EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

AMGAA: Attention-Guided Multi-Target Generative Adversarial Attack for Vision Transformers

Dongbo Ou iD, Jintian Lu, Shihui Zhou, Ying Zeng, Dongwan Liao, Haoyin Liu, Chao Yang, Yingsheng He, Jie Tian iD

Entropy · 12.06.2026 · 28 (6) · 680

DOI: 10.3390/e28060680 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Entropy, online erschienen am 12.06.2026. Verfasst von Dongbo Ou und 8 weiteren Autor:innen. Beteiligte Einrichtung: School of Computer Science and Engineering, Jishou University, Jishou 416000, China. Die Arbeit nennt 46 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Vision Transformers (ViTs) have achieved strong performance in computer vision, but their adversarial robustness remains underexplored. Existing ViT-oriented attacks mainly rely on iterative optimization, leading to high generation cost and limited transferability. Moreover, most generative attacks target a single class, making them inefficient for multi-target scenarios. To address these issues, we propose Attention-Guided Multi-Target Generative Adversarial Attack (AMGAA). AMGAA leverages ViT self-attention to guide target feature fusion and adaptively selects important source patches for perturbation generation. It jointly optimizes adversarial, attention constraint, and total variation losses to improve targeted attack success while preserving visual naturalness. Experiments on CIFAR-10 and ImageNet show that AMGAA achieves average attack success rates (ASRs) of 43.2% and 39.0% in ImageNet single-target transfer attacks and CIFAR-10 multi-target attacks, respectively. Compared with the generative attack methods evaluated in our experiments, AMGAA improves ASR by 5.7 percentage points in the multi-target setting and by 8.1 percentage points in unknown-class generalization. AMGAA also obtains a low LPIPS of 0.018, indicating good visual imperceptibility. Ablation studies confirm the effectiveness of its key components.

Schlagwörter

Stichwörter der Autor:innen

adversarial attackattention mechanismmulti-target attacksecurity of AItransfer attackvision transformer

Autor:innen und Institutionen

  • School of Computer Science and Engineering, Jishou University, Jishou 416000, China.

ORCID-Profile: Dongbo Ou, Jie Tian

Zitieren

APA

Ou, D., Lu, J., Zhou, S., Zeng, Y., Liao, D., Liu, H., Yang, C., He, Y., & Tian, J. (2026). AMGAA: Attention-Guided Multi-Target Generative Adversarial Attack for Vision Transformers. Entropy, 28(6), 680. https://doi.org/10.3390/e28060680

Vancouver

Ou D, Lu J, Zhou S, Zeng Y, Liao D, Liu H, et al. AMGAA: Attention-Guided Multi-Target Generative Adversarial Attack for Vision Transformers. Entropy. 2026;28(6):680. doi:10.3390/e28060680

BibTeX
@article{Ou2026,
  author = {Ou, Dongbo and Lu, Jintian and Zhou, Shihui and Zeng, Ying and Liao, Dongwan and Liu, Haoyin and Yang, Chao and He, Yingsheng and Tian, Jie},
  title = {AMGAA: Attention-Guided Multi-Target Generative Adversarial Attack for Vision Transformers},
  journal = {Entropy},
  year = {2026},
  volume = {28},
  number = {6},
  pages = {680},
  doi = {10.3390/e28060680}
}
Kostenlos starten →