Improving Integrated Gradient-based Transferable Adversarial Examples by Refining the Integration Path
Proceedings of the AAAI Conference on Artificial Intelligence · 11.04.2025 · 39 (7) · 6731-6739
DOI: 10.1609/aaai.v39i7.32722 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 11.04.2025. Verfasst von Yuchen Ren und 6 weiteren Autor:innen. Die Arbeit wurde bisher 17-mal zitiert (Crossref).
Abstract (Originalsprache)
Transferable adversarial examples are known to cause threats in practical, black-box attack scenarios. A notable approach to improving transferability is using integrated gradients (IG), originally developed for model interpretability. In this paper, we find that existing IG-based attacks have limited transferability due to their naive adoption of IG in model interpretability. To address this limitation, we focus on the IG integration path and refine it in three aspects: multiplicity, monotonicity, and diversity, supported by theoretical analyses. We propose the Multiple Monotonic Diversified Integrated Gradients (MuMoDIG) attack, which can generate highly transferable adversarial examples on different CNN and ViT models and defenses. Experiments validate that MuMoDIG outperforms the latest IG-based attack by up to 37.3% and other state-of-the-art attacks by 8.4%. In general, our study reveals that migrating established techniques to improve transferability may require non-trivial efforts.
Zitieren
APA
Ren, Y., Zhao, Z., Lin, C., Yang, B., Zhou, L., Liu, Z., & Shen, C. (2025). Improving Integrated Gradient-based Transferable Adversarial Examples by Refining the Integration Path. Proceedings of the AAAI Conference on Artificial Intelligence, 39(7), 6731-6739. https://doi.org/10.1609/aaai.v39i7.32722
Vancouver
Ren Y, Zhao Z, Lin C, Yang B, Zhou L, Liu Z, et al. Improving Integrated Gradient-based Transferable Adversarial Examples by Refining the Integration Path. Proceedings of the AAAI Conference on Artificial Intelligence. 2025;39(7):6731-6739. doi:10.1609/aaai.v39i7.32722
BibTeX
@article{Ren2025,
author = {Ren, Yuchen and Zhao, Zhengyu and Lin, Chenhao and Yang, Bo and Zhou, Lu and Liu, Zhe and Shen, Chao},
title = {Improving Integrated Gradient-based Transferable Adversarial Examples by Refining the Integration Path},
journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
year = {2025},
volume = {39},
number = {7},
pages = {6731--6739},
doi = {10.1609/aaai.v39i7.32722}
}
EUVIMED