Coherent Knowledge Processing at Maximum Entropy by SPIRIT

We present a probabilistic expert system permitting the acqusition and processing of uncertain rules. A joint distribution is computed which fits the constraints given by the rules best in an information theoretic sense. So no external (in-)dependencies between the variables are assumed. The knowledge base is decomposed into a system of marginal distributions to allow local computations. This expert system is applied to build up a probabilistic model using statistical data from a medical investigation in a German town. The model is visualized by an undirected graph and reveals both qualitative and quantitative dependencies....