半监督学习方法  

Semi-Supervised Learning Methods

作　　者：刘建伟[1] 刘媛[1] 罗雄麟[1] 

机构地区：[1]中国石油大学(北京)自动化研究所,北京102249

出　　处：《计算机学报》2015年第8期1592-1617,共26页Chinese Journal of Computers

基　　金：国家"九七三"重点基础研究发展规划项目基金(2012CB720500);国家自然科学基金(21006127);中国石油大学(北京)基础学科研究基金项目(JCXK-2011-07)资助~~

摘　　要：半监督学习研究如何同时利用有类标签的样本和无类标签的样例改进学习性能,成为近年来机器学习领域的研究热点.鉴于半监督学习的理论意义和实际应用价值,系统综述了半监督学习方法.首先概述了半监督学习的相关概念,包括半监督学习的定义、半监督学习研究的发展历程、半监督学习方法依赖的假设以及半监督学习的分类,然后分别从分类、回归、聚类和降维这4个方面详述了半监督学习方法,接着从理论上对半监督学习进行了分析并给出半监督学习的误差界和样本复杂度,最后探讨了半监督学习领域未来的研究方向.Semi-supervised learning is used to study how to improve performance in the presence of both examples and instances,and it has become a hot area of machine learning field.In view of the theoretical significance and practical value of semi-supervised learning,semi-supervised learning methods were reviewed in this paper systematically.Firstly,some concepts about semi-supervised learning were summarized,including definition of semi-supervised learning,development of research,assumptions relied on semi-supervised learning methods and classification of semisupervised learning.Secondly,semi-supervised learning methods were detailed from four aspects,including classification,regression,clustering,and dimension reduction.Thirdly,theoretical analysis on semi-supervised learning was studied,and error bounds and sample complexity were given.Finally,the future research on semi-supervised learning was discussed.

关 键 词：半监督学习 有类标签的样本 无类标签的样例 类标签 成对约束 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]