基于中心凸包算法与增量学习的SVM算法研究

Study on SVM Algorithm using Center Convex Vector and Incremental Learning

基于计算几何理论,在分析支持向量与凸包向量关系的基础上,提出了一种基于中心凸包算法与增量学习的SVM学习算法。在确保分类器达到可靠精度的前提下,为解决学习中时耗过长的问题,在对当前训练集计算凸包的基础上采用欧式中心距离淘汰法对训练样本进一步精简,并且每次进行增量学习的样本都包含前次训练样本集中违背KKT条件的样本,在UCI数据库上进行算法对比实验,结果表明算法的可行性和有效性。

Based on analysis of the support vector geometry significance and convex vector relationship, a learning algorithm of fast incremental convex vector algorithm based on SVM is proposed. In order to enhance training speed while SVM incremental learning accuracy is ensured, the training set is acquired based on computing the convex vector of the center distance method in the training sample set, and each incremental training sample contains the last training sample set against the KKT conditions of the sample, various algorithms then are used for comparison experiment under the UCI standard database, the result shows the feasibility and validity of this algorithm.