一种改进的LLE方法

局部线性嵌入(Locally Linear Embedding,简称LLE)是一种较好的非线性降维方法,这种方法对于位于某种非线性流形上的数据的降维有着比较好的效果.但是这种方法对于其中一个重要参数——近邻个数,太过敏感.文章将另一种非线性降维方法Conformal-Isomap中的一种度量数据之间距离的方法引入到LLE方法中.经过实验发现,新引入的距离对于近邻个数的选择有比较好的效果,可以使得实验的结果对近邻个数的选择不那么敏感.

One improved LLE method

Locally Linear Embedding(LLE)is a good nonlinear dimensionality reduction method,this method is good for the data that lies on a nonlinear manifold.But it was too sensitive for one of the important parameters-the number of nearest neighbors.This paper used a method which was introduced in another nonlinear dimensionality reduction method called Conformal-Isomap to measure the distance between two samples in LLE algorithm.Through experiments,we find that this distance measurement is good for selecting the number of nearest neighbors,the results of experiments are not so sensitive to the number of the nearest neighbors.