Romane Giard

We consider the statistical modeling of a dataset. Gaussian Mixture Models are commonly used for this purpose. In particular, they enable clustering, that is, the separation of data into several groups, which has many applications. 
The estimation of a Gaussian Mixture Model raises both theoretical and algorithmic questions. We study a method based on solving an optimization problem over the space of measures, and provide reconstruction guarantees in this setting. 

Keywords: Mixture Models;  sparse regularization; Fisher-Rao geometry