自然语言处理中的对抗攻击与防御研究综述

Summary of Research on Adversarial Attack and Defense in Natural Language Processing

自2014年以来,对抗样本在计算机视觉领域已经得到了广泛的应用,并取得巨大的成功。近几年,对抗样本在自然语言处理(NLP)中的应用引起广泛的研究兴趣。与图像领域相比,应用于NLP任务的对抗攻击方法既有相似的思路,也有其不同的特点。从多个角度综述应用于NLP的对抗攻防的最新进展,并探讨可能的研究方向。

Adversarial examples have been used in computer vision since 2014,and achieved great success.In recent years,adversarial examples ap⁃plied to natural language processing(NLP)has attracted a wide range of research interests.Compared with it in image processing,the ad⁃versarial attack methods applied to NLP have both similar processing ideas and unique features.In this article,we give an overview to these recent advances on adversarial attack and defense in NLP from various perspectives,and discuss some possible research directions.