摘要
操作系统漏洞长期威胁用户隐私安全,给用户带来的不仅是经济损失,更有个人隐私泄漏等威胁。基于操作系统的内存管理、进程管理、IO管理三大核心功能,分析针对各功能的攻击方案的表现形式,提出利用深度学习技术,搭建包含基础数据层、特征提取层、特征学习层、防御行为层的操作系统多维度安全保护架构。基础数据层收集操作系统运行数据,并将原始数据提供给特征提取层;特征提取层对原始数据进行变换,向特征学习层提供易于理解与学习的特征;特征学习层借助深度神经网络,理解特征提取层提供的数据,并将检测结果反馈给防御行为层;防御行为层产生防御行为,并优化特征学习层,降低误判概率,提升检测精度。该架构通用性强,可部署性良好,能检测到未曾发现的攻击。若辅以GPU硬件支持,神经网络的算力资源开销将降低80%。
Operating system vulnerabilities threaten users’privacy security for a long time,bringing not only economic losses but also privacy leakage to users.Based on the three core functions of the operating system,i.e.memory management,process management and IO management,the expression forms of attack schemes are analyzed for each function,so as to build a multi-dimensional operating system security protection architecture including basic data layer,feature extraction layer,feature learning layer and defensive behavior layer by using deep learning technology.The basic data layer collects the operation data of the operating system and provides the original data to the feature extraction layer;the feature extraction layer transforms the original data and provides the features that are easy to understand and learn to the feature learning layer;the feature learning layer uses the deep neural network to understand the data provided by the feature extraction layer and feedback the detection results to the defense behavior layer;the defense behavior layer generates defense behavior,and optimize the feature learning layer,as to reduce the probability of misjudgment and improve the detection accuracy.Such architecture has strong generality,good deployability,and can detect previous undetected attacks.If supported by the GPU hardware,the computational cost of neural network will be reduced by 80%.
作者
曹玉红
陈思羽
CAO Yu-hong;CHEN Si-yu(The Chinese Institute of Electronics,Beijing 100036,China)
出处
《工业技术创新》
2021年第1期90-95,共6页
Industrial Technology Innovation
关键词
操作系统
多维度安全保护
深度学习
可部署性
GPU
神经网络
Operating System
Multi-dimensional Security Protection
Deep Learning
Deployability
GPU
Neural Network
