Online Program Home
My Program

Abstract Details

Activity Number: 156
Type: Topic Contributed
Date/Time: Monday, August 1, 2016 : 10:30 AM to 12:20 PM
Sponsor: Section on Nonparametric Statistics
Abstract #320109
Title: Nonparametric Mixture Models with Conditionally Independent Multivariate Component Densities
Author(s): Didier Chauveau* and Vy Thuy Lynh Hoang
Companies: University of Orleans and University of Orleans
Keywords: nonparametric mixture ; multivariate mixture ; EM algorithm ; multivariate kernel density estimation

Non-parametric multivariate finite mixture models often assume independent coordinates conditional on the subpopulation from which each observation is drawn, so that the dependence structure comes only from the mixture. We propose a more general model where this assumption is relaxed, allowing for conditionally independent multivariate blocks of coordinates with multivariate and nonparametric density functions allowing within-block dependences. To estimate the parameters of this model we present an EM-like algorithm and a nonlinear smoothed majorization-minimization algorithm. The performance of this model and new algorithms is illustrated through several simulations and an actual dataset from an unsupervised clustering perspective.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2016 program

Copyright © American Statistical Association