【摘要】 作为现阶段人工智能研究的热点，对抗样本攻击方法已广为人知并处于一个不断发展的阶段。然而对抗样本的存在会给实际生活中的人工智能应用带来安全隐患。所以研究对抗样本的生成方式，研究对抗的原理以寻找对应的防御方法来提升深度学习图像分类任务的安全性迫在眉睫。论文首先介绍了对抗技术的相关知识，围绕当前图像对抗样本研究的发展趋势，列举并分析了一些重要的对抗算法，然后分析了现实场景下对抗算法的应用以及对抗防御的现状，最后总结了对抗技术现状并提出未来研究的发展方向。更多还原

【Abstract】 As a hot spot of artificial intelligence research at this stage,adversarial sample attack methods have been widely known and are in a continuous development stage. However,the existence of adversarial samples will bring security risks to artificial intelligence applications in real life. Therefore,it is imminent to study how adversarial samples are generated and to investigate the principles of adversarial samples to find defensive methods to improve the security of deep learning image classification tasks. This paper first begins with an introduction to the relevant knowledge of adversarial technology,enumerates and analyzes some important adversarial algorithms around the current development trend of image adversarial sample research,then analyzes the application of adversarial algorithms in real scenarios and the current state of adversarial defense,and finally summarizes the current technology and the development direction of future research.更多还原