On the Bayesian analysis of species sampling mixture models for density estimation

Abstract

The mixture of normals model has been extensively applied to density estimation problems.
\nThis paper proposes an alternative parameterisation that naturally leads to new forms
\nof prior distribution. The parameters can be interpreted as the location, scale and smoothness
\nof the density. Priors on these parameters are often easier to specify. Alternatively, improper
\nand default choices lead to automatic Bayesian density estimation. The ideas are extended to
\nmultivariate density estimation.