基于不动点方程的非负矩阵分解算法

(厦门大学数学科学学院,福建 厦门 361005)

非负矩阵分解; 线性互补问题; 不动点方程; 最速下降法; 最小梯度法

Non-negative Matrix Factorization Algorithms Based on Fixed Point Equation
LIU Weiqiang,LIN Lu*

(School of Mathematical Sciences,Xiamen University,Xiamen 361005,China)

nonnegative matrix factorization; linear complementarity problem; fixed point equation; steepest descent; minimal gradient

DOI: 10.6043/j.issn.0438-0479.201712022

备注

从线性互补问题出发,通过非负矩阵分解问题与线性互补问题的关系,分别提出不动点方程的最速下降算法与最小梯度算法,证明了这两种算法的收敛性,并进行了数值实验.

Based on the linear complementarity problem,this paper proposes a steepest descent algorithm and a minimum gradient algorithm of the fixed point equation by the relationship between the NMF problem and the linear complementarity problem.The convergence of these two algorithms is proved and numerical simulations are carried out.