Cookies

We use cookies to ensure that we give you the best experience on our website. You can change your cookie settings at any time. Otherwise, we'll assume you're OK to continue.

Department of Mathematical Sciences

Seminar Archives

On this page you can find information about seminars in this and previous academic years, where available on the database.

Statistics Seminars: Credal networks for sensitivity analysis in graphical models

Presented by Cassio de Campos, Queen's University Belfast

26 October 2015 14:00 in CM221

Probabilistic graphical models such as Bayesian networks are important
tools in AI for reasoning with uncertainty. The quantification of a
Bayesian network requires sharp (i.e., precise) assessments of the
model local conditional probabilities. Credal networks have been
proposed as a generalization of Bayesian networks based on imprecise
probabilities: credal sets are used instead of single distributions,
thus providing higher expressiveness and allowing different inference
tasks to be performed. In this talk we discuss about credal networks
and how they can be used for global sensitivity analysis of
probabilistic graphical models with respect to perturbations of
parameters. We describe an exact algorithm to check whether MAP
configurations are robust with respect to given perturbations. This
algorithm has essentially the same complexity as that of obtaining the
MAP configuration itself. We apply this approach in two practical
scenarios: the prediction of facial action units using posed images
and the classification task in multiple public real data sets.

Contact sunil.chhita@durham.ac.uk for more information