Computing with activities III: chunking and aspect integration of complex situations by a new kind of kohonen map with whu-structures

In this paper we present a new theory and give first examples of a new kind of network structure, which shows oscillatory behaviour and codes the aspects of complex situations more exactly than it can be expected by a common neural net. It will be shown, that by introducing lateral `wide hook up??-structure (resp. whu-structures) on SOMs, an aspect integration can be done which generates very filigree classification and/or action strategies and makes it possible to supervise time trajectories. In addition these structures enable to integrate several aspects of a scene into a mega-concept corresponding to the Chunking process of the brain. The categorization of concepts will hereby be done by active whu-structures which become dominant since they sum up the activities of the involved neurons instead of considering one and only one high-active winner neuron.