反向梯度优化深度学习的病毒数据对抗方法  

Virus data countermeasure method using reverse gradient optimization and deep learning

作　　者：赵荷[1] 盖玲[2] ZHAO He;GAI Ling(Department of Computer Science and Engineering,Chengdu Neusoft University,Chengdu 611844,China;School of Management,Shanghai University,Shanghai 200444,China)

机构地区：[1]成都东软学院计算机科学与工程系,四川成都611844 [2]上海大学管理学院,上海200444

出　　处：《计算机工程与设计》2020年第6期1575-1580,共6页Computer Engineering and Design

基　　金：四川省科技厅重大科技专项基金项目(18ZDZX0078)。

摘　　要：为解决传统病毒攻击样本生成算法存在适用范围较小的问题,提出一种基于反向梯度优化深度学习算法的数据病毒攻击样本生成方法。通过自动微分计算对应感兴趣服务的梯度,为降低算法复杂度并提高算法适用范围,采用反向梯度算法对学习过程进行反转以训练神经网络。为验证所提方法的有效性,设置垃圾邮件过滤、恶意软件检测和手写数字识别3个数值算例,实验结果表明,所提算法具有较好的适用性。To solve the problem that the traditional virus attack sample generation algorithm has a small scope of application,a poisoning data sample generation method for deep learning algorithms based on reverse gradient optimization was proposed.The gradient of service of interest was calculated by automatic differentiation.To reduce the complexity of the algorithm and improve the application scope of the algorithm,the inverse gradient algorithm was used to reverse the learning process to train the neural network.To verify the effectiveness of the proposed method,three numerical examples of spam filtering,malware detection and handwritten numeral recognition were set up.The results show that the proposed algorithm has good applicability.

关 键 词：反向梯度优化 深度学习算法 数据病毒样本生成 对抗性算法训练 双层优化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]