A0762
Title: Flexible modeling of grouped multivariate data via Bayesian shared-atom nested mixture models
Authors: Federica Zoe Ricci - University of California Irvine (United States)
Laura D Angelo - Universita di Milano Bicocca (Italy) [presenting]
Abstract: The use of hierarchical mixture priors with shared atoms has recently flourished in the Bayesian literature for partially exchangeable data. Leveraging on nested levels of discrete random measures, these models allow the estimation of a two-layered data partition: across groups and among observations. We illustrate the properties of such modeling strategies when the mixing weights are assigned either a finite-dimensional Dirichlet distribution or a Dirichlet process prior. We discuss the use of hierarchical nonparametric priors based on a finite set of shared atoms, showing how this specification preserves the flexibility of the induced random measure and allows the derivation of fast posterior inference. Specifically, we develop a mean-field variational algorithm for posterior inference to boost the applicability to large multivariate data. This allows us to fit a nested mixture model to a real dataset of Spotifys song features, simultaneously segmenting artists and songs with similar characteristics.
