流形学习算法中的参数选择问题研究  

ON PARAMETER SELECTION IN MANIFOLD LEARNING ALGORITHM

作　　者：王泽杰[1] 胡浩民[1] 

机构地区：[1]上海工程技术大学计算中心,上海201620

出　　处：《计算机应用与软件》2010年第6期84-85,102,共3页Computer Applications and Software

基　　金：上海市选拔培养优秀青年教师基金项目(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。等度规特征映射Isomap(Isometric feature mapping)和局部线性嵌入LLE(Locally Linear Embedding)是两种典型的流形学习算法。通过实验比较和分析两种算法中邻接参数K和采样点数N的选取对降维结果以及执行时间的影响,实验结果表明Isomap对邻接参数K和采样点数N具有较高的容忍度,而LLE算法在计算速度上优势明显。Manifold learning algorithms are nonlinear dimensionality reduction machine learning algorithms rising in recent years.Isometric feature mapping（Isomap） and local linear embedding（LLE） are two typical manifold learning algorithms.Comparison and analysis of the effect of the selection of adjoining parameter K and sampling point number N in two algorithms on the reduction results and computational efficiency are performed through experiment.Experimental results suggested that Isomap has higher tolerance to the parameter K and sampling points number N than LLE,but LLE has conspicuous advantage in computational speed.

关 键 词：等度规特征映射 局部线性嵌入 流形学习 非线性降维 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]