Bayesian Networks

In the mid-1980s a new trend in probabilistic reasoning with uncertainty in knowledge-based systems became discernable taking a graphical representation of knowledge as a point of departure. We use the phrase network models to denote this type of model In the preceding sections, we have concentrated primarily on models for plausible reasoning that were developed especially for knowledge-based systems using production rules for knowledge representation. In contrast, the network models depart from another knowledge-representation formalism: the so-called Bayesian network. Common synonyms for the formalism are: belief network, probabilistic network, Bayesian belief network, and causal probabilistic network. Informally speaking, a Bayesian network is a graphical representation of a problem domain consisting of the statistical variables discerned in the domain and their probabilistic interrelationships. The relationships between the statistical variables are quantified by means of ‘local’ probabilities together defining a total probability function on the variables. This section presents a brief introduction to network models. In Section 2 we shall discuss the way knowledge is