基于Help-Training的半监督支持向量回归  

Semi-supervised support vector regression based on Help-Training

作　　者：程玉虎[1] 冀杰[1] 王雪松[1] 

机构地区：[1]中国矿业大学信息与电气工程学院,江苏徐州221116

出　　处：《控制与决策》2012年第2期205-210,226,共7页Control and Decision

基　　金：国家自然科学基金项目(60804022;60974050;61072094);教育部新世纪优秀人才支持计划项目(NCET-08-0836);教育部博士点基金项目(20110095110016);霍英东NCET-10-0765青年教师基金项目(121066)

摘　　要：提出一种基于Help-Training的半监督支持向量回归算法,包含最小二乘支持向量回归(LS-SVR)和近邻(NN)两种类型学习器.主学习器LS-SVR通过选择高置信度的未标记样本加以标记,并将其添加到已标记样本集,使训练样本的规模不断扩大,以提高LS-SVR的函数逼近性能.辅学习器NN用以协助LS-SVR从训练样本比较密集的区域选取未标记样本加以置信度评估,可以减弱噪声对学习效果的负面影响.实验结果表明所提算法具有良好的回归估计性能,学习精度较高.A semi-supervised support vector regression based on Help-Training is proposed,which includes two kinds of learners： a least squares support vector regression（LS-SVR） and a-nearest neighbor（NN）.As a main learner,the LSSVR chooses unlabeled samples with the highest confidence to label and adds these samples to the labeled sample set,which is repeated for given iterations to enlarge the scale of the training samples so as to improve the property of function approximation of the LS-SVR.As an auxiliary learner,the NN is used to help the LS-SVR choose unlabeled samples to evaluate confidence from a high-density region of training samples,which can weaken the negative influence of noise on the learning performance of the LS-SVR.Experimental results show that the Help-Training LS-SVR has advantages of good regression performance and high learning accuracy.

关 键 词：半监督学习 助训练 支持向量回归 近邻 置信度 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]