摘要
主要研究支持向量机方法与聚类算法的配合问题.支持向量机的训练代价太大,如果直接把成千上万个特征向量直接用作训练,运算时间难以忍受.采取的策略是用聚类算法获得较少的聚类中心,然后将聚类中心作为支持向量机的训练样本.事实上,这样的组合方式有待改进.每一聚类的样本数有多有少,所以每一个聚类中心所体现出来的权重不一样.反映在支持向量机的算法中,改进思路为:在支持向量机的训练中,除了原有点以外,加入人工样本点,人工样本点的位置就是这些原有点之一,各个位置的数量与聚类大小成比例.
In this thesis,we studies the cooperation of support vector machines and clustering method.The cost of the training of support vector machine is too much,if we use thousands of support vectors directly,the computation time would be too long.The strategy of this thesis is to get fewer clustering center points,then take the clustering points as the train sample of support vector machine.In fact,such method is not perfect.The sample number of each cluster is not the same,therefore the weight of each clustering center point is different.In the support vector machine method,we have two issues to improve.The issue is,we don't take each point as the same,and fix the classify error of objective function with the point weight.
出处
《广西师范大学学报（自然科学版）》
CAS
2003年第A01期66-69,共4页
Journal of Guangxi Normal University:Natural Science Edition
关键词
统计学习理论
支持向量机
聚类
statistical learning theory
support vector machines
clustering
