《计算机应用研究》|Application Research of Computers

对抗样本生成及攻防技术研究

Survey of generation,attack and defense of adversarial examples

免费全文下载 (已被下载 次)  
获取PDF全文
作者 刘小垒,罗宇恒,邵林,张小松,朱清新
机构 电子科技大学 a.信息与软件工程学院;b.网络空间安全研究中心,成都 611731
统计 摘要被查看 次,已被下载
文章编号 1001-3695(2020)11-001-3201-05
DOI 10.19734/j.issn.1001-3695.2019.07.0252
摘要 基于对抗样本的攻击方法是机器学习算法普遍面临的安全挑战之一。以机器学习的安全性问题为出发点,介绍了当前机器学习面临的隐私攻击、完整性攻击等安全问题,归纳了目前常见对抗样本生成方法的发展过程及各自的特点,总结了目前已有的针对对抗样本攻击的防御技术,最后对提高机器学习算法鲁棒性的方法作了进一步的展望。
关键词 对抗样本; 机器学习; 深度学习
基金项目 国家自然科学基金资助项目(61572115)
四川省苗子工程创新基金资助项目(2019JDRC0069)
本文URL http://www.arocmag.com/article/01-2020-11-001.html
英文标题 Survey of generation,attack and defense of adversarial examples
作者英文名 Liu Xiaolei, Luo Yuheng, Shao Lin, Zhang Xiaosong, Zhu Qingxin
机构英文名 a.School of Information & Software Engineering,b.Center for Cyber Security,University of Electronic Science & Technology of China,Chengdu 611731,China
英文摘要 Attack methods based on adversarial samples are one of the security challenges that machine learning algorithms are commonly facing. This paper took the security of machine learning as a starting point, introduced the current security issues such as privacy attack and integrity attack that were faced by machine learning, summarized the development process and respective characteristics of current adversarial sample generation methods, and summarized the existing defense techniques for adversarial sample attacks, finally made a further look at how to improve the robustness of machine learning algorithms.
英文关键词 adversarial examples; machine learning; deep learning
参考文献 查看稿件参考文献
 
收稿日期 2019/7/23
修回日期 2019/9/11
页码 3201-3205,3212
中图分类号 TP393.04
文献标志码 A