[Forum SIS] Annuncio Seminario @DSS-Sapienza: A REVIEW OF BAYESIAN VARIABLE SELECTION METHODS (prof. Fouskakis)

[Luca Tardella]

--------------------------------------------------------------------------
Dipartimento di Scienze Statistiche
Sapienza Università di Roma
--------------------------------------------------------------------------
Prof. Dimitris Fouskakis (National Technical University of Athens, Greece)
--------------------------------------------------------------------------
A REVIEW OF BAYESIAN VARIABLE SELECTION METHODS AND
INTRODUCTION TO THE POWER-EXPECTED-POSTERIOR PRIOR METHODOLOGY

Venerdì 30 Settembre 2016, ore 11.00

aula 34 - IV piano - Edificio CU002 - Dipartimento di Scienze Statistiche
Città Universitaria - piazzale Aldo Moro 5, Roma

Abstract: The problem of variable selection is one of the most fundamental
and widespread model selection problems in statistics. There are numerous
classical approaches to variable selection that have several drawbacks;
exact significance level cannot be calculated and more importantly model
uncertainty is ignored. This talk will initially present how the Bayesian
community deals with the variable selection problem. Under the Bayesian
framework this problem is transformed to the form of parameter estimation:
rather than searching for the single optimal model, a Bayesian will attempt
to estimate the posterior probability of all models within the considered
class of models. Then using Bayesian model averaging techniques we end up
with a coherent mechanism for accounting for model uncertainty.
Focus will be given on Objective Bayesian methods in which vague prior
information is assumed. We will revisit the Expected-Posterior Prior (EPP)
methodology and we will introduce two variants: the
Power-Expected-Posterior (PEP) Prior and the
Power-Conditional-Expected-Posterior (PCEP) Prior. Under the first
approach, ideas from Power-Prior and Unit-Information Prior methodologies
are combined to simultaneously (a) produce a minimally-informative prior
and (b) diminish the effect of training samples. Under the second approach,
g-priors are used with an extra hierarchical level that accounts for the
imaginary data uncertainty. Finally we will investigate the use of
sufficient statistics as a way to redefine the EPP and PEP prior. By this
way we may reduce the dimensionality of the problem drastically and
therefore we simplify computations.
This work contributes to the theoretical understanding of Objective
Bayesian Model Selection methodologies, that are currently becoming
increasingly practical.

Link al PDF dell'abstract del seminario ->
http://www.dss.uniroma1.it/it/node/6591
Link alla mappa di Sapienza  -> http://www.virtualtour.uniroma1.it
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.stat.unipg.it/pipermail/sis/attachments/20160929/c13c15ad/attachment.html>