Variational Gaussian mixtures for blind source detection

N Nasios; A G Bors

Variational Gaussian mixtures for blind source detection

Computer Science

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

Bayesian algorithms have lately been used in a large variety of applications. This paper proposes a new methodology for hyperparameter initialization in the Variational Bayes (VB) algorithm. We employ a dual expectation-maximization (EM) algorithm as the initialization stage in the VB-based learning. In the first stage, the EM algorithm is used on the given data set while the second EM algorithm is applied on distributions of parameters resulted from several runs of the first stage EM. The graphical model case study considered in this paper consists of a mixture of Gaussians. Appropriate conjugate prior distributions are considered for modelling the parameters. The proposed methodology is applied on blind source separation of modulated signals.

Original language	English
Title of host publication	2003 IEEE INTERNATIONAL CONFERENCE ON SYSTEMS, MAN AND CYBERNETICS, VOLS 1-5, CONFERENCE PROCEEDINGS
Place of Publication	NEW YORK
Publisher	IEEE
Pages	474-479
Number of pages	6
ISBN (Print)	0-7803-7952-7
Publication status	Published - 2003
Event	IEEE International Conference on Systems, Man and Cybernetics (SMC 03) - WASHINGTON Duration: 5 Oct 2003 → 8 Oct 2003

Conference

Conference	IEEE International Conference on Systems, Man and Cybernetics (SMC 03)
City	WASHINGTON
Period	5/10/03 → 8/10/03

Keywords

Gaussian mixtures
Bayesian inference
variational learning
expectation-maximization algorithm
MODELS

Cite this

@inproceedings{162b6af5a8f8441c8825b33f676e0957,

title = "Variational Gaussian mixtures for blind source detection",

abstract = "Bayesian algorithms have lately been used in a large variety of applications. This paper proposes a new methodology for hyperparameter initialization in the Variational Bayes (VB) algorithm. We employ a dual expectation-maximization (EM) algorithm as the initialization stage in the VB-based learning. In the first stage, the EM algorithm is used on the given data set while the second EM algorithm is applied on distributions of parameters resulted from several runs of the first stage EM. The graphical model case study considered in this paper consists of a mixture of Gaussians. Appropriate conjugate prior distributions are considered for modelling the parameters. The proposed methodology is applied on blind source separation of modulated signals.",

keywords = "Gaussian mixtures, Bayesian inference, variational learning, expectation-maximization algorithm, MODELS",

author = "N Nasios and Bors, {A G}",

year = "2003",

language = "English",

isbn = "0-7803-7952-7",

pages = "474--479",

booktitle = "2003 IEEE INTERNATIONAL CONFERENCE ON SYSTEMS, MAN AND CYBERNETICS, VOLS 1-5, CONFERENCE PROCEEDINGS",

publisher = "IEEE",

address = "United States",

note = "IEEE International Conference on Systems, Man and Cybernetics (SMC 03) ; Conference date: 05-10-2003 Through 08-10-2003",

}

TY - GEN

T1 - Variational Gaussian mixtures for blind source detection

AU - Nasios, N

AU - Bors, A G

PY - 2003

Y1 - 2003

N2 - Bayesian algorithms have lately been used in a large variety of applications. This paper proposes a new methodology for hyperparameter initialization in the Variational Bayes (VB) algorithm. We employ a dual expectation-maximization (EM) algorithm as the initialization stage in the VB-based learning. In the first stage, the EM algorithm is used on the given data set while the second EM algorithm is applied on distributions of parameters resulted from several runs of the first stage EM. The graphical model case study considered in this paper consists of a mixture of Gaussians. Appropriate conjugate prior distributions are considered for modelling the parameters. The proposed methodology is applied on blind source separation of modulated signals.

AB - Bayesian algorithms have lately been used in a large variety of applications. This paper proposes a new methodology for hyperparameter initialization in the Variational Bayes (VB) algorithm. We employ a dual expectation-maximization (EM) algorithm as the initialization stage in the VB-based learning. In the first stage, the EM algorithm is used on the given data set while the second EM algorithm is applied on distributions of parameters resulted from several runs of the first stage EM. The graphical model case study considered in this paper consists of a mixture of Gaussians. Appropriate conjugate prior distributions are considered for modelling the parameters. The proposed methodology is applied on blind source separation of modulated signals.

KW - Gaussian mixtures

KW - Bayesian inference

KW - variational learning

KW - expectation-maximization algorithm

KW - MODELS

M3 - Conference contribution

SN - 0-7803-7952-7

SP - 474

EP - 479

BT - 2003 IEEE INTERNATIONAL CONFERENCE ON SYSTEMS, MAN AND CYBERNETICS, VOLS 1-5, CONFERENCE PROCEEDINGS

PB - IEEE

CY - NEW YORK

T2 - IEEE International Conference on Systems, Man and Cybernetics (SMC 03)

Y2 - 5 October 2003 through 8 October 2003

ER -