Symmetry-Driven Multimodal Adversarial Attacks: An Information-Theoretic Perspective on Cross-Modal Invariance and Robustness
Entropy · 04.05.2026 · 28 (5) · 521
DOI: 10.3390/e28050521 ↗Kurz gefasst
Fachartikel in der Zeitschrift Entropy, online erschienen am 04.05.2026. Verfasst von Jin Wei und 3 weiteren Autor:innen. Beteiligte Einrichtung: School of Computer Science and Technology, Zhejiang University of Water Resources and Electric Power, Hangzhou 310018, China. Die Arbeit nennt 35 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Zhejiang University of Water Resource and Electric Power.
Abstract (Originalsprache)
Multimodal models such as CLIP and ALBEF essentially maximize cross-modal mutual information to align heterogeneous modalities, utilizing semantic consistency as an implicit prior. However, this alignment mechanism creates a structural vulnerability: the models rely heavily on invariant information coupling. In this work, we investigate this vulnerability and propose a symmetry-driven adversarial attack framework. Unlike standard methods that inject high-entropy unstructured noise, our approach designs collaborative perturbations by modeling semantic-consistent mappings between geometric image transformations and syntactic text variations. By explicitly exploiting the information redundancy inherent in cross-modal symmetries, our method effectively reduces the entropy of the adversarial search space. This reveals a fundamental trade-off between information invariance and robustness, achieving state-of-the-art attack success rates with imperceptible perturbations.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- School of Computer Science and Technology, Zhejiang University of Water Resources and Electric Power, Hangzhou 310018, China.
- Department of Accounting and Management Engineering, Hebei Institute of Mechanical and Electrical Technology, Xingtai 054000, China.
- Institute of Cyberspace Technology, Hong Kong College of Technology, Hong Kong, China.
Förderung
- Zhejiang University of Water Resource and Electric Power
Zitieren
APA
Wei, J., Wang, X., Xu, L., & Li, Y. (2026). Symmetry-Driven Multimodal Adversarial Attacks: An Information-Theoretic Perspective on Cross-Modal Invariance and Robustness. Entropy, 28(5), 521. https://doi.org/10.3390/e28050521
Vancouver
Wei J, Wang X, Xu L, Li Y. Symmetry-Driven Multimodal Adversarial Attacks: An Information-Theoretic Perspective on Cross-Modal Invariance and Robustness. Entropy. 2026;28(5):521. doi:10.3390/e28050521
BibTeX
@article{Wei2026,
author = {Wei, Jin and Wang, Xinyuan and Xu, Liam and Li, Yunfei},
title = {Symmetry-Driven Multimodal Adversarial Attacks: An Information-Theoretic Perspective on Cross-Modal Invariance and Robustness},
journal = {Entropy},
year = {2026},
volume = {28},
number = {5},
pages = {521},
doi = {10.3390/e28050521}
}
EUVIMED