EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

Leveraging Generalizability of Image-to-Image Translation for Enhanced Adversarial Defense

Haibo Zhang, Zhihua Yao, Kouichi Sakurai, Takeshi Saitoh

SN Computer Science · 07.07.2026 · 7 (6)

DOI: 10.1007/s42979-026-05198-9 ↗

Kurz gefasst

Fachartikel in der Zeitschrift SN Computer Science, online erschienen am 07.07.2026. Verfasst von Haibo Zhang und 3 weiteren Autor:innen. Die Arbeit nennt 54 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: Japan Society for the Promotion of Science.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In the rapidly advancing domain of artificial intelligence, machine learning stands out as a crucial technology, notable for its immense potential and associated risks. The stability and reliability of these models are important, given that they are often aimed at by security threats. Adversarial attacks, which were initially thoroughly defined by Ian Goodfellow et al. in 2013, highlight a significant vulnerability: they can trick machine learning models into making incorrect predictions by applying nearly invisible perturbations to images. Although many studies have focused on constructing sophisticated defensive mechanisms to mitigate such attacks, they often overlook the substantial time and computational costs of training and maintaining these models. In an ideal scenario, a defense method should be able to mechanism ought to effectively generalize across different, including previously un-encountered, adversarial attacks while maintaining minimal overhead. Building on our previous work on image-to-image translation-based defenses, this study introduces an improved model that incorporates residual blocks to enhance generalizability. The proposed method requires training only a single model, effectively defends against diverse attack types, and is well-transferable between different target models. Experiments show that our model can restore the classification accuracy from near zero to an average of 72% while maintaining competitive performance compared to state-of-the-art methods. Significantly, our model operates more efficiently, reducing the time needed to process individual images and speeding up the training process to achieve faster convergence. Robustness tests further confirm stable performance under varying attack strengths, demonstrating the model’s practical value in real-world adversarial settings. The code is avaible at https://github.com/Haiboz0105/img2imgAdvDef.

Förderung

  • Japan Society for the Promotion of Science

Zitieren

APA

Zhang, H., Yao, Z., Sakurai, K., & Saitoh, T. (2026). Leveraging Generalizability of Image-to-Image Translation for Enhanced Adversarial Defense. SN Computer Science, 7(6). https://doi.org/10.1007/s42979-026-05198-9

Vancouver

Zhang H, Yao Z, Sakurai K, Saitoh T. Leveraging Generalizability of Image-to-Image Translation for Enhanced Adversarial Defense. SN Computer Science. 2026;7(6). doi:10.1007/s42979-026-05198-9

BibTeX
@article{Zhang2026,
  author = {Zhang, Haibo and Yao, Zhihua and Sakurai, Kouichi and Saitoh, Takeshi},
  title = {Leveraging Generalizability of Image-to-Image Translation for Enhanced Adversarial Defense},
  journal = {SN Computer Science},
  year = {2026},
  volume = {7},
  number = {6},
  doi = {10.1007/s42979-026-05198-9}
}
Kostenlos starten →