Speaker
Description
We study the semiparametric estimation of finite location mixtures on the real line when the component density is unknown but symmetric. The method is based on a smoothed maximum likelihood approach over finite-dimensional mixture parameters and an infinite-dimensional functional parameter, leading to a two-step iterative algorithm: a functional update for the component density and a Euclidean update for the mixture weights and locations. The algorithm satisfies a descent property for the population criterion and for its plug-in empirical version. We quantify the effect of numerical approximations in the Euclidean step and show that the resulting practical algorithm retains an approximate descent guarantee. Numerical experiments on synthetic data illustrate the method.
| Thématiques | Semiparametric estimation ; Finite mixture models ; Maximum likelihood ; Kernel smoothing ; Tempered distributions |
|---|---|
| Mes travaux sont plutôt de la statistique ... | Théorique |