基于认知的流形学习方法概要  

Overview of the Manifold Learning Algorithm

作　　者：周谆[1] 杨炳儒[1] 

机构地区：[1]北京科技大学信息工程学院,北京100083

出　　处：《计算机科学》2009年第5期234-237,264,共5页Computer Science

基　　金：国家自然科学基金重点项目(69835001);教育部科技重点项目([2000]175);北京市自然科学基金项目(4022008)资助

摘　　要：流形学习是一种新出现的机器学习方法,近年来引起越来越多的计算机科学工作者和认知科学工作者的重视。为了加深对流形学习的认识和理解,从流形与流形学习的基本概念入手,追溯它的发展历程。针对目前的几种主要的流形算法,分析它们各自的优势和不足,然后引用LLE的应用示例,说明流形学习较之于传统的线性降维方法如PCA等,能够有效地发现非线性高维数据的本质维数,可以有效地进行维数约简和数据分析。最后对流形学习未来的研究方向做出展望,以期进一步拓展流形学习的应用领域。As a new machine learning method, manifold learning is capturing increasing interests of researchers in the field of computer sciences and cognitive sciences. To understand manifold learning better, the concept of manifold and manifold learning was presented, and then its history was traced. Several major different manifold learning algorithms were introduced,whose advantages and disadvantages were pointed out respectively. Then a typical application of LLE was indicated. The results show that compared with traditional linear dimensionality reduction methods such as PCA, manifold learning can discover the intrinsic dimensionality better. Finally, the proposal of manifold learning was discussed for the application.

关 键 词：维数约简 机器学习 流形学习 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论] TP18[自动化与计算机技术—计算机科学与技术]