We propose a new approach to sample from posterior distributions of statistical models with intractable normalizing constants. The algorithm builds on the extended Wang-Landau algorithm of Atchade and Liu (07) which provides, using a single Monte Carlo run, an efficient estimate of the intractable normalizing constant at every point of the parameter space. We show that the method is a valid adaptive MCMC method. We illustrate the method with an application to image segmentation.
Joint with: Nicolas Lartillot, Universite de Montpellier, Christian P. Robert, Universite Paris Daphine.