摘要
局部线性嵌套LLE(locally linear embedding)是一种经典的流形学习方法。对于从单个流形上采样得到的数据集,它能够有效地学习其内在低维结构,然而当数据集是从多个流形上采样得到时,LLE的效果并不理想。提出了一种基于距离度量学习的改进方法:Metric LLE,它利用部分数据点的相似信息来学习距离度量。实验结果表明Metric LLE在应用中有很好的性能:分类能力比LLE好;在可视化方面,效果比Supervised LLE好。
Locally linear embedding （ LLE ） is a classical manifold learning method. It is efficient in learning internal low-dimensional structure for data set sampled from a single global manifold. But when data sets are laid on （ or near） multiple manifolds, it often performs poor. In this paper, a semi-supervised variant of LLE called Metric LLE is proposed based on distance metric learning, which learns distance metric by similar information from partial data points. It is shown by the experiment that in application the Metric LLE performs better than LLE in terms of classification and Supervised LLE in terms of visualization.
出处
《计算机应用与软件》
CSCD
北大核心
2008年第10期9-10,18,共3页
Computer Applications and Software
基金
国家自然科学基金项目(60473104)
关键词
流形学习
距离度量
局部线性嵌套
多流形
Manifold learning Distance metric Locally linear embedding Multiple manifolds
