Distinguished Lecture

[Thomas Bartold]

From: "Susan K. Yale" <sky...@eecs.umich.edu>
To: <eecs-a...@eecs.umich.edu>; <cs...@eecs.umich.edu>;
<cspl_...@eecs.umich.edu>; <es...@eecs.umich.edu>;
<system_...@eecs.umich.edu>
Subject: LUCENT Distinguished Lecture Friday 10/6
Date: Tuesday, October 03, 2000 10:08 AM

LUCENT DISTINGUISHED LECTURE SERIES IN COMMUNICATIONS AND NETWORKS
presents

ROBERT M. GRAY
Stanford University

"GAUSS MIXTURE VECTOR QUANTIZATION"

Abstract-
The "worst case" attribute of Gaussian vectors for data
compression/source coding originally developed by Sakrison and
Lapidoth using Shannon rate-distortion theory is developed using the
high rate quantization theory of Bennett, Zador, and Gersho and
extended to Gauss mixtures, providing an approach to robust data
compression for nonGaussian sources such as images. The analysis
provides several interesting side results, including a new
interpretation of the minimum discrimination information distortion
(MDI) measure and its application to clustering models and
constructing Gauss mixture models based on training data. High rate
quantization theory provides a mathematical connection between the
distortion and the performance of a classified vector quantizer for
nonGaussian data designed using Gaussian distributions. Although the
primary application is compression and classification, several ideas
relating maximum entropy density estimation, the MAXDET problem, and
Markov mesh random fields arise in the analysis. At this time no
experimental evidence exists that the approach works for image
coding, the formulation of the theory, but the theory provides a
hindsight explanation for why CELP speech coder work as well as they
do.

Friday -- October 6
11:00 AM
1005 EECS

ALL ARE INVITED TO ATTEND

--
CSEG Web Page: http://www.eecs.umich.edu/CSEG
See the CSEG web page for unsubscription instructions.

Please support CSEG by buying your soda from CSEG vending machines.