Scholarlink

This chapter introduces the relevant concepts of adversarial attacks and presents related examples. It elaborates in detail on the principles of three common types of attack algorithms, namely optimization-based, gradient-based, and generation-based algorithms. Moreover, it explains the principles of common defense algorithms such as defense distillation, adversarial training, denoising networks, and adversarial sample detectors. Subsequently, in terms of the application of GAN attacks, three GAN-based adversarial sample generation models are introduced, including Perceptual-Sensitive GAN, Natural GAN, and AdvGAN. Regarding the defense of GANs, taking APE-GAN as an example, its defensive effect is demonstrated. Finally, a practical introduction to the widely used open-source tool AdvBox is provided.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Adversarial Attack

  • 章节

  • 发表日期：

    2026年

查看全文
  • Peng Long,
  • Xiaozhou Guo

摘要

This chapter introduces the relevant concepts of adversarial attacks and presents related examples. It elaborates in detail on the principles of three common types of attack algorithms, namely optimization-based, gradient-based, and generation-based algorithms. Moreover, it explains the principles of common defense algorithms such as defense distillation, adversarial training, denoising networks, and adversarial sample detectors. Subsequently, in terms of the application of GAN attacks, three GAN-based adversarial sample generation models are introduced, including Perceptual-Sensitive GAN, Natural GAN, and AdvGAN. Regarding the defense of GANs, taking APE-GAN as an example, its defensive effect is demonstrated. Finally, a practical introduction to the widely used open-source tool AdvBox is provided.