摘要
针对传统支持向量机无法适应大规模问题,通过引入加权线性损失函数,取代标准支持向量机的Hinge损失,提出一种加权线性损失支持向量机WLSVM(Weighted Linear Loss Support Vector Machine)。它的主要方法是:(1)通过对线性损失增加权重,提出对不同位置上的训练点给出不同惩罚,在一定程度上避免了过度拟合,增强了泛化能力。(2)仅需计算非常简单的数学表达式就可获得分类超平面,且方便解决大规模问题。通过在合成和真实数据集上的试验,结果表明:WLSVM的分类精度高于SVM和LSSVM,且减少了计算时间,尤其对于大规模问题。
In view of that traditional support vector machine( SVM) cannot adapt to large-scale problems,by introducing the weighted linear loss function to replace the Hinge loss function in standard support vector machine,the paper proposes a weighted linear loss support vector machine( WLSVM). Its main features are:( 1) by adding the weight on linear loss,it suggests giving different penalties on training points at different positions,thus avoids over fitting to a certain extent and enhances the generalisation ability.( 2) it can obtain the classified hyperplane by just calculating every simple mathematical expression,and is convenient in solving large-scale problems. Through the experiments on both synthetic and real datasets it is shown that the classification accuracy of WLSVM is higher than that of SVM and LS-SVM,and the computation time is reduced as well,especially for large-scale problems.
出处
《计算机应用与软件》
CSCD
2015年第12期114-117,共4页
Computer Applications and Software
基金
国家自然科学基金项目(61272500)
关键词
模式识别
支持向量机
线性损失
加权因子
大规模问题
Pattern recognition
Support vector machine
Linear loss
Weighed coefficient
Large-scale problems