On Bayesian thresholding and truncation methods

초록

Thresholding dynamic process is a powerful statistical tool for modeling structural nonlinear relationships. We here discuss a Bayesian formalism to give rise to a type of threshold estimation in dynamic process with spatial structure. A prior distribution is imposed on the unknown parameter of process model, designed to capture the sparseness of process parameters that is common to most application. For the prior specified, the posterior distribution yields a thresholding procedure. In this paper, we introduce a general approach in which the truncation step is directly implanted to MCMC procedure by thresholding the MCMC outputs. The proposed thresholding approach is applied to the basal topography of the Northeast Ice-Stream in Greenland by using Daubechies wavelet-based analysis.

키워드