Dear all,
the Young Statisticians Section announces the next pre-ordinary meeting:
19 October 2011,
RSS, 12 Errol Street, London EC1Y 8LX, 3pm - 5 pm
Pre-Ordinary Meeting
Improving algorithms for Bayesian model selection –
Catching up faster by switching sooner
Registration is free, but please register at the RSS webpage or by
email: [log in to unmask] The aim of a Pre-Ordinary Meeting is to make the Read Paper of the Ordinary Meeting that follows more accessible to PhD students, young statisticians and those not familiar with the area discussed in the paper. Please feel free to advertise in your Department and among colleagues.
Schedule:
3:00-3:30pm Philip Dawid, Statistical Laboratory, University of Cambridge
As an introduction to the read paper, I will discuss and compare various statistical model selection criteria, including Bayes factors, AIC, cross-validation and prequential assessment.
3:30-4:00pm Peter Grunwald, Centrum Wiskunde & Informatica, Amsterdam
Prediction and estimation based on Bayesian model averaging is sometimes slow to converge: we give examples showing that, while in the infinite sampe limit, Bayesian predictions become optimal, other methods may perform significantly better at smaller samples. We show how the slow convergence can be understood in terms of Dawid's prequential approach to statistics and how it can be reconciled with claims that Bayesian inference is, in some sense, optimal. This serves as a preparation to the main meeting, in which we propose the switch distribution, a modification of the Bayesian predictive distribution which in many cases achieves faster convergence.
4.00-5.00pm - Tea and coffee
5.00-7.00pm RSS Ordinary Meeting
“Catching up faster by switching sooner: a predictive approach to adaptive estimation with an application to the Akaike information criterion-Bayesian information criterion dilemma”
by Tim van Erven, Peter Grunwald and Steven de Rooij
With best wishes,
Oliver Ratmann
On behalf on the Young Statisticians Section, Royal Statistical Society
You may leave the list at any time by sending the command
SIGNOFF allstat
to [log in to unmask], leaving the subject line blank.
|