Learning General Gaussian Kernels by Optimizing Kernel Polarization  

作　　者：WANG Tinghua HUANG Houkuan TIAN Shengfeng DENG Dayong 

机构地区：[1]School of Computer and Information Technology, Beijing Jiaotong University, Beijing 100044, China [2]School of Mathematics and Information Engineering, Zhejiang Normal University, Jinhua 321004, China

出　　处：《Chinese Journal of Electronics》2009年第2期265-269,共5页电子学报（英文版）

摘　　要：The problem of model selection for Support vector machines （SVM） with general Gaussian kernels is considered. Unlike the conventional standard single scale Gaussian kernels, where all the basis functions have a common kernel width, the general Gaussian kernels adopt some linear transformations of input space such that not only the scaling but also the rotation is adapted. We pro- posed a gradient-based method for learning the optimal general Gaussian kernels by optimizing kernel polarization. This method can find a more powerful kernel for a given classification problem without designing any classifier. Experiments on both synthetic and real data sets demonstrate that tuning of the scaling and rotation of Gaussian kernels using our method can yield better generalization performance of support vector machines.

关 键 词：General Gaussian kernels Kernel polar- ization Support vector machines （SVM） Model selection. 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP278[自动化与计算机技术—计算机科学与技术]