViTs to Boost Adversarial Transferability (NeurIPS 2025 · similarity 0.89)
- Boosting the Transferability of Adversarial Attack on Vision Transformer with Adaptive Token Tuning (NeurIPS 2024 · similarity 0.89)
- Ensemble Diversity Facilitates Adversarial Transferability (CVPR 2024 · similarity 0.88)
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.
