Parallel Markov Chain Monte Carlo computation for varying-dimension signal analysis

Parallel implementation of Markov Chain Monte Carlo (MCMC) algorithms for Bayesian inference has been effective but is usually restricted to the case where the dimension of the parameter vector is fixed. We propose an efficient parallel solution for the varying-dimension problem by constructing multiple within-model MCMC chains and then combining the separate results to analyze the posterior distribution of dimensionality. We aim for parallel speed-up by reducing the length of the burn-in period and the individual chains in comparison with a serial, reversible jump MCMC (RJMCMC) algorithm. The parallel methodology is illustrated with application to a benchmarking, change point problem.