Unifying graphical models by credal networks: algorithms and applications
Graph-based approaches to uncertainty and statistics usually fall under the headline of graphical models. There are many graphical models. In this proposal we focus on Bayesian networks, qualitative networks, influence diagrams and decision nets, and, moreover, on credal networks. All these models, but the last one, are based on traditional, precise probability, and they aim at computing inferences or decisions under risk (i.e., determinate uncertainty). Credal networks are an imprecise probability model: i.e., a credal net can be regarded as a set of Bayesian nets. Our recent research has provided basic tools to map the following models into each other: credal networks, decision networks, models of incomplete information in Bayesian networks; the same tools are expected to lead to bridging credal and qualitative nets. This project is concerned with exploring the mapping as well as the opportunities originated by the cross-fertilization among these fields of research.
People Sun Yi

Marco Zaffalon



Copyright © 2003 - 2012 IDSIA. All Rights Reserved.