Introduction du Retour d'Experience dans les Reseaux Bayesiens (original) (raw)

Projet IS2, INRIA Rhône-Alpes / EDF R&D, département MRI 655 av. de l'Europe, Montbonnot/ 6, quai Watier 38334 Saint Ismier cedex, Abstract This paper presents experience feedback integration in Bayesian Networks. Bayesian Networks structure and probabilities are designed with experts judgement only. In this paper, we assume that networks structure are fixed and known. With growing experience, data are registered and must be used to improve the model. In order to model the ignorance of an expert, we choose Jeffreys law. Therefore, we choose a procedure to parameter the confidence in experts judgement in a rapid way. This parameter represents an imaginary sample size. We give a simple example of experience feedback integration and an application on system of a Reactor Coolant Plant.