首页 | 本学科首页   官方微博 | 高级检索  
     

多分类孪生支持向量机研究进展
引用本文:丁世飞,张健,张谢锴,安悦瑄.多分类孪生支持向量机研究进展[J].软件学报,2018,29(1):89-108.
作者姓名:丁世飞  张健  张谢锴  安悦瑄
作者单位:中国矿业大学 计算机科学与技术学院, 江苏 徐州 221116;中国科学院计算技术研究所 智能信息处理重点实验室, 北京 100190,中国矿业大学 计算机科学与技术学院, 江苏 徐州 221116,中国矿业大学 计算机科学与技术学院, 江苏 徐州 221116,中国矿业大学 计算机科学与技术学院, 江苏 徐州 221116
基金项目:国家自然科学基金(61379101,61672522);国家重点基础研究发展计划(973)(2013CB329502)
摘    要:孪生支持向量机因其简单的模型、快速的训练速度和优秀的性能而受到广泛关注.该算法最初是为解决二分类问题而提出的,不能直接用于解决现实生活中普遍存在的多分类问题.近来,学者们致力于将二分类孪生支持向量机扩展为多分类方法并提出了多种多分类孪生支持向量机.多分类孪生支持向量机的研究已经取得了一定的进展.本文主要工作是回顾多分类孪生支持向量机的发展,对多分类孪生支持向量机进行合理归类,分析各个类型的多分类孪生支持向量机的理论和几何意义.本文以多分类孪生支持向量机的子分类器组织结构为依据,将多分类孪生支持向量机分为:基于“一对多”策略的多分类孪生支持向量机、基于“一对一”策略的多分类孪生支持向量机、基于“一对一对余”策略的多分类孪生支持向量机、基于二叉树结构的多分类孪生支持向量机和基于“多对一”策略的多分类孪生支持向量机.基于有向无环图的多分类孪生支持向量机训练过程与基于“一对一”策略的多分类孪生支持向量机类似,但是其决策方式有其特殊的优缺点,因此本文将其也独立为一类.本文分析和总结了这六种类型的多分类孪生支持向量机的算法思想、理论基础.此外,还通过实验对比了分类性能.本文工作为各种多分类孪生支持向量机之间建立了联系比较,使得初学者能够快速理解不同多分类孪生支持向量机之间的本质区别,也对实际应用中选取合适的多分类孪生支持向量机起到一定的指导作用.

关 键 词:多分类  孪生支持向量机  多生支持向量机  支持向量机
收稿时间:2017/1/19 0:00:00
修稿时间:2017/4/16 0:00:00

Survey on Multi Class Twin Support Vector Machines
DING Shi-Fei,ZHANG Jian,ZHANG Xie-Kai and AN Yue-Xuan.Survey on Multi Class Twin Support Vector Machines[J].Journal of Software,2018,29(1):89-108.
Authors:DING Shi-Fei  ZHANG Jian  ZHANG Xie-Kai and AN Yue-Xuan
Affiliation:School of Computer Science and Technology, China University of Mining and Technology, Xuzhou 221116, China;Key Laboratory of Intelligent Information Processsing, Institute of Computing Technology, The Chinese Academy of Sciences, Beijing 100190, China,School of Computer Science and Technology, China University of Mining and Technology, Xuzhou 221116, China,School of Computer Science and Technology, China University of Mining and Technology, Xuzhou 221116, China and School of Computer Science and Technology, China University of Mining and Technology, Xuzhou 221116, China
Abstract:Twin support vector machines have drawn extensive attention because of the simple model, high training speed and good performance. The initial twin support vector machine is designed for binary classification. However, multi class classification problems are also common in practice. In recent years, researchers have devoted themselves to the study of multi class twin support vector machines. Various mulit class twin support vector machines have been proposed. The study of multi class twin support vector machines has made great progress. This paper aims to review the development of multi class twin support vector machines, group them reasonably and analyze them with the respect to the basic theories and geometric meanings. According to the structures, this paper divides them into the following groups:"one-versus-all" strategy based multi class twin support vector machines, "one-versus-one" strategy based multi class twin support vector machines, binary tree based multi class twin support vector machines, "one-versus-one-versus-rest" strategy based multi class twin support vector machines and "all-versus-one" strategy based multi class twin support vector machines. Although the training processes of direct acyclic graph based multi class twin support vector machines are much similar with that of "one-versus-one" based approachs, the decision processes have their own characteristics and disadvantages, so they are divided into a separate group. This paper analyzes and summarizes the ideas and theories of different multi class twin support vector machines and presents experimental results to compare the performances. This paper can make it easy for novices to understand the essential differences and help to choose the suitable multi class twin support vector machine for a practical problem.
Keywords:multi classs classification  twin support vector machine  multiple birth support vector machine  support vector machine
点击此处可从《软件学报》浏览原始摘要信息
点击此处可从《软件学报》下载全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号