基于超球支持向量机的类增量学习算法研究  

Study on Class Incremental Learning Algorithm Based on Hyper-sphere Support Vector Machines

作　　者：秦玉平[1] 李祥纳[2] 王秀坤[1] 王春立[1] 

机构地区：[1]大连理工大学电子与信息工程学院,大连116024 [2]渤海大学信息科学与工程学院,锦州121000

出　　处：《计算机科学》2008年第8期116-118,共3页Computer Science

基　　金：国家自然科学基金项目(No.60603023);国家基础研究重大项目(973)研究专项(No.2001CCA00700)

摘　　要：提出了一种超球支持向量机类增量学习算法。对每一类样本,利用超球支持向量机在特征空间中求得包围该类尽可能多样本的最小超球,使各类样本之间通过超球隔开。类增量学习过程中,只对新增类样本进行训练,使得该算法在很小的样本集、很小的空间代价下实现了类增量学习,大大降低了训练时间,同时保留了历史训练结果。分类过程中,通过计算待分类样本到各超球球心的距离判定其所属类别,分类简单快捷。实验结果证明,该算法不仅具有较高的训练速度,而且具有较高的分类速度和分类精度。A new class incremental learning algorithm based on hyper-sphere support vector machines is proposed in this paper. For every class, hyper-sphere support vector machine is used to get the smallest hyper-sphere in feature space that contains most samples of a class, which can divide the class samples from others. In the process of class incre- mental learning, only are the samples that belong to the new incremental class trained. Therefore, the class incremental learning can be realized in a small set of samples and a small memory space. The training time is reduced largely; be sides, the history results of the classes that have nothing to do with the incremental class are saved at the same time. For the sample to be classified, the distances from it to the centre of every hyper-sphere are used to confirm the class that the sample belongs to. The classification method is simple and the speed is fast. The experiment results show that the algorithm has a higher performance on training speed, classification speed, and classification precision.

关 键 词：支持向量机 类增量学习 超球 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] O174.56[自动化与计算机技术—控制科学与工程]