【摘要】 流形学习算法分为两类,一类是等距映射算法,一类是等角映射算法,它们都有各自适用的数据类型.现有的流形学习算法都是直接处理流形数据,不对数据集作数据类型判定,使得算法在处理一些数据集时,降维结果较差.本文提出首先判定数据集的类型,然后根据数据集的类型,采用合适的流形学习算法进行降维的思想,给出了中心对称流形的定义和中心对称流形数据类型的判别定理,提出一个中心对称流形数据类型的判别算法.对人工数据集的实验表明,该算法能够准确地判定中心对称流形数据集的类型.更多还原

【Abstract】 There are two kinds of manifold learning algorithms: isometric mapping and conformal mapping. They are applicable to some type of datasets. Instead of identifying the datasets type,the manifold learning algorithms available are used to processing manifold datasets directly,which results in a bad dimensionality reduction when processing some datasets. A solution is proposed that identifying the datasets type firstly,then according to the type of datasets,taking the right manifold learning algorithm. The definition of central symmetry manifold and the discriminant theorem of data types of central symmetry manifold are given and a discriminant algorithm of data types of central symmetry manifold is proposed. Experimental results on a series of synthetic datasets verify that the discriminant algorithm of central symmetry manifold could identify the datasets type accurately.更多还原