-----------------------------------------------------------------
University of Edinburgh Statistics & BioSS Seminar programme
-----------------------------------------------------------------
Friday 31st of January, at 3:10pm, room JCMB5323.
Speaker: Chris Holmes (University of Oxford)
Title:
Approximate Probabilistic Models and Robust Inference
Abstract:
The rise of approximate probabilistic modeling approaches, such as INLA,
Variational Bayes, ABC, Composite Likelihoods, Gibbs posteriors, etc.
has been driven by the computational constraints of modern data analysis
problems, including applications involving “big-data”. Statisticians are
taught from an early stage that “all models are wrong” but little formal
guidance exists on how to assess the impact of model misspecification,
or how to proceed when optimal actions appear sensitive to model
fidelity. We present formal methods for assessing robustness under model
misspecification by quantifying stability of conclusions to local
perturbations of the approximating model within an information
divergence (Kullback-Leibler ball) around the approximating model. We
derive analytic results for local-minimax outcomes and present Monte
Carlo methods using Bayesian nonparametric priors to sample from
distributions (models) of a given KL divergence from the approximating
model. Our work draws on recent research in the robust control,
macroeconomics and financial mathematics literature. We adopt a Bayesian
approach throughout although the methods are agnostic to this position.
This seminar is part of the Maxwell institute seminar series.
Website: http://www.maths.ed.ac.uk/events/statistics
Map: http://www.ed.ac.uk/polopoly_fs/1.20490!/fileManager/campus_maps.pdf
Please note: seminars may be recorded
--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
You may leave the list at any time by sending the command
SIGNOFF allstat
to [log in to unmask], leaving the subject line blank.
|