标记样本的Adaboost算法  

An Adaboost Algorithm with Sample Marked

作　　者：郑诚[1] 张瑞[2] 陈娟娟[1] 

机构地区：[1]安徽大学计算智能与信号处理教育部重点实验室,安徽合肥230039 [2]安徽大学计算机科学与技术学院,安徽合肥230039

出　　处：《计算机技术与发展》2008年第7期109-111,115,共4页Computer Technology and Development

基　　金：国家自然科学基金资助项目(60475017);安徽省高等学校自然科学研究项目(2006kj055B)

摘　　要：提升(Boost)学习算法中,可以划分为多数提升和Adaboost两类。Adaboost是目前比较流行的分类方法,目前在多媒体和计算机视觉领域得到了广泛的应用。文中介绍了Adaboost方法的原理与方法,通过在提升过程中对训练集中部分样本的标记,提出了一种新的Adaboost算法的训练方法,并且用实验数据对该方法进行验证。该方法通过对前一轮提升后权值较小的那部分样本作标记,增加了后一轮提升抽样的有效容量,从而使算法中的分类器能够更快速地关注那些很难分类的样本。In multimedia and computer fields,Adahoost algorithm is the popular boost learning method for classification. It can be divided into two series：Boost by majority and Adaboost. The algorithm of Adaboost method was described in detail, A new training method has been raised through the marks on the part of training samples and proved by experiment. The former＇ s effective capacity of sample has been improved by the new algorithm, which is made on the former samples with less weights. Thus, the classification algorithm can focus on those samples faster which are hard to classify.

关 键 词：ADABOOST算法 提升 抽样 有效容量 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]