具有磁场效应的p-间隔核学习机  

p-margin kernel learning machine with magnetic field effect

作　　者：陶剑文[1,2] 王士同[1] 

机构地区：[1]江南大学信息工程学院,江苏无锡214122 [2]浙江工商职业技术学院信息工程学院,浙江宁波315012

出　　处：《控制与决策》2011年第4期623-628,共6页Control and Decision

基　　金：国家自然科学基金项目(60975027,60903100);宁波市自然科学基金项目(2009A610080)

摘　　要：针对模式分类问题,提出一种具有磁场效应的p-间隔核学习机(p-MKLM),旨在寻求一个具有磁场效应的最优超平面,受其吸引,使得一类模式离该平面的距离尽可能的小,而另一类模式受其排斥,离该平面的间隔尽可能的大,从而最大可能地实现模式分类.通过引入一个可调节的磁场强度q,减小一类模式的数据分布,从而提升分类性能.分别采用人工数据和实际数据进行实验,所得结果显示,p-MKLM在一类和二类模式分类上的性能均优于或等同于相关方法.A p-margin kernel learning machine（p-MKLM） with magnetic field effect is proposed for pattern classification problem in this paper.The basic idea is to find a optimal superplane with magnetic field effect such that the distance between one class and the hyperplane is as small as possible due to the magnetic attractive effect,while at the same time the margin between the hyperplane and the other classes is as large as possible due to magnetic repulsion,thus implementing pattern classification as much as possible.Moreover,a magnetic field density q is introduecd to compact the data distribution of one class,thus improving the classification performance of p-MKLM even more.Exprimental results obtained with synthetic and real data show that the proposed algorithms are effective and competitive to other related diagrams in such cases as two-class and one-class pattern classification respectively.

关 键 词：磁场效应 模式分类 新奇检测 支持向量机 核方法 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]