Feature Compression May Be the Root Cause of Adversarial Fragility in Neural Network Classifiers (Student Abstract)
Proceedings of the AAAI Conference on Artificial Intelligence · 14.03.2026 · 40 (48) · 41212-41213
DOI: 10.1609/aaai.v40i48.42217 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 14.03.2026. Verfasst von Jingchao Gao und 8 weiteren Autor:innen.
Abstract (Originalsprache)
In this paper, we study the adversarial robustness of deep neural networks (DNN) for classification against optimal classifiers. We look at the smallest magnitude of possible additive perturbations that can change a classifier's output. We provide a matrix-theoretic explanation of the adversarial fragility of DNNs for classification. In particular, our theoretical results show that the adversarial robustness of a neural network can degrade as the input dimension d increases. Analytically, we show that the adversarial robustness of neural networks can be only 1/√d of the best possible adversarial robustness of optimal classifiers. Our theories match remarkably well with empirical results. The matrix-theoretic explanation aligns with an earlier information-theoretic feature-compression-based explanation for the adversarial fragility of neural networks.
Zitieren
APA
Gao, J., Lu, Z., Mudumbai, R., Wu, X., Yi, J., Cho, M., Xu, C., Xie, H., & Xu, W. (2026). Feature Compression May Be the Root Cause of Adversarial Fragility in Neural Network Classifiers (Student Abstract). Proceedings of the AAAI Conference on Artificial Intelligence, 40(48), 41212-41213. https://doi.org/10.1609/aaai.v40i48.42217
Vancouver
Gao J, Lu Z, Mudumbai R, Wu X, Yi J, Cho M, et al. Feature Compression May Be the Root Cause of Adversarial Fragility in Neural Network Classifiers (Student Abstract). Proceedings of the AAAI Conference on Artificial Intelligence. 2026;40(48):41212-41213. doi:10.1609/aaai.v40i48.42217
BibTeX
@article{Gao2026,
author = {Gao, Jingchao and Lu, Ziqing and Mudumbai, Raghu and Wu, Xiaodong and Yi, Jirong and Cho, Myung and Xu, Catherine and Xie, Hui and Xu, Weiyu},
title = {Feature Compression May Be the Root Cause of Adversarial Fragility in Neural Network Classifiers (Student Abstract)},
journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
year = {2026},
volume = {40},
number = {48},
pages = {41212--41213},
doi = {10.1609/aaai.v40i48.42217}
}
EUVIMED