两类非线性降维流形学习算法的比较分析  

Comparison and Analysis of Two Categories of Manifold Learning Algorithms for Nonlinear Dimensionality Reduction

作　　者：王泽杰[1] 

机构地区：[1]上海工程技术大学计算中心,上海201620

出　　处：《上海工程技术大学学报》2008年第1期54-59,共6页Journal of Shanghai University of Engineering Science

基　　金：上海市选拔培养优秀青年教师科研专项基金资助项目(06XPYQ48)

摘　　要：流形学习(Manifold Learning)算法是近年来发展起来的非线性降维机器学习算法。目前的流形学习算法大体可以分为两类:全局的(如等度规映射)和局部的(如局部线性嵌套),它们有各自的优点和不足。以等度规映射(ISOMAP)和局部线性嵌套(LLE)为例,通过实验比较分析了这两类算法在参数选择、前提条件和执行效率上的特点,期望为不同应用提供参考。Manifold learning algorithms are nonlinear dimensionality reduction machine learning algorithms proposed in recent years. They fall broadly into two categories which have advantages and disadvantages respectively ： global（such as ISOMAP） and local（such as LLE）. Comparison and analysis of these two categories on parameter selection, premise and computational efficiency were given based on experimental results on ISOMAP and LLE. It is expected to give some insight into different applications.

关 键 词：流形学习 机器学习 非线性降维 等度规映射 局部线性嵌套 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TP181[自动化与计算机技术—计算机科学与技术]