Extended self organizing maps for structured domain: Models and learning

Abstract

Modelling of neural networks is still a very interesting and important field in the area of computing models. If input domains of neural networks are data structures represented by graphs and output domain is expected in a similar form, it is necessary to consider it in the process of neural network modelling. We propose four models of extended Self Organizing Maps (SOM) that can be applied to graph data structures as input and output domains together with learning algorithms. Extensions of the SOM model are based on the idea to remember information of connections in data structures (using some context neurons). With regards to the evaluation of developed models and trained structures, we used data from the study programs of the Faculty of Science, P. J. Šafárik University in Košice. We evaluated the ability of models to enumerate output descendants of a node in a graph structure and the ability to interpret structures by developed neural networks. We also evaluated the quality of the developed networks in a learning process.