And/or trees for knowledge representation

Abstract

Graph modelling is a modern branch of probability theory concerned with representations for the probability distributions as a product of functions of several variables as a base for possible ways to store high-dimensional distributions by means of a small number of parameters. During the last years, several attempts have been proposed as alternatives in solving this kind of problems [1]–[4]. In most of the practical applications of interest, dependency structures expressed in terms of probability distributions are too complex to allow convenable representations; in such cases, a possible approach could be realised by approximating them keeping the computations at a certain level of complexity but at a convenable accuracy too.

The aim of the paper is to formulate an informational -based approach in decomposing probability distributions using tree-like structures. Our model will be stated in terms of features or actions meaning possible alternatives and their corresponding response effects.