Document worth reading: “Bayesian Networks, Total Variation and Robustness”

Now that Bayesian Networks (BNs) have become broadly used, an appreciation is creating of merely how essential an consciousness of the sensitivity and robustness of positive objective variables are to modifications throughout the model. When time sources are restricted, such factors impression instantly on the chosen stage of complexity of the BN along with the quantity of missing potentialities we’re able to elicit. Currently most such analyses are carried out as quickly as your complete BN has been elicited and are based mostly totally on Kullback-Leibler data measures. In this paper we argue that robustness methods based as an alternative on the acquainted complete variation distance current straightforward and further useful bounds on robustness to misspecification which are every formally justifiable and clear. We show how such formal robustness considerations may be embedded contained in the technique of setting up a BN. Here we give consideration to 2 particular alternatives a modeller should make: the collection of the mom and father of each node and the number of ranges to determine on for each variable contained in the system. Our analyses are illustrated all by way of using two BNs drawn from the present literature. Bayesian Networks, Total Variation and Robustness