功耗分析攻击中机器学习模型选择研究  

Research on Model Selection of Machine Learning in Power Analysis Attack

作　　者：段晓毅[1] 陈东 高献伟[1] 范晓红[1] 周玉坤[1] DUAN Xiaoyi;CHEN Dong;GAO Xianwei;FAN Xiaohong;ZHOU Yukun(Department of Electronics and Information Engineering,Beijing Electronic Science and Technology Institute,Beijing 100070,China)

机构地区：[1]北京电子科技学院电子信息工程系

出　　处：《计算机工程》2019年第11期144-151,158,共9页Computer Engineering

基　　金：国家自然科学基金(61701008);中央高校基本科研业务费专项资金(2017LG05)

摘　　要：根据密码芯片功耗曲线的特性,对支持向量机、随机森林、K最近邻、朴素贝叶斯4种机器学习算法进行分析研究,从中选择用于功耗分析攻击的最优算法。对于机器学习算法的数据选取问题,使用多组数量相同但组成元素不同的数据集的十折交叉验证结果进行模型选择,提高测试公平性及测试结果的泛化能力。为避免十折交叉验证过程中出现测试集误差不足以近似泛化误差的问题,采用Friedman检验及Nemenyi后续检验相结合的方法对4种机器学习算法进行评估,结果表明支持向量机是适用于功耗分析攻击的最优机器学习算法。According to the characteristics of the power curve of crypto chips,four kinds of machine learning algorithms including Support Vector Machine(SVM),Random Forest(RF),K-Nearest Neighbor(KNN)and Naive Bayes(NB)are analyzed and studied,and the optimal algorithm for power analysis attack is selected.For the data selection problem of the machine learning algorithm,the 10-fold cross-validation results of multiple sets of data sets with the same number but different constituent elements are used to select the model,which improves the test fairness and the generalization ability of the test results.In order to avoid the problem that the test set error is not enough to approximate the generalization error during the 10-fold cross-validation process,the four kinds of machine learning algorithms are evaluated by the combination of Friedman test and Nemenyi post-hoc test.The results show that the SVM is the optimal machine learning algorithm for power analysis attack.

关 键 词：机器学习 十折交叉验证 Friedman检验 Nemenyi后续检验 功耗分析攻击 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]