Advantages of Hierarchical Organisation in Neural Networks

Artificial neural networks, which are inspired by the structure and functioning of the vertebrate-brain, are powerful modelling tools. However, the black-box representation they provide does not allow the usage of the huge accumulation of theoretical knowledge on system dynamics. Similarly, they also do not seem to provide any clue for the symbolic operations typical for the higher functioning mode of the human brain.

In this study a "chaos control" problem is used as a test case to demonstrate the viability of extracting an analytical model from an artificial neural network. The results are used to comment on the advantages of hierarchical organisation not only in artificial but also natural neural networks.