Title :
Boundedness of modified multiplicative updates for nonnegative matrix factorization
Author :
Katayama, Jiro ; Takahashi, Naoyuki ; Takeuchi, Jun
Author_Institution :
Kyushu Univ., Fukuoka, Japan
Abstract :
There have been proposed various types of multiplicative updates for nonnegative matrix factorization. However, these updates have a serious drawback in common: they are not defined for all pairs of nonnegative matrices. Furthermore, due to this drawback, their global convergence in the sense of Zangwill´s theorem cannot be proved theoretically. In this paper, we consider slightly modified versions of various multiplicative update rules, that are defined for all pairs of matrices in the domain, and show that many of them have the boundedness property. This property is a necessary condition for update rules to be globally convergent in the sense of Zangwill´s theorem.
Keywords :
matrix decomposition; NMF; Zangwill theorem; global convergence; modified multiplicative updates boundedness; nonnegative matrix factorization; Conferences; Convergence; Educational institutions; Euclidean distance; Information processing; Minimization; Optimization;
Conference_Titel :
Computational Advances in Multi-Sensor Adaptive Processing (CAMSAP), 2013 IEEE 5th International Workshop on
Conference_Location :
St. Martin
Print_ISBN :
978-1-4673-3144-9
DOI :
10.1109/CAMSAP.2013.6714055