基于密度信息的改进降维方法

Improvement method of dimensionality reduction through mining density information

扩散映射(diffusion maps)是一种基于流形学习的非线性降维方法。为了提高降维的效果,根据近邻点的选取对diffusion maps的降维效果影响,利用数据近邻点分布的不同,挖掘该数据点局部的密度信息,能够更好地保持数据的流形结构。利用样本点聚类后的类别信息构造密度信息指数,提出了一种改进的diffusion maps算法,有效地保持了高维数据中的流形结构,所提的新算法在多种实验中得到了证实。

Diffusion maps is based on manifold learning nonlinear dimensionality reduction method. The effective of the diffusion maps dimension reduction is affected by the distribution of the neighboring points, mining local density information , it can better maintain the structure of the data. Using sample points after clustering to construct density coefficient, this paper proposed an improved method of diffusion maps algorithm, effective to keep the high dimension data of the manifold structure, and confirmed the proposed method in the experiments.