摘要
提出了一种基于拆分、组装神经网络的入侵检测方法。神经网络克服了以前在网络训练中易出现的训练时间过长、陷入局部极小的问题,使网络训练效率大大提高。该方法以网络数据源为主,同时考虑反映主机性能的数据源,并对特征数据进行了预处理,利用改进算法的学习能力和快速识别能力,实现了对用户行为的检测,尤其是在识别以前没有观察到的未知攻击方面具有较好性能。
This paper describes new neural network-based intrusion detection approach. This kind of neural network solved the problem of training's long time, trapped in the state of local minimum which exist in the former network training. It improved the efficiency of network training. Main in the network data source, also take the data source, which reflect host performance into consideration. It is based on the feature data that should be pretreated, the approach employs improving arithmetic that provides the capability of learning and quick classification, and can be used for detection of user behavior. In particular, the approach gained good performance in recognizing future unseen attacks.
出处
《武汉理工大学学报》
EI
CAS
CSCD
北大核心
2005年第1期83-86,共4页
Journal of Wuhan University of Technology
基金
国家"十五"重点科技攻关项目(2001BA307B010201).
关键词
入侵检测
神经网络
特征提取
拆分和组装
intrusion detection
neural networks
feature extraction
split and assembling
