CMStatistics 2019: Start Registration
View Submission - CMStatistics
B0653
Title: Semiparametric copula-based mixture models Authors:  Gildas Mazo - INRA (France) [presenting]
Abstract: Faced with non-Gaussian clusters (or more generally, non-elliptical clusters), natural alternatives to Gaussian mixture models include copula-based mixture models and nonparametric mixture models. The first can be difficult to calibrate. The last are built on a conditional independence assumption. With semiparametric copula-based mixture models, one aims at getting benefits from both approaches: exploiting the underlying dependence structure and getting some nonparametric flexibility. Semiparametric copula-based mixture models are presented and algorithms are given to do the inference. These are illustrated on simulated and real data.