Connectionist Taxonomy Learning

Abstract

The thesis presents an unsupervised connectionist network using spreading activation mechanism. By means of self-organization, the network is capable of creating a taxonomy of concepts which serves as a backbone for a respective ontology. The system is a biologically inspired constructivist hybrid between connectionist networks using distributed and localist data representation. Unlike most currently developed models it is capable to deal with analog signals and displays cognitive properties of categorization process. The thesis presents the general overview over the system’s architecture and method of network build-up and shows results of several experiments exploring the nature of categorization performed with the use of the described network.