Modelling with Mixtures
Mike West · 1992
Abstract Discrete mixtures of distributions of standard parametric forms commonly arise in statistical modelling and with methods of analysis that exploit mixture structure. This paper discusses general issues of modelling with mixtures that arise in fitting mixtures to data distributions, using mixtures to approximate functional forms, such as posterior distributions in parametric models, and development of mixture pruning methods useful for reducing the number of components of large mixtures. These issues arise in problems of density estimation using mixtures of Dirichlet processes, adaptive importance sampling function design in Monte Carlo integration, and Bayesian discrimination and cluster analysis.