Avviso di Seminario 

When

September 8th 2016, 14:30

Where 

Sapienza Università di Roma 

facoltà di Economia

Aula Fanfani, V piano 

Viale del Castro Laurenziano, 9

Speaker 

Federico Camerlenghi

Università Luigi Bocconi, Milano

Titolo 

Distribution theory for hierarchical processes with statistical applications

Abstract

In Bayesian nonparametrics, a large amount of literature has been carried out under the assumption that observations are exchangeable. However in a large variety of applications this may appear too restrictive, a typical example is when data are generated by different experiments: even if these experiments may be related, they have some specific features that induce heterogeneity across observations. In such a situation data are usually divided into different groups, and a more appropriate assumption is partial exchangeability. In particular the construction of dependent random probability measures to deal with partially exchangeable observations has recently attracted great attention in Bayesian nonparametric literature. In this talk we define and investigate general classes of nonparametric priors, namely hierarchical models, based on transformations of completely random measures. We are able to study all the analytical properties, i.e. the random partition structure, the predictive distributions and the posterior characterization of the processes. Finally some relevant applications to species sampling problems and survival analysis will be discussed.

(This is a joint work with A. Lijoi, P. Orbanz and I. Prunster.)