Bayesian inference for an extended class of hidden Markov random field models using a spike-and-slab formulation

For an extended class of hidden Markov random field models incorporating weighted dependence among latent variables, we develop a Bayesian framework based on spike-and-slab prior distributions that accommodate both independence and dependence structures, allowing automatic model selection. The proposed approach unifies finite mixture and hidden Markov random field models within a single framework. We illustrate this approach using both simulated data and a real data application.