基于深度学习的漏洞检测中样本集预处理的方法研究  

Study on method of sample set preprocessing in vulnerability detection based on deep learning

作　　者：王于叶 张皓天 许泽遥 Wang Yuye;Zhang Haotian;Xu Zeyao(Nanjing University of Posts and Telecommunications,Nanjing 210023,China)

机构地区：[1]南京邮电大学,江苏南京210023

出　　处：《无线互联科技》2020年第9期123-125,共3页Wireless Internet Technology

基　　金：国家大学生创新训练计划支持项目,项目编号:201810293024。

摘　　要：基于深度学习的二进制程序漏洞检测是漏洞检测研究的一种新趋势。目前基于深度学习的漏洞检测技术主要选用源代码或二进制代码作为样本进行分析处理。文章提出一种用Valgrind框架对二进制程序进行处理,从而得到中间语言代码作为深度学习样本的思路,并以该思路为核心开发出一种数据处理方法建立样本集,最后,选择Bi-LSTM深度学习算法进行对比试验,验证了该方法的效果。The detection of binary vulnerability based on deep learning is a new trend of vulnerability detection research.At present,the vulnerability detection technology based on deep learning mainly uses source code or binary code as samples for analysis and processing.In the article,we provided a idea to process binary program with Valgrind framework,and get intermediate language code as deep learning sample.Based on this idea,develop a data processing methods to establish sample set.Finally,choose Bi-LSTM deep learning algorithm to performed a comparative experiment,and the validity of the method has been verified.

关 键 词：深度学习 漏洞挖掘 中间语言 

分 类 号：TP309[自动化与计算机技术—计算机系统结构] TP18[自动化与计算机技术—计算机科学与技术]