Finally, we study degeneracies in the source separation problem where the characterization ... PC1Ъ can be diagonalized in the orthogonal group: PCЪ Ю Ð. Я.
Penalized maximum likelihood for multivariate Gaussian mixture Hichem Snoussi and Ali Mohammad-Djafari
Laboratoire des Signaux et Systèmes (L2S), Supélec, Plateau de Moulon, 91192 Gif-sur-Yvette Cedex, France Abstract. In this paper, we first consider the parameter estimation of a multivariate random process distribution using multivariate Gaussian mixture law. The labels of the mixture are allowed to have a general probability law which gives the possibility to modelize a temporal structure of the process under study. We generalize the case of univariate Gaussian mixture in [1] to show that the likelihood is unbounded and goes to infinity when one of the covariance matrices approaches the boundary of singularity of the non negative definite matrices set. We characterize the parameter set of these singularities. As a solution to this degeneracy problem, we show that the penalization of the likelihood by an Inverse Wishart prior on covariance matrices results to a penalized or maximum a posteriori criterion which is bounded. Then, the existence of positive definite matrices optimizing this criterion can be guaranteed. We also show that with a modified EM procedure or with a Bayesian sampling scheme, we can constrain covariance matrices to belong to a particular subclass of covariance matrices. Finally, we study degeneracies in the source separation problem where the characterization of parameter singularity set is more complex. We show, however, that Inverse Wishart prior on covariance matrices eliminates the degeneracies in this case too.
INTRODUCTION We consider a double stochastic process:
, with taking its values in the discrete set . A discrete process !
which is white conditionally to the first process • A continuous process !
and following a distribution: •
" $#
&%
('*),+,
In the following, without loss of generality of the considered model, we restrict the # / +10 2 +, function % to be a Gaussian: % . .This double process is called in literature "Mixture model".+ When the hidden process 3 / + 072 +, 3 + " " 4 65 is white, we have an i.i.d mixture model: and when is Markovian, the model is called HMM (Hidden Markov Model). For application of these two models see [2] and [3]. Mixture models present an interesting alternative to non parametric modeling. By increasing the number of mixture components, we are able to approximate any probability density and the time dependence structure of the hidden 3 process allows to take into account a correlation structure of the resulting process. In the following, for clarity of demonstrations, we assume an i.i.d. mixture model.
CHARACTERIZATION OF LIKELIHOOD DEGENERACY We consider 8 observations Gaussian mixture law:
!
"
" Where A@
+
?
;
of a random 9 -vector following a multivariate : + =
To get the best image quality from an optical system in terms of contrast, it is required to deconvolve the image from the point spread function (PSF). For this ...
and flexible programming language for maximum likelihood estimation (MLE). In this document, I describe the basic syntax elements that allow you to .... function and the fifth line ends the program. The action, of course, is in the fourth line.
presence of the interference terms is the main issue for the implementation of the maximum likelihood technique. In this work, we propose to insert null data ...
tion has recently been exploited by other authors [7], where the source PDF ... represents the joint PDF of N samples of the source s ..... 1712-1725, July 1997. 5.
1. Imperfect Measurements, 8 AN, 20 RCN, size 20Ã20m, Ï â¼ U(0.1,3). C irc u la r E rro r P ro b a b ility. (C. E. P. ) Positioning Error (m). 50%-60%. 70%-80%. 90%-100% ..... Positioning Aspects for Future Sensor Networksâ, IEEE SP Mag., vol.
an MMSE equalizer is used, either alone or in an iterative procedure involving convolutional coding. Now, the issue is the application of ML techniques to FBMC.
Aug 11, 2014 - ing very recent and strong contractions may be difficult to achieve. However ..... hidden Markov models (e.g., Dutheil et al. 2009; Mailund et al.
has been studied in depth for orthogonal frequency division ... project N°211887, Phydyas. ..... 3 as long as the SNR at the receiver output does not exceed 20.
Dec 10, 2010 - and this effect does not disappear with increasing sample sizeâ (page 47). (v). .... Furthermore, in the case of embedded models, M2 and M1 â M2, ..... 2010. On resolving the SavageâDickey paradox. Electron. J. Statist.
discrete distributions referred to as multivariate mixed Poisson dis- tributions (MMPDs). ... pression, the computation of its probability distribution function (pdf).
Bernardoff recently defined MGDs by the form of their mo- ment generating ... The key element of the image registration problem, is the es- timation of the ..... radar backscatter from selected agricultural targets,â NASA STI/Recon. Technical ...
Description of the first axis: negative side ( 3 / 8 ). The following ..... Distribution of the individuals per cluster for the variable walk.2.to.5.km group 1 group 2.
The reader is invited to check that the model remains ..... Waelbroeck, editor, Nonlinear Operators and the Calculus of Variations, Springer-Verlag series Lec-.
Our Example Spec File: Subpackages Galore! ............................... ...... First, RPM displays the package's record number. ..... In the following examples, we'll use the letters X, ..... query, verify, update, and uninstall individual softwar
18, 212±216. Pannu, N. S., Murshudov, G. N., Dodson, E. J. & Read, R. J. (1998). Acta Cryst. D54, 1285±1294. Pannu, N. S. & Read, R. J. (1996). Acta Cryst.
