A novel EM algorithm for stable optimum

Y. Xiao, G. R. Xuan, Z. Q. Yang, Y. Q. Shi

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

The main contribution of this paper is to propose a novel EM algorithm that utilizes ideas of EM algorithm and maximum-entropy uniform distribution to find a stable optimum in an uncomplicated way. The conventional EM algorithms may suffer from the following two problems: first, it may converge to an undetermined local maximum; second, the algorithm may suffer from singularity. The novel EM algorithm is deterministic that the solution is determined solely by the initial condition. In our novel EM algorithm, a stable and optimal solution can be obtained by using a uniform distribution instead of special initial condition. In addition, a positive perturbation scheme is adopted to avoid singularity. Experimental results have demonstrated that the novel EM is uncomplicated and effective for stable optimum compared with some prior arts.

Original languageEnglish (US)
Title of host publicationNetwork Security and Communication Engineering - Proceedings of the 2014 International Conference on Network Security and Communication Engineering, NSCE 2014
EditorsKennis Chan
PublisherCRC Press/Balkema
Pages189-192
Number of pages4
ISBN (Print)9781138028210
DOIs
StatePublished - 2015
EventInternational Conference on Network Security and Communication Engineering, NSCE 2014 - Hong Kong, China
Duration: Dec 25 2014Dec 26 2014

Publication series

NameNetwork Security and Communication Engineering - Proceedings of the 2014 International Conference on Network Security and Communication Engineering, NSCE 2014

Other

OtherInternational Conference on Network Security and Communication Engineering, NSCE 2014
Country/TerritoryChina
CityHong Kong
Period12/25/1412/26/14

All Science Journal Classification (ASJC) codes

  • Computer Networks and Communications
  • Aerospace Engineering

Fingerprint

Dive into the research topics of 'A novel EM algorithm for stable optimum'. Together they form a unique fingerprint.

Cite this