Gibbs sampling for parameters estimation and change-points detection in inverse problems. Application to electromagnetic imaging

In the framework of Bayesian inverse problems, we investigate the use of su itable prior probabilities for modeling the presence of abrupt changes in t he distribution of the non-observed data sequence. We adopt a Gibbs-type sa mpling method for estimating the posterior distribution of this sequence. I n the second part, we apply recent results on stochastic versions of the we ll-known EM algorithm with averaging and acceleration techniques, to estima te some parameters of the model. A numerical example for the magnetotelluri c inverse problem is proposed. (C) 1999 Elsevier Science B.V. All rights re served.