Título: Sensitivity analysis in graphical models: An algebraic approach
Ponente: Manuele Leonelli (lecturer in the School of Mathematics and Statistics at the University of Glasgow, UK)
Lugar y hora: ETSI Informáticos de la UPM, Aula 3341 (Sala de Juntas), 12:30
Abstract: The assessment of the validity of a statistical model outputs, usually referred to as sensitivity analysis, is a critical task of any applied analysis. This consists of checking that a model produces outputs that are in line with current understanding, following a defensible and expected mechanism. Useful techniques to do so are nowadays well-established for various probabilistic graphical models, in particular for discrete Bayesian networks. In this talk I will provide an overview of the steps of a sensitivity analysis for discrete graphical models, based on a monomial representation of their atomic probabilities. Furthermore, a new information-geometry characterisation of sensitivity analysis for these models will be introduced and its usefulness discussed.