基于FCM的分布式学习方法  

Distribute learning algorithm based on fuzzy C-means

作　　者：吴静[1,2] 刘衍珩[1,2] 吕荣[1,2] 

机构地区：[1]吉林大学计算机科学与技术学院,长春130012 [2]吉林大学符号计算与知识工程教育部重点实验室,长春130012

出　　处：《吉林大学学报（工学版）》2010年第1期171-175,共5页Journal of Jilin University:Engineering and Technology Edition

基　　金：高等学校博士学科点专项科研基金项目(20060183043);国家自然科学基金项目(60573128)

摘　　要：在大规模网络入侵检测应用中,针对模块化神经网络算法学习精度高但效率相对较低的特点,采用SOM算法作为基本学习方法并利用其神经元竞争特性,引入模糊聚类FCM对SOM输出权值进行融合学习分类,在保持SOM高精度的基础上,大幅减少了学习时间。最后采用入侵检测数据集KDDCUP99进行测试,结果验证了算法的高检测率和较好的效率。In intrusion detection in large-scale network, the accuracy of modular neural network learning algorithm is high but its efficiency is relatively low. In this paper, Self-Organizing Mapping （SOM） is adopted as a basic learning method to take the advantage of competitive characteristics of its neurons. Then Fuzzy C-Means （FCM） is introduced to carry out fusion learning and classification of the output weights of SOM. This method can not only maintain high accuracy of the SOM but also significantly reduce the learning time. The proposed method is tested using KDDCUP99 dataset. Results show that the detection rate of this method is high and its efficiency is improved.

关 键 词：计算机应用 入侵检测 分布式学习 神经网络 模糊C均值(FCM) 

分 类 号：TP393.08[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]