Abstract

The goal of the paper is to introduce a program system, MUDIM, and to show how it can be used for multidimensional probabilistic model construction. The system is being developed with the goal to gain a tool for experimental computations with compositional models which are, in a way, an alternative to Bayesian networks. These models are based on the idea of composing a multidimensional distribution from a great number of low-dimensional ones. When considering knowledge-based systems, this approach quite naturally cope with the difficulty of expressing global knowledge about a field of practise. We have only to work with a system of pieces of local knowledge from which the global knowledge must be assembled