Aprendizaje de conceptos a partir de representaciones basadas en grafos

Resumen

In this paper a new system of learning from examples is presented. The input data are attribute-value arrays that codify observations of a given world, where the attributes can be linear, nominal or structured ones. The system proceeds in there steps. First of all we build up an inheritance net describing the input data. We endow this net with a semantic interpretation allowing us to read the net as a first draft of classification rules. Later on, by means of a finite automaton, we will rewrite those rules in order to compact its syntatic description; this is our second step. Finally we will apply two generalization principles (closing the interval for linear attributes and climbing in the attribute structure).The rules are generalized so as to obtain the definite intensional description of the concepts so learned. To close the paper we also present different experiments made with the system.