Presentation 2014-07-11
Derivation of New Update Rules based on KL, γ, Renyi Divergences for Nonnegative Matrix Factorization
Masato SEKI, Norikazu TAKAHASHI,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Three new update rules based on Kullback-Leibler divergence, γ-divergence and Renyi divergence for nonnegative matrix factorization are presented in this report. Multiplicative update rules based on these divergences have already been proposed by Yang and Oja. However, their error functions have a common problem that we can increase the values of variables as much as we want while keeping the value of the error function fixed. In addition, the global convergence of their multiplicative update rules is not guaranteed. In order to solve these problems, we add a penalty term to each of the original error functions, and derive three new update rules based on the method of Yang and Oja and the one of Gillis and Glineur. We also prove that the new update rules have the boundedness property which means that any sequence of solutions generated by the new update rules is contained in a bounded region.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) nonnegative matrix factorization / multiplicative update / boundedness / global convergence
Paper # CAS2014-47,VLD2014-56,SIP2014-68,MSS2014-47,SIS2014-47
Date of Issue

Conference Information
Committee MSS
Conference Date 2014/7/2(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Mathematical Systems Science and its applications(MSS)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Derivation of New Update Rules based on KL, γ, Renyi Divergences for Nonnegative Matrix Factorization
Sub Title (in English)
Keyword(1) nonnegative matrix factorization
Keyword(2) multiplicative update
Keyword(3) boundedness
Keyword(4) global convergence
1st Author's Name Masato SEKI
1st Author's Affiliation Graduate School of Natural Science and Technology, Okayama University()
2nd Author's Name Norikazu TAKAHASHI
2nd Author's Affiliation Graduate School of Natural Science and Technology, Okayama University
Date 2014-07-11
Paper # CAS2014-47,VLD2014-56,SIP2014-68,MSS2014-47,SIS2014-47
Volume (vol) vol.114
Number (no) 125
Page pp.pp.-
#Pages 6
Date of Issue