基于向量投影的支持向量机增量算法  

Incremental Learning Algorithm of Support Vector Machine Based on Vector Projection

作　　者：张健沛 赵莹[1] 杨静[1] 

机构地区：[1]哈尔滨工程大学,哈尔滨150001

出　　处：《计算机科学》2008年第3期164-166,共3页Computer Science

基　　金：国家自然科学基金(60673131);黑龙江省自然科学基金(F2005-02)的资助

摘　　要：提出一种新的基于向量投影的支持向量机增量式学习算法。该算法根据支持向量的几何分布特点,采用向量投影的方法对初始样本及增量样本在有效地避免预选取失效情况下进行预选取。选取最有可能成为支持向量的样本形成边界向量集,并在其上进行支持向量机训练。通过对初始样本是否满足新增样本集KKT条件的判断,解决非支持向量向支持向量转化的问题,有效地处理历史数据。实验表明,基于向量投影的支持向量机增量算法可以有效地减少训练样本数,积累历史信息,提高训练的速度,从而具有更好的推广能力。A new incremental learning algorithm of support vector machine based on vector projection is proposed. The geometric character of the support vectors is used to extract the simples which can be the support vectors most possibly from the original simples and the incremental simples, avoiding the invalid per-extracting. The simples pre-extracted are used to generate the bound support vector set, on which the support vector machine training. It is used to solve the transferring between the support vectors and no-support vectors whether the original simples satisfy the KKT condition of the incremental simples, which can deal with the history data efficiently. The experiment shows that the algorithm can greatly reduce the number of training data in incremental SVM training, speeding up the training process, which has the better generalization performance.

关 键 词：增量算法 支持向量机 向量投影 预选取失效 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] O18[自动化与计算机技术—控制科学与工程]