Topic: Title: Reparameterisations for location-scale mixtures Authors:  Kate Lee - Auckland University of Technology (New Zealand) [presenting]
Abstract: Mixture models have been used in a wide variety of applications. The construction of reference Bayesian analysis for mixture models has been very challenging and still is not unsolved. We introduce the global mean-variance reparameterisation for mixture models that main consequence is to have all other parameters within a compact space. For Gaussian mixtures, a genuine non-informative prior is developed and the posterior distribution associated with this prior is almost surely proper as it is supported by simulation studies. While we only study the Gaussian case, extension to other classes of location-scale mixtures is straightforward. The MCMC implementation constrained on simplex equations to sample from a complex space is discussed.