图像分类中的对抗补丁研究综述

近年来,深度神经网络模型的安全性与鲁棒性成为了备受关注的重要问题。从前人们常探讨的对抗样本攻击,在物理世界中真正实施攻击的能力较弱,这促使了对抗补丁攻击的出现。文中介绍了图像分类模型中典型的对抗补丁攻击与防御方法相关研究,对其中的特点进行了分析,最后总结了对抗补丁研究中仍存在的主要挑战,并对未来的研究方向进行了展望。

Review of Adversarial Patches for Image Classification

In recent years, the security and robustness of deep neural network models have become an important issue of great interest. The adversarial sample attack, which was often explored before, is actually weak in implementing the attack in the physical world, which has prompted the emergence of adversarial patch attacks. This paper presents research related to typical adversarial patch attacks and defense methods in image classification, analyzes the characteristic. Finally, we will summarize the main challenges that still exist in adversarial patch research and provides an outlook on future research directions.