EUVIMEDEUVIMED entdecken
Crossref

A Reinforcement Learning-Driven Adversarial Attack Methods With Dynamic Perturbation Optimization

Lulu Wang, Xiaoyi Wang, Xiaoning Du, Jianming Chang, Bixin Li

IEEE Transactions on Reliability · 2026 · 75 · 2668-2682

DOI: 10.1109/tr.2026.3712009 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Reliability, 2026. Verfasst von Lulu Wang und 4 weiteren Autor:innen. Die Arbeit nennt 57 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Pretrained programming language models (PLMs) have shown strong performance in software engineering tasks, but remain vulnerable to adversarial attacks. Traditional methods rely on narrow, static perturbation strategies, leading to poor diversity and a combinatorial explosion of the search space when multiple operations are applied. To address these limitations, we propose CodeRL-IA, an adversarial attack method that integrates reinforcement learning and importance analysis to dynamically optimize semantics-preserving perturbation strategies, achieving high attack effectiveness while maintaining code quality. Extensive experiments on code summarization, code translation, and defect detection tasks demonstrate that CodeRL-IA outperforms the baselines, with average attack success rates of 9.8% and 14.87% higher than those of the baseline methods, while keeping comparable code quality.

Zitieren

APA

Wang, L., Wang, X., Du, X., Chang, J., & Li, B. (2026). A Reinforcement Learning-Driven Adversarial Attack Methods With Dynamic Perturbation Optimization. IEEE Transactions on Reliability, 75, 2668-2682. https://doi.org/10.1109/tr.2026.3712009

Vancouver

Wang L, Wang X, Du X, Chang J, Li B. A Reinforcement Learning-Driven Adversarial Attack Methods With Dynamic Perturbation Optimization. IEEE Transactions on Reliability. 2026;75:2668-2682. doi:10.1109/tr.2026.3712009

BibTeX
@article{Wang2026,
  author = {Wang, Lulu and Wang, Xiaoyi and Du, Xiaoning and Chang, Jianming and Li, Bixin},
  title = {A Reinforcement Learning-Driven Adversarial Attack Methods With Dynamic Perturbation Optimization},
  journal = {IEEE Transactions on Reliability},
  year = {2026},
  volume = {75},
  pages = {2668--2682},
  doi = {10.1109/tr.2026.3712009}
}
Kostenlos starten →