机构地区: 暨南大学计算机科学系
出 处: 《控制与决策》 2005年第10期1129-1132,1136,共5页
摘 要: 经典的支持向量机(SVM)算法在求解最优分类面时需求解一个凸二次规划问题,当训练样本数量很多时,算法的速度较慢,而且一旦有新的样本加入,所有的训练样本必须重新训练,非常浪费时间.为此,提出一种新的SVM快速增量学习算法.该算法首先选择那些可能成为支持向量的边界向量,以减少参与训练的样本数目;然后进行增量学习.学习算法是一个迭代过程,无需求解优化问题.实验证明,该算法不仅能保证学习机器的精度和良好的推广能力,而且算法的学习速度比经典的SVM算法快,可以进行增量学习. A kind of algorithm for support vector machine (SVM) is proposed, which can train SVM fast and incrementally. The new algorithm selects border vectors which may be support vectors, so as to reduce training samples and advance training speed. Then an incremental algorithm is presented to train SVM by using the selected border vectors. Experiment results show that the algorithm not only acquirs the same precision with that of the classical algorithms, but also is faster than that of the classical algorithms.
领 域: [自动化与计算机技术] [自动化与计算机技术]