基于填充函数法的光滑分段半监督支持向量机  

Smooth piecewise semi-supervised support vector machine based on filled function

作　　者：马斌奇[1] 王彦彦[2] 

机构地区：[1]西安邮电大学自动化学院,陕西西安710121 [2]西安邮电大学通信与信息工程学院,陕西西安710121

出　　处：《西安邮电大学学报》2015年第5期62-66,共5页Journal of Xi’an University of Posts and Telecommunications

基　　金：国家自然科学基金资助项目(61100165;61100231;61472307;51205309);陕西省自然科学基金资助项目(2014JM8313);陕西省教育厅科学研究计划资助项目(2013JK1023)

摘　　要：构造一个光滑分段函数,以逼近梯度直推式支持向量机模型中的不可微项,得到一种新的光滑分段半监督支持向量机模型。根据新模型的非凸特性,采用无参数填充函数法对其求解,以此避免参数调节的复杂过程,减小计算量.采用UCI数据库中不同规模的数据集,对新模型进行训练和测试,结果表明新模型可降低向量机的平均误分率,提高其分类性能。A smooth piecewise function is constructed to approximate the non-differentiable part of the gradient transductive support vector machine, and then, a new smooth semi-supervised support vector machine model is proposed. According to the non-convex character of the new model, a new parameter free filled function method can be used to solve the smooth non-convex programming, and thus the difficulties of parameter adjustment can be overcome, which makes the model simple and easy to implement. Several data sets of different size from the UCI database are used to train and test the new model. The results show that the error classification rate becomes more lower, and that the classification performance are improved.

关 键 词：半监督支持向量机 梯度直推式支持向量机 光滑分段函数 逼近性能 填充函数 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]