In this section some aspects of the model are considered. This analysis presents some background to the network behaviour, and also addresses the question: why does the network work? - focusing is on aspects of speed, learning and generalization. Theoretical limitations of the model are also considered.
The structure of the proposed network is based on modules with each input being connected to a single module. This feature results in a faster training procedure and advances the generalization performance, but also introduces certain limitations. The number of connections within the modular network is significant less than in a comparable monolithic architecture. A concrete comparison is given in chapter 7.