Bayesian networks : a better than frequentist approach for parametrization, and a more accurate structural complexity measure than the number of parameters

2005
Gelly, Sylvain
Teytaud, Olivier

Abstract: Résumé : We propose and justify a better-than-frequentist approach for bayesian network parametrization, and propose a structural entropy term that more precisely quantifies the complexity of a BN than the number of parameters. Algorithms for BN learning are deduced.