The elicitation of prior information may be a resource against the curse of dimensional- ity affecting score-based (Bayesian) structural learning of Bayesian networks. A Bayesian elicitation procedure focused on network features has been recently proposed which is based on a collection of propositions characterizing a-priori plausible structures. In this work we address issues related to revi- sion of the elicited values and we introduce a parameterization derived from log-linear models that may be useful both to assist the expert in the revision of an elicited distribution or in the elicitation of be- liefs characterized by low order interactions among reference features. A case study dealing with breast cancer is shortly reconsidered to illustrate some features of the proposed parameterization. The discus- sion includes issues related to the development of an open source project devoted to the feature-based elicitation.
The revision of elicited beliefs on the structure of a bayesian network / F. M. Stefanini. - STAMPA. - (2009), pp. 1-6. (Intervento presentato al convegno S.CO 09 - Complex Data Modeling and Computationally Intensive Statistical Methods for Estimation and Prediction tenutosi a Milano nel Settembre 2009).
The revision of elicited beliefs on the structure of a bayesian network
STEFANINI, FEDERICO MATTIA
2009
Abstract
The elicitation of prior information may be a resource against the curse of dimensional- ity affecting score-based (Bayesian) structural learning of Bayesian networks. A Bayesian elicitation procedure focused on network features has been recently proposed which is based on a collection of propositions characterizing a-priori plausible structures. In this work we address issues related to revi- sion of the elicited values and we introduce a parameterization derived from log-linear models that may be useful both to assist the expert in the revision of an elicited distribution or in the elicitation of be- liefs characterized by low order interactions among reference features. A case study dealing with breast cancer is shortly reconsidered to illustrate some features of the proposed parameterization. The discus- sion includes issues related to the development of an open source project devoted to the feature-based elicitation.I documenti in FLORE sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.