一种基于Vague-Sigmoid核的支持向量机研究  

A Novel SVM Based on Vague-sigmoid Kernel

作　　者：张永[1] 迟忠先[2] 李克秋[2] 

机构地区：[1]辽宁师范大学计算机系,辽宁大连116029 [2]大连理工大学计算机科学与工程系,辽宁大连116024

出　　处：《小型微型计算机系统》2008年第7期1277-1281,共5页Journal of Chinese Computer Systems

基　　金：国家科技型中小企业技术创新基金项目(05C26212120357)资助

摘　　要：Sigmoid核最初起源于神经网络,目前在支持向量机中也得到了广泛应用,但由于核矩阵的非半正定性,其应用受到一些限制.研究表明Sigmoid核可以用简单的模糊三角隶属函数来近似替代,使得其学习效率能进一步提高.本文首先分析模糊支持向量机的特性,将模糊理论用于支持向量机的核中,并在此基础上提出了基于Vague-Sigmoid核函数的支持向量分类器.该方法充分结合了Vague集的自身优势,用基于Vague集的相似度量来代替了常规中的样本间的点积计算方法.将文中提出的方法应用于标准数据集中,并与传统的Sigmoid核方法、Fuzzy-Sigmoid核方法进行了实验分析,实验表明文中提出的方法在不损失精度的情况下,能较好的提高算法的执行效率,取得了较好的实验结果;同时也表明在支持向量机中能利用Vague-Sigmoid核取代替传统的Sigmoid核,从而减少对Sigmoid核的限制.Support vector machine （SVM）, proposed by Vapnik based on statistical learning theory （SLT）, is a novel machine learning method which has been applied to many application fields successfully. The sigmoid kernel was quite popular for sup- port vector machines due to its origin from neural networks. Although it is known that the kernel matrix may not be positive semi-definite （PSD）, the sigmoid kernel matrix is conditionally positive definite （CPD） in certain parameters and thus is valid kernel there. Research shows that the sigmoid activation function can be approximated by using a simple sigmoid-like nonlinear activation function, which would improve its learning efficiency. The slgmold kernel is combined with fuzzy logic methodology first, which makes the computation of SVM simple and of ease implementation in hardware. Cornpared with Camps-Valls＇s fuzzy sigmoid kernel, this paper propose the vague sigmoid kernel based on similarity measure. The proposed method uses the characteristic of vague set, and replaces the traditional inner product with vague set similarity measure between training sampies. The experiments show that the average CPU time can be decreased markedly, in spite of small decrease in prediction precision. Results suggest that further and elaborated versions of the fuzzy tanh kernel could provide a SVM framework with a sigmoidal global kernel to overcome the current limitations of the sigmoid kernel.

关 键 词：支持向量机 模糊核 VAGUE集 分类器 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]