基于NMF初始化的非负双奇异值分解算法分析

Analysis of non-negative double singular Value decomposition algorithm based on NMF initialization

随着大数据技术的发展,非负矩阵分解(NMF)日益成为目前最流行的模式识别方法之一,并广泛应用于文档聚类、图像处理、人脸识别、信号分析等多个领域。针对NMF中双因子矩阵的初始化问题,对非负双奇异值分解算法进行分析,数值实验表明该算法可以快速降低众多基于NMF衍生算法的近似误差。

With the development of big data technology,Non-negative matrix factorization(NMF)has increasingly become one of the most popular pattern recognition methods,and is widely used in document clustering,image processing,face recognition,signal analysis and other fields.In order to initialize the two-factor matrix in NMF,this paper analyzes the non-negative double singular value decomposition algorithm,which can quickly reduce the approximate error of many NMF variants.