基于最近边界向量的SVM增量学习算法  

Incremental Learning Algorithm of Support Vector Machine Based on Nearest Border Vectors

作　　者：刘叶青[1] 刘三阳[2] 谷明涛 

机构地区：[1]河南科技大学数学与统技学院 [2]西安电子科技大学理学院 [3]解放军96251部队

出　　处：《数学的实践与认识》2011年第2期110-114,共5页Mathematics in Practice and Theory

基　　金：河南科技大学博士科研启动基金;河南科技大学青年基金(2008QN205)

摘　　要：为了减少求支持向量过程中二次规划的复杂度,利用训练样本集的几何信息,选出两类中离另一类最近的边界向量集合,它是样本中最有可能成为支持向量的一部分,用它代替原样本集进行训练.对新增样本,若存在违反KKT条件的样本,只对这部分新样本进行学习.同时找出原样本中可能转化为支持向量的非支持向量样本.基于分析结果,提出了一种新的基于最近边界向量的增量式支持向量机学习算法.对标准数据集的实验结果表明,算法是可行的,有效的.In order to reduce the time consumed in solving quadratic programming problems, a set of nearest border vectors were extracted from the training samples by using the geometric information in these samples.The original sample set was replaced by the obtained nearest border vector set in the process of training.The nearest border vector set is most likely to become the support vectors.For new samples,those was learned which do not satisfy Karush-Kuhn-Tucker（KKT） conditions.Besides support vectors,those was learned which maybe convet support vectors in the original samples.Based on the analysis results,a incremental learning algorithm of support vector machine（SVM） based on nearest border vectors is presented.The experimental results with the standard dataset indicate the effectiveness of the proposed algorithm.

关 键 词：支持向量机 增量算法 数据挖掘 分类 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]