EUVIMEDEUVIMED entdecken
Crossref

Strongly concealed adversarial attack against text classification models with limited queries

Yao Cheng, Senlin Luo, Yunwei Wan, Limin Pan, Xinshuai Li

Neural Networks · 2025

DOI: 10.1016/j.neunet.2024.106971 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Neural Networks, 2025. Verfasst von Yao Cheng und 4 weiteren Autor:innen. Beteiligte Einrichtung: School of Information and Electronics, Beijing Institute of Technology, Beijing 100081, PR China. Electronic address: chengyao_bit@126.com. Hauptthemen laut MEDLINE: Neural Networks, Computer, Computer Security. Die Arbeit wurde bisher 9-mal zitiert (Crossref) und nennt 39 Literaturangaben.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In black-box scenarios, adversarial attacks against text classification models face challenges in ensuring highly available adversarial samples, especially a high number of invalid queries under long texts. The existing methods select distractors by comparing the confidence vector differences obtained before and after deleting words, and the query increases linearly with the length of the text, making it difficult to apply to attack scenarios with limited queries. Generating adversarial samples based on a thesaurus can lead to semantic inconsistencies and even grammatical errors, making it easy for the target model to recognize adversarial samples and resulting in a low success rate of attacks. A parallel and highly stealthy Adversarial Attack against Text Classification Model (AdATCM) is proposed, which reinforces dual-task of attack and generation. This method does not require querying the target model during the selection of distractors. Instead, it directly uses contextual information to calculate the importance of words and selects distractors in one go, strengthening the concealment of attacks. Integrating KL divergence loss, cross entropy loss, and adversarial loss to construct an objective function for training an adversarial sample attack model, generating adversarial samples that can fit the original sample distribution and strengthen the success rate of attacks. The experimental results show that this method has a high success rate and strong concealment, effectively reducing the number of attack queries under long text conditions.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Neural Networks, ComputerComputer SecurityHumansSemanticsAlgorithms

Stichwörter der Autor:innen

Adversarial LossAdversarial SamplesDiffusion ModelMask-GuidedText Classification Model

Autor:innen und Institutionen

  • School of Information and Electronics, Beijing Institute of Technology, Beijing 100081, PR China. Electronic address: chengyao_bit@126.com.
  • School of Information and Electronics, Beijing Institute of Technology, Beijing 100081, PR China. Electronic address: luosenlin2019@126.com.
  • School of Information and Electronics, Beijing Institute of Technology, Beijing 100081, PR China. Electronic address: francishijack@gmail.com.
  • School of Information and Electronics, Beijing Institute of Technology, Beijing 100081, PR China. Electronic address: panlimin2022@gmail.com.
  • School of Information and Electronics, Beijing Institute of Technology, Beijing 100081, PR China. Electronic address: bfs_lxs@163.com.

Zitieren

APA

Cheng, Y., Luo, S., Wan, Y., Pan, L., & Li, X. (2025). Strongly concealed adversarial attack against text classification models with limited queries. Neural Networks. https://doi.org/10.1016/j.neunet.2024.106971

Vancouver

Cheng Y, Luo S, Wan Y, Pan L, Li X. Strongly concealed adversarial attack against text classification models with limited queries. Neural Networks. 2025. doi:10.1016/j.neunet.2024.106971

BibTeX
@article{Cheng2025,
  author = {Cheng, Yao and Luo, Senlin and Wan, Yunwei and Pan, Limin and Li, Xinshuai},
  title = {Strongly concealed adversarial attack against text classification models with limited queries},
  journal = {Neural Networks},
  year = {2025},
  doi = {10.1016/j.neunet.2024.106971}
}
Kostenlos starten →