期刊导航

论文摘要

非稀疏多核组合的支持向量回归方法

Non-sparseMultipleKernelLearningApproachforSupportVectorRegression

作者:胡庆辉(武汉大学计算机学院;桂林航天工业学院广西高校机器人与焊接技术重点实验室培育基地;武汉大学软件工程国家重点实验室);丁立新(武汉大学计算机学院;武汉大学软件工程国家重点实验室);刘晓刚(桂林航天工业学院广西高校机器人与焊接技术重点实验室培育基地);李照奎(武汉大学计算机学院;武汉大学软件工程国家重点实验室)

Author:HuQinghui(SchoolofComputer,WuhanUniv.;GuangxiCollegesandUniversitiesKeyLab.BreedingBaseof RobotandWeldingTechnol.,GuilinUniv.ofAerospaceTechnol.;StateKeyLab.ofSoftwareEng.,WuhanUniv.);DingLixin(SchoolofComputer,WuhanUniv.;StateKeyLab.ofSoftwareEng.,WuhanUniv.);LiuXiaogang(GuangxiCollegesandUniversitiesKeyLab.BreedingBaseof RobotandWeldingTechnol.,GuilinUniv.ofAerospaceTechnol.);LiZhaokui(SchoolofComputer,WuhanUniv.;StateKeyLab.ofSoftwareEng.,WuhanUniv.)

收稿日期:2014-08-28          年卷(期)页码:2015,47(4):91-97

期刊名称:工程科学与技术

Journal Name:Advanced Engineering Sciences

关键字:多核学习;支持向量回归;非稀疏核组合;两步优化

Key words:multiplekernellearning;supportvectorregression;non-sparsekernelscombination;two-stepoptimization

基金项目:国家自然科学基金资助项目(11301106);广西自然科学基金资助项目(2014GXNSFAA1183105);广西高校科研重点项目资助(ZD2014147); 桂林航天工业学院科研基金资助项目(Y12Z028)

中文摘要

为了改善支持向量回归机的性能,提出一种利用多核学习解决回归问题的算法(NS-MKR)。算法对基本核函数的组合系数施加了Lp范数的约束(p>1),以得到组合系数的非稀疏解,并采用了两步优化方法,首先求解基于加权组合核的标准支持向量回归问题,用于学习拉格朗日乘子,然后采用简单的计算,求得基本核函数的组合系数,这2个步骤交替进行,直到满足事先定义的收敛准则。在人工数据集和真实数据集上的实验表明,相对于传统的单核和稀疏多核支持向量回归方法,提出的算法有更好的泛化性能。

英文摘要

In order to improve the performance of support vector regression machine,a novel algorithm was proposed to solve the support vector regression problem with multiple kernel leaning method (NS-MKR).Lp-norm constraints were imposed on kernel weights withp>1to obtain the non-sparse solution.Two-step optimization method was used to solve the modal.Firstly,standard support vector regression based on a weighted mixture kernel was optimized to obtain Lagrange multipliers.Secondly,kernel weights were solved only with some simple calculation.These two steps were executed alternately until some predefined criterions were satisfied.Experiments on artificial datasets and real datasets showed that the proposed algorithm has better performance than the existing ones which based on single kernel or sparse multiple kernel leaning method.

关闭

Copyright © 2020四川大学期刊社 版权所有.

地址:成都市一环路南一段24号

邮编:610065