Explaining slow convergence of EM in low noise linear mixtures

Kaare Brandt Petersen, Ole Winther

AbstractThis report conducts an investigation of the convergence properties of the EM algorithm used for linear mixture models. Since the linear mixture model is a rather general approach, the analysis is relevant for a wide range of models which to some degree are subsets of each other: Independent Component Analysis (ICA), probabilistic PCA, Factor Analysis (FA), Independent Factor Analysis (IFA) and Mean Field ICA.
TypeTechnical report
PublisherInformatics and Mathematical Modelling, Technical University of Denmark, DTU
AddressRichard Petersens Plads, Building 321, DK-2800 Kgs. Lyngby
SeriesIMM-Technical report-2005-02
Electronic version(s)[pdf]
BibTeX data [bibtex]
IMM Group(s)Intelligent Signal Processing