Probability Assessment with MaximumEntropy in Bayesian Networks

Bayesian networks are widely accepted as tools for probabilistic modeling. In building Bayesian networks in collaboration with domain experts, the de nition of the graphical structure is usually relatively easy. The assessment of the conditional probability tables (CPT) is often a much more diÆcult task, even when there is a lot of statistical information available as domain knowledge. The problem is that in many cases it is not possible to ll this information directly into the CPTs. In this paper we propose a method to t the CPTs such that the model reproduces the available information as accurate as possible. We will discuss some criteria to do this and we illustrate the methods in an example.