支持向量机的增量学习和减量学习  

Incremental and Decremental Learning with Support Vector Machine

作　　者：段华[1,2] 侯伟真[2] 贺国平[2] 廉文娟[2] 

机构地区：[1]上海交通大学数学系,上海200240 [2]山东科技大学信息学院,山东青岛266510

出　　处：《哈尔滨工程大学学报》2006年第B07期415-421,共7页Journal of Harbin Engineering University

基　　金：国家自然科学基金资助项目（10571109,60503002）.

摘　　要：分别介绍了支持向量机的增量学习和减量学习的两种训练方法，即在线递归训练法和最小二乘支持向量机．递归法只能处理在线（每次只处理一个样本）增量学习或减量学习，而最小二乘法即可处理在线又可处理成批增量学习或减量学习．递归法得到的解是精确的但是以时间为代价的，最小二乘法花费的时间少，但得到的解不如递归法的精确．并通过标准模式分类库中数据集进行数值试验比较．Two training methods for incremental learning and decremental learning of support vector machine are discussed, which are on-line recursive training algorithm and least squares support vector machine. On-line recursive training algorithm can only process on-line incremental learning and deremental learning, in which process only one vector at a time, while least squares support vector machine can process incremental learning and deremental learning for both on-line and batchs. The solution obtained by on-line recursive training algorithm is exact, but with the cost of time. The time spent by least squares support vector machine is short, but the solution obtained is not as exact as recursive training algorithm. The experiment has been done to prove the advantages and disadvantages for two training methods, which is based on the data set from Benchmark Repository.

关 键 词：支持向量机 增量学习 减量学习 最小二乘法 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]