Improving transferability of adversarial examples with mixed representation attack
The Visual Computer · 2026-05 · 42 (7)
DOI: 10.1007/s00371-026-04464-4 ↗Kurz gefasst
Fachartikel in der Zeitschrift The Visual Computer, online erschienen am 19.04.2026. Verfasst von Yunfei Long und 4 weiteren Autor:innen. Die Arbeit nennt 49 Literaturangaben. Gefördert von: National Natural Science Foundation of China under Grants, Fundamental Research Funds for the Central Universities, Hainan Provincial Natural Science Foundation of China.
Abstract (Originalsprache)
Although deep neural networks (DNNs) have achieved remarkable performance in the image classification task, they remain highly vulnerable to adversarial examples, which are crafted by adding human-imperceptible perturbations to benign samples. An important aspect is their transferability, which refers to the ability to deceive target black-box models, enabling attacks in the black-box setting to assess and understand the robustness of DNNs. Recent methods have been proposed to boost adversarial transferability, among which input transformation is one of the most effective approaches. We observe that most existing methods in this direction perform geometric transformations in the spatial domain, ignoring the potential transformation in the latent space, which may limit the transferability of adversarial examples. To tackle this issue, we propose a novel mixed representation attack (MRA) to augment input diversity by exploiting transformations on latent representations. Specifically, MRA leverages a variational autoencoder to generate representations of the input image and images randomly sampled from different categories and then reconstructs images based on mixed representations. Instead of directly computing the average gradient over the reconstructed images, MRA calculates the gradient on the original input mixed with each reconstructed image to generate more transferable adversaries. Extensive experiments on the ImageNet-compatible dataset demonstrate that our MRA achieves state-of-the-art transferability, significantly outperforming various input transformation attacks. Source code will be released in https://github.com/unclelongheu/Mixed-Representation-Attack
Förderung
- National Natural Science Foundation of China under Grants
- Fundamental Research Funds for the Central Universities
- Hainan Provincial Natural Science Foundation of China
Zitieren
APA
Long, Y., Tian, Z., Zhang, L., Xu, H., & Xu, H. (2026). Improving transferability of adversarial examples with mixed representation attack. The Visual Computer, 42(7). https://doi.org/10.1007/s00371-026-04464-4
Vancouver
Long Y, Tian Z, Zhang L, Xu H, Xu H. Improving transferability of adversarial examples with mixed representation attack. The Visual Computer. 2026;42(7). doi:10.1007/s00371-026-04464-4
BibTeX
@article{Long2026,
author = {Long, Yunfei and Tian, Zilin and Zhang, Liguo and Xu, Haocheng and Xu, Huosheng},
title = {Improving transferability of adversarial examples with mixed representation attack},
journal = {The Visual Computer},
year = {2026},
volume = {42},
number = {7},
doi = {10.1007/s00371-026-04464-4}
}
EUVIMED