Parameter Selection Strategy for Sparsity Enforcing Prior Models

Abstract

The Bayesian framework and heavy tailed distributions expressed as continuous Gaussian scale mixtures have been used intensively in the sparsity context. The Posterior Mean corresponding iterative algorithms strongly depend on the parameter selection. We propose a parameter selection strategy based on the link of the mixing and prior distribution. We compare it with other parameter selection strategies for three prior models obtained as particular cases of the Generalized Hyperbolic distribution and show that the proposed parameter selection strategy seem to be more suitable for the sparsity context.