非负矩阵分解算法可靠性分析生物医学工程专业论文.docxVIP

非负矩阵分解算法可靠性分析Reliability 非负矩阵分解算法可靠性分析 Reliability Investigation of Non—negative Matrix Factorization Algorithms Abstract This research aims to use cluster analysis to evaluate the stability and reliability of different Non—negative Matrix Factorization(NMF)algorithms．Furthermore，trying different clustering methods to find more suitable cluster analysis for NMF algorithms． This study presents develop and using ofthe new method named Non-negative Matrix Factorization(NMF)technique on electroencephalogram(EEG)data．NMF，relatively novel paradigm for dimensionality reduction，and recently，it has been used since development．Since the NMF paradigm has been developed，a large number of NMF algorithms based on different constraints conditions have been generated．NMF incorporates the non—negativity constraint and thus obtains the parts-based representation as well as enhancing the interpretability of issue correspondingly．It can extract some features and source signals from a group of observed signals．Moreover，NMF algorithm is based on data characteristics without any limitations such as，independency and orthogonality．However，most of NMF algorithms adaptive and the initialization trends to be random．Therefore，NMF could decompose great mount ofuncertainties．It means that twice running NMF decomposition on the same dataset may produce different results． I have organized a package that can compare different NMF methods with cluster analysis．First，we can select an NMF algorithm，and then repeat，for instance 50 times，running the NMF method decomposition on the same dataset．Therefore，we got many non—negative component matrix and coefficient matrix，and then connect these matrix in a big matrix．At last clustering analysis the big matrix，we can compare stability of different NMF methods from the result of clustering． This study just introduce cluster analysis on stability of NMF，I don’t make improvement and optimization on the NMF that has low stability．Accordance with the thinking，1 wil