FW: mathematics and cogntion seminar *TODAY*

From: Thomas Taylor
Sent: Tuesday, February 01, 2005 10:23 AM
To: William Uttal; ataxr@IMAP1.ASU.EDU; pikabruce@asu.edu; cheal@IMAP1.ASU.EDU; cinfante@cox.net; cochran@asu.edu; Root Gorelick; David Mix; Warren Van Egmond; Esma Gel; Erik Wennstrom; flavio.dasilva@asu.edu; hasan.davulcu@asu.edu; Jennifer Fewell; Jennifer Neakrase; jml@infinet-is.com; John Black; Junko Murakami; kathy.brower@asu.edu; Kevin Dooley; Peter Killeen; leachman@digitalf.com; leonidas.jassemidis@asu.edu; John Anderies; mgacula@IMAP3.ASU.EDU; mmcbeath@asu.edu; msitomer@asu.edu; Nia Amazeen; An Papandreou-Suppappola; Peter Crouch; Diana Posadas-Sanchez; Ronald Rutowski; rclement@asu.edu; renaut@asu.edu; rherron@mum.edu; Susan Bertram; Jennie Si; sjk1336@aol.com; George Stelmach; Steve.HelmsTillery@asu.edu; susan.somerville@asu.edu; Thomas Taylor; tsfrank@uswest.net; Nia Amazeen; Edward Castaneda; hdavulcu@asu.edu; Stephen Goldinger; Douglas Kenrick; paulmaycock@compuserve.com; presson@asu.edu; ozel@asu.edu; albert.thompson@asu.edu; kevin.gluck@mesa.afmc.af.mil; schvan@asu.edu; ncooke@asu.edu; Janet Neisewander; richardeverett8@netscape.net; lunatok@aol.com; m.anderies@asu.edu; William Griffin; kanav@asu.edu; Priyamvada Tripathi; Brad Armendt; Clark.Vangilder@asu.edu; Gayla Chandler; Steven Neuberg; macduff@asu.edu; jeanson@asu.edu; Timothy Newman; Erin Gaekel; msitomer@asu.edu; Federico Sanabria; Jennie Si; jcritt@asu.edu; Olga Kornienko; pgreenw@math.la.asu.edu; Bruce Long; Jerry Coursen; bradscientist@juno.com; Mariano Phielipp; Mmcbeath@asu.edu
Subject: mathematics and cogntion seminar *TODAY*

Mathematics and Cognition Seminar

Spring 2005

Tuesdays 12:00 GWC 604

Seminar Schedule:<http://math.la.asu.edu/~tom/cognition/math+cogsched.html>
On Tuesday, February 1, at 12:00 Noon in GWC 604,
the Mathematics and Cognition Seminar will
present brief discussion with Junko Murakami and Tom Taylor,
both of the Department of Mathematics and Statistics
and the Center for Social Dynamics and Complexity
on the topic of

"Least Mean Squares Error Parameter Estimation in Hidden Markov Models"

Abstract

Hidden Markov models (HMM) are a class of stochastic models which generalize Markov models, and may be viewed as a type of latent Markov variable model. They are widely applied in a variety of applied situations, including voice and image recognition. The predominant method of estimating the parameters of the HMM from observations is Maximum Likelyhood, implemented via some variant of the Expectation Maximization algorithm. It is known that this method will sometimes find local maxima of the likelyhood, and that sometimes the method works well and sometimes not. We present the results of some experiments with EM estimation of HMM, propose a reason why it will sometimes fail, and discuss LMSE estimation as an alternative to EM.