项目名称: 基于学习的智能化漏洞挖掘关键技术研究
项目编号: No.61772308
项目类型: 面上项目
立项/批准年度: 2018
项目学科: 自动化技术、计算机技术
项目作者: 张超
作者单位: 清华大学
项目金额: 17万元
中文摘要: 软件安全漏洞是网络空间攻防双方共同关心的战略资源之一,因而漏洞挖掘技术的研究意义重大。当下最流行的漏洞挖掘方案大多基于程序分析技术,如静态分析、模糊测试及符号执行等。然而,这些技术面临着严重的瓶颈问题,如路径爆炸等,制约了其漏洞挖掘的效率。此外,自动化漏洞挖掘技术方兴未艾,吸引了大批研究人员的关注,但是现有方案的智能化程度较低,仍存在很大的提升空间。针对这些需求,本项目将在申请人前期基于程序分析的漏洞挖掘及自动化漏洞挖掘研究工作的基础上,研究基于学习的智能化漏洞挖掘关键技术,尝试突破程序分析技术(如符号执行和模糊测试)的瓶颈;重点从面向漏洞挖掘的深度学习方法、漏洞数据采集、漏洞数据预处理三个方面展开研究,设计并实现一个智能化漏洞挖掘系统原型,提高安全测试的路径覆盖率并提升漏洞挖掘效率,形成一个可支撑未来更多研究的漏洞数据集,最终实现提升软件和系统安全性的目的。
中文关键词: 漏洞挖掘;深度学习;符号执行;强化学习;模糊测试
英文摘要: Software vulnerabilities are one of the most important resources in cyber-space. It is critical to study on finding vulnerabilities in software. State-of-art vulnerability discovery solutions mostly focus on program analysis, including static analysis, fuzzing and symbolic execution. However, they are facing critical challenges, e.g., the path explosion problem, and are not effective to find vulnerabilities in many cases. Moreover, automated vulnerability discovery drew many researchers' attentions. But state-of-art automated solutions are not intelligent, and could be improved. So, in this project, we focus on studying learning-based intelligent vulnerability discovery solutions, based on our previous work on vulnerability discovery. More specially, we will perform research on 1) learning algorithms towards vulnerability discovery, 2) vulnerability data collection, 3) vulnerability data pre-processing. With these efforts, we try to build a prototype of intelligent vulnerability discovery solution, breaking the bottlenecks of existing vulnerability discovery solutions, and build a set of vulnerability data for future research.
英文关键词: Vulnerability Discovery;Deep Learning;Symbolic Execution;Reinforcement Learning;Fuzzing