Creating a Bayesian Inference Engine for KAMET

Probabilistic networks have shown to be an important tool to work with probability-measured uncertainty. However, quality of probabilistic networks depends on a correct knowledge acquisition and modelling. KAMET is a model-based methodology designed to manage knowledge acquisition from multiple knowledge sources [1]. After this methodology is applied, a graphical model representing causal relations is obtained. Up to now, all inference methods developed for these models were rule-based, and therefore eliminated most of the probabilistic information. In this article we present a way to combine the benefits of probabilistic networks and KAMET, and reduce their problems, through a transformation that generates directed acyclic graphs, the basic structure of Bayesian networks [3], and conditional probability tables, from KAMET models. Thus, inference methods for probabilistic networks may be used in KAMET models.