面向自然语言处理领域的对抗攻击研究与展望

Research and Prospect of Adversarial Attack in the Field of Natural Language Processing

随着人工智能的不断发展,深度学习已经被应用到各领域当中.然而,近年来有研究表明,深度学习易受到对抗攻击的影响,这些对抗攻击可以欺骗深度学习模型,使模型对样本类别产生错误判断.目前,有关计算机视觉对抗攻击的研究已经逐渐趋于成熟,而由于文本数据的结构性质比较特殊,自然语言处理领域的对抗攻击的研究还处于发展阶段.因此,通过介绍对抗攻击的概念及其在计算机视觉领域的应用,引出自然语言处理领域的对抗攻击的研究现状,并根据具体的自然语言处理下游任务进行分析,研究目前流行的对抗攻击方案.最后对自然语言处理领域的对抗攻击发展提出展望.对自然语言处理对抗攻击领域的研究人员具有参考价值.

With the continuous development of artificial intelligence, deep learning has been applied to various fields. However, in recent years, related research have shown that deep learning is susceptible to adversarial attack, which can trick deep learning models into misjudging sample classes. At present, the research of computer vision adversarial attack has gradually become mature, and due to the special structure of text data, the research of natural language processing adversarial attack is still in the development stage. Therefore, by introducing the concept of adversarial attack and its application in the field of computer vision, this paper introduces the current research status of adversarial attack in the field of natural language processing, and investigates popular adversarial attack schemes according to specific downstream tasks of natural language processing. Finally, prospects for the development of adversarial attack in the field of natural language processing are proposed. This paper has reference value for researchers in the field of natural language processing adversarial attack.