流形学习算法及泛化研究
首发时间:2008-12-09
摘要:本文介绍了流形学习以及目前流形学习研究中的几种常见的算法:等度归映射算法(Isomap,Isometric Mapping),拉普拉斯特征映射算法(LE,Laplacian Eigenmaps),局部线性嵌套算法(LLE,Locally linear Embedding),局部切空间排列算法(LTSA,Local tangent space alignment)等。分析了这几种算法的特点和复杂度,并重点研究了其中的LLE算法,介绍了LLE算法的主要步骤以及几种已有的针对LLE算法的泛化方法。根据LLE算法的特点和目前LLE算法泛化中存在的问题,本文提出了一种新的针对LLE算法的泛化方案,本方案能够显著提高LLE的泛化性能,在流形没有被很好的采样,无法充分地代表潜在的流形时也能够有效工作,同时具有较小的运算复杂度,能够快速精确实现LLE的泛化。
For information in English, please click here
Rearch on Mainfold Learning Algorithms and Generalizing Method
Abstract:Manifold learning and several common algorithms of manifold learning are introduced in this paper: Isomap (Isometric Mapping), LE (Laplacian Eigenmaps), LLE (Locally linear Embedding), LTSA (Local tangent space alignment). The characteristics and complexities of these algorithms are analyzed. This paper focuses on the LLE (Locally linear Embedding) algorithm and the generalizing method and introduces some generalizing methods for LLE. According to the characteristics of LLE algorithms and the disadvantages of existing generalizing methods, a new generalizing method for LLE algorithms is proposed. It can improve the performance of generalizing method, resolve some problems faced in the exciting methods, have lower complexity and achieve the LLE generalizing accurately.
Keywords: Manifold-learing Dimension-reducing LLE Generalizing
基金:
论文图表:
引用
No.2641236943612288****
同行评议
共计0人参与
勘误表
流形学习算法及泛化研究
评论
全部评论0/1000