一种实用快速非负矩阵分解算法  

A practical fast NMF algorithm

作　　者：程明松[1] 刘勺连[1] 

机构地区：[1]大连理工大学数学科学学院,辽宁大连116024

出　　处：《大连理工大学学报》2013年第1期151-156,共6页Journal of Dalian University of Technology

基　　金：中央高校基本科研业务费专项资金资助项目(DUT10LK04)

摘　　要：提出了一种基于快速非负矩阵分解算法的实用新算法.该实用快速非负矩阵分解算法扩展了快速非负矩阵分解算法的约束条件,并且保持了较高的收敛速度,更具一般性和实用性.然后对该新算法进行了一些稀疏非负矩阵分解的扩展应用.数值实验显示该实用快速非负矩阵分解算法和快速非负矩阵分解算法具有相近的收敛速度,与其他经典非负矩阵分解算法相比其收敛速度有明显的提高,同时对添加稀疏性约束条件的实验也有很好的效果.Based on the fast non-negative matrix factorization （NMF） algorithm, a new practical fast NMF algorithm is developed. The new algorithm extends the constrained conditions of the fast NMF algorithm, and it keeps having the fast convergence speed, so it is more general and practical. Then, some extended applications for sparse NMF are proposed based on the new algorithm. Numerical experiments show that the practical fast NMF algorithm and the fast NMF algorithm have similar convergence speed, which is improved significantly compared with other classical NMF algorithms. Moreover, the practical fast NMF algorithm also has good experimental results for the cases with sparse constrained conditions being added.

关 键 词：非负矩阵分解 快速非负矩阵分解算法 实用快速非负矩阵分解算法 稀疏非负矩阵分解 

分 类 号：O241.6[理学—计算数学]