Bayesian computational approaches to model selection (Q2712134)

The authors begin by noting that `A fundamental task in signal and data processing and science is, in general, to develop models for signals which are observed and to determine whether the model function that one is using to describe the data is actually appropriate for the particular problem under investigation'. Within a Bayesian framework, consideration is paid to the problems of parameter estimation and model selection. The question of the determination of prior distributions on models and their parameters is addressed. Numerical methods for the computation of the Bayes factors and posterior model probabilities are discussed, with special attention being paid to the use of Markov chain Monte Carlo methods. Finally the methodology is applied to the detection of sinusoids in noise and the determination of the components of Gaussian mixture models.NEWLINENEWLINEFor the entire collection see [Zbl 0958.00020].