|
| 摘要: |
| 在非负矩阵分解中, 初值的选择对于算法效果有很大的影响. 一些基于奇异值分解的初始化方法已有人提出, 但当矩阵维数过大时, 直接对原矩阵进行奇异值分解是耗时的. 本文提出了一种更节时的初始化方法(KFV-NMF), 而且通过数值实验, 此算法既在一定程度上保持了计算精度, 也节省了计算时间. |
| 关键词: 非负矩阵分解 初始化 奇异值分解 FKV |
| DOI： |
| 分类号:O241.6 |
| 基金项目: |
|
| MONTE CARLO BASED INITIALIZATION FOR NONNEGATIVE MATRIX FACTORIZATION |
|
chenhongli
|
| Abstract: |
| The selection of initial values is crucial for nonnegative matrix factorization (NMF), because it significantly influences the effectiveness of NMF algorithms. Some initialization methods based on singular value decomposition (SVD) have been proposed. However, when the dimension of the matrix is very large, it is time-consuming to compute the SVD of original matrix directly. In this paper, we propose a more time-saving initialization method (KFV-NMF). Numerical experiments show that our initialization algorithm needs less time and the accuracy is also maintained to some extent. |
| Key words: nonnegative matrix factorization initialization SVD FKV |
