Venerdi' 19 settembre 2014, alle ore 11, presso l'aula Beltrami del Dipartimento di Matematica dell'Universita' degli Studi di Pavia (via Ferrata 1, Pavia)
Nicolas Bousquet (EDF Research and Development, Dpt. of Industrial Risk Management, Chatou)
terrà il seminario dal titolo
*Some ideas about prior elicitation in Bayesian risk and reliability analysis*
When dealing with small samples and large uncertainties in statistical risk and reliability analysis (typically when a highly reliable industrial component is studied, or when focusing on extreme events with the appropriate statistical theory), the help of Bayesian analysis, allowing to incorporate expert assessments based on technical knowledge, past ground experiments, meta-analyses.. , can be useful. Technically, it is needed to encode prior information into one or several statistical distributions. We review here some difficulties arising in practice, both technical and from the point of view of a control authority. Some methodological ideas about how eliciting a "good" prior are discussed, especially in the context where several sampling models can be compared (and their priors must be calibrated to do so). A major idea is speaking in terms of virtual data. Applications will deals with conjugate priors, some discrete models and Weibull, Gamma and extreme values models.
Gli interessati sono invitati a partecipare.