非负矩阵分解毕业论文P145.docxVIP

Ⅱ  STYLEREF 标题 1 \* MERGEFORMAT Error! No text of specified style in document.  STYLEREF 标题 1 \* MERGEFORMAT Error! No text of specified style in document. 摘 要 在需求处理大量高维数据的今天，需要优秀的降维技术来解决数据问题，如主成分分析，奇异值分解等等，这些数据分析方法的共同点是无法保证数据的非负性，而负值往往没有物理意义。非负矩阵分解（Non-negative Matrix Factorization）是一种保证非负属性的优秀降维方法。 本文中先对非负矩阵分解做一个简短的介绍，讲述了非负矩阵分解的一些算法。而这些算法都可以在效率或其他属性上进行改进，故本文的主要目的在于获取更高效率的非负矩阵分解算法，数值实验证明，改进后的算法PCGSNMF算法在效率和稀疏度上展现出优越性。最后，对文章进行了总结。 关键词：非负矩阵分解 降维技术 梯度 ALS算法 Abstract With the processing of large amounts of high dimensional data, we need good dimensionality reduction techniques to solve this problem, such as PCA, SVD, and so on, because these methods cant guarantee the non-negative property, so it has a guarantee of non-negative constraints of the dimensionality reduction method, that is, non-negative matrix factorization (NMF) algorithm. In this paper, we first do a brief introduction summarized, The main purpose of this paper is to put forward to a new NMF algorithm which has higher efficiency. The numerical experiments will demonstrate the superiority of the new algorithm. And then briefly introduces some of the NMF application now. Finally, some conclusions about NMF algorithm are described. Keywords: nonnegative matrix factorization Dimensionality reduction techniques Gradient alternating nonnegative least squares method PAGE \* MERGEFORMATx  STYLEREF 标题 1 \* MERGEFORMAT Error! No text of specified style in document.  STYLEREF 标题 1 \* MERGEFORMAT Error! No text of specified style in document. PAGE \* MERGEFORMATix 目 录 TOC \o 1-3 \u  第一章 绪论  PAGEREF _Toc422137806 \h 1 1.1 NMF发展概况和研究现状  PAGEREF _Toc422137807 \h 1 1.2 NMF的数学问题介绍  PAGEREF _Toc422137808 \h 3 1.3 NMF的应用介绍  PAGEREF _Toc422137809 \h 4 1.4 KKT条件  PAGEREF _Toc422137810 \h 6 1.5 本文的研究内容及结构安排  PAGEREF _Toc422137811 \h 7 第二章 NMF算法  PAGEREF _Toc422137812 \h 9 2.1 LS基础NMF算法  PAGEREF _Toc422137813 \h 9 2.2 ALS交替最小二乘  PAGEREF _