基于支持向量机的增量学习算法  

Incremental Learning Algorithm Based on Support Vector Machine

作　　者：李永丽[1,2] 刘衍珩[1] 肖见涛[2] 李向涛[2] 关伟洲[2] 

机构地区：[1]吉林大学计算机科学与技术学院,长春130012 [2]东北师范大学计算机科学与信息技术学院,长春130117

出　　处：《吉林大学学报（理学版）》2010年第3期464-467,共4页Journal of Jilin University:Science Edition

基　　金：国家科技支撑计划项目基金(批准号:3G008G341421)

摘　　要：通过对支持向量机KKT条件和样本间关系的研究,分析了新增样本加入训练集后支持向量的变化情况,提出一种改进的Upper Limiton Increment增量学习算法.该算法按照KKT条件将对应的样本分为3类:位于分类器间隔外,记为RIG;位于分类间隔上,记为MAR;位于分类间隔内,记为ERR.并在每次训练后保存ERR集,将其与下一个增量样本合并进行下一次训练.实验证明了该算法的可行性和有效性.The relationship between KKT conditions and the studied sample and the analysis of the change of support vector after the addition of incremental samples to the training set on the basis of an improved Upper Limit on Increment incremental learning algorithm. According to the KKT conditions for this algorithm the corresponding samples were divided into three categories：the RIG distributed outside the interval of classifier; the MAR at intervals on the classification and the ERR inside the intervals of classification. And ERR set after each training was preserved and combined with the incremental sample of the next training. Experiments show that the algorithm is feasible and effective.

关 键 词：支持向量 增量学习 支持向量机(SVM) 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]