[Date Prev][Date Next]
[Chronological]
[Thread]
[Top]
No Subject
****************************************************************
CON PREGHIERA DI DIFFUSIONE
****************************************************************
AVVISO SEMINARI
Il Prof J.O. Berger, Purdue University, terra',
presso il Dipartimento di Statistica, Probabilita'
e Statistiche Applicate, Universita' di Roma "La Sapienza",
Piazza Aldo Moro 5, due seminari, secondo il seguente calendario:
Lunedi 20 maggio ore 11, Aula 30
"On the Choice of Hyperpriors in Normal Hierarchical Models"
Lunedi 27 maggio ore 11, Aula 30
"Recent Developments in Bayesian Model Selection"
Tutti gli invitati sono invitati a partecipare.
(Brunero Liseo)
RIASSUNTI
On the Choice of Hyperpriors in Normal Hierarchical Models
Abstract
Hierarchical modelling is wonderful and here to stay, but we usually
"cheat" in choosing the prior distributions for hyperparameters.
By "cheating" I mean that we usually choose hyperparameter priors in
a casual fashion, often feeling that the choice is not too important.
Unfortunately, as the number of hyperparameters grows, the effects
of casual choices can multiply, leading to considerably inferio
r performance. As an extreme but not uncommon example, use of the
wrong hyperparameter priors can even lead to impropriety of
the posterior. Finding a solution to this problem is, unfortunately,
difficult; indeed, it is not even clear how to attack the problem.
In this talk we simply give some illustrations of the problem, and
some "solutions" in special cases. Among the topics to be discussed
along the way are reference priors for covariance matrices,
and propriety and admissibility of priors in exchangeable
hierarchical normal models.
Recent Developments in Bayesian Model Selection
Abstract
There have been rapid advances in default Bayesian model selection
and hypothesis testing in recent years, due to the development of
the "intrinsic Bayes factor" and "fractional Bayes factor" approaches.
This talk will cover some of this recent work, including discussion of
the "median intrinsic Bayes factor", an especially simple and almost
universally applicable method; discussion of situations in which certain
default Bayes factors fail; and simplifications that can result when
the models being considered have suitable invariance structures.
On the Choice of Hyperpriors in Normal Hierarchical Models
Abstract
Hierarchical modelling is wonderful and here to stay, but we
usually "cheat" in choosing the prior distributions for hyperparameters.
By "cheating" I mean that we usually choose hyperparameter priors in a
casual fashion, often feeling that the choice is not too important.
Unfortunately,
as the number of hyperparameters grows, the effects of casual choices can
multiply, leading to considerably inferior performance. As an extreme but
not uncommon example, use of the wrong hyperparameter priors can even lead
to impropriety of the posterior.
Finding a solution to this problem is, unfortunately, difficult;
indeed, it is not even clear how to attack the problem. In this talk we
simply give some illustrations of the problem, and some "solutions" in
special cases. Among the topics to be discussed along the way are reference
priors for covariance matrices, and propriety and admissibility of priors in
exchangeable hierarchical normal models,