基于RSOM聚类的局部线性嵌入算法

LLE algorithm based on RSOM clustering

局部线性嵌入算法(locally linear embedding,LLE)是一种非线性降维方法。当数据量较大时,算法计算效率较低,算法运行所占用的内存空间较大。为了提高LLE算法的计算效率和减小算法运行时占用的内存空间,给出了基于RSOM(Recursive SOM)树聚类的LLE算法,通过RSOM树对数据集进行聚类,在保证输入样本依概率分布的同时显著降低算法复杂度,提高了映射效果。仿真实验表明,基于RSOM树聚类的LLE算法相对于原始的LLE算法,其算法效率有了显著提高,明显降低了算法运行所占用的内存空间,同时很好地学习了高维数据的流形结构。

Locally linear embedding（LLE）is one of nonlinear dimensionality reduction technique. When large database is performed, the algorithm is time--consuming and huge memory space is occupied. In order to improve the efficiency of the LLE algorithm, a LLE algorithm based on RSOM tree clustering is proposed. Through clustering of RSOM tree, the computation complexity of the I.LE algorithm is reduced and the proba- bility of the database is retained. Experiments show that, compared to the original LLE algorithm, the efficien- cy of the RSOM tree clustering based LLE algorithm is improved remarkably and the memory space is reduced. The manifold structure of the database is also learned correctly.