集成学习的能量分析攻击研究  

Research on Energy Analysis Attacks Based on Integrative Learning

作　　者：范晓红[1] 黄烨 韩文楷 FAN Xiaohong;HUANG Ye;HAN Wenkai(Beijing Electronic Science and Technology Institute,Beijing 100070,P.R.China)

机构地区：[1]北京电子科技学院,北京市100070

出　　处：《北京电子科技学院学报》2023年第2期20-29,共10页Journal of Beijing Electronic Science And Technology Institute

基　　金：中央高校基本科研业务费专项资金(项目编号:328202207)。

摘　　要：在针对密码芯片的攻击中,能量分析攻击是一种行之有效的攻击方式。传统的能量分析攻击以差分能量分析攻击(Differential Power Analysis,DPA)、相关功耗分析攻击(Correlation Power Analysis,CPA)、模板攻击(Template Attacks,TA)为主。随着科技的进步,机器学习迈入时代舞台。越来越多的能量分析攻击也开始基于机器学习实现,而集成学习是机器学习中一种有效的方法。集成学习可以综合多种机器学习算法,产生效果更优的能量分析攻击。本文首先对能量分析攻击及机器学习算法进行研究;其次,实现了基于支持向量机、随机森林等多种机器学习的能量分析攻击;最后,利用集成学习投票法生成组合攻击模型,并对各种攻击模型进行测试比对。实验结果表明,集成学习效果在绝大多数情况下,优于单一机器学习攻击效果;不同的集成学习组合效果不同,当机器学习算法原理相近时,会导致集成学习效果不佳。Energy analysis attack is an effective attack method against cryptographic chips and tradition-al energy analysis attacks mainly include the DPA(Differential Power Analysis),the CPA(Correlation Power Analysis),and the TA(Template Attacks).With the technological advancement,machine learning becomes in focus,based on which,an increasing number of energy analysis attacks are implemented.Integrative learning is an effective method in machine learning,which could synthesize multiple machine learning algorithms to achieve better energy analysis attacks.In this paper,energy analysis attacks and machine learning algorithms are first studied,and energy analysis attacks based on multiple machine learning algorithms are then implemented,including vector machines,ran-dom forests,etc.Finally,integrative learning voting method is utilized to generate combined attack models,which are tested and compared experimentally.Experiment results show that in most cases,integrative learning achieves better attack effect than single machine learning.Different integrative learning combinations exhibit different effects,and poor effect occurs in the case of similar machine learning algorithms.

关 键 词：机器学习 集成学习 能量分析攻击 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]