基于改进型循环神经网络的恶意代码分类检测

近年来,随着恶意代码检测技术的提升,网络攻击者开始倾向构建能自重写和重新排序的恶意代码,以避开安全软件的检测。传统的机器学习方法是基于安全人员自主设计的特征向量来判别恶意代码,对这种新型恶意代码缺乏检测能力。为此,文中提出了一种新的基于代码时序行为的检测模型,并采用回声状态网络、最大池化和半帧结构等方式对神经网络进行优化。与传统的检测模型相比,改进后的模型对恶意代码的检测率有大幅提升。

Classification and detection of malicious code based on improved recurrent neural network

In recent years,with the improved alicious code detection technology,network attackers tend to build malicious code that can rewrite and reorder itself to avoid detection of security software.The traditional machine learning method is based on the feature vector designed by the security personnel to identify malicious code,so it is disable to detect this new type of malicious code.Therefore,a new detection model based on code timing behavior is proposed,and the neural network are optimized by means of echo state network,maximum pooling and half-frame structure.Compared with the traditional detection model,the improved model greatly improves the detection rate of malicious code.