This project is inspired by one main property of the nervous system: Modularity. The concept of modularity is investigated in the context of artificial neural networks.
First well known architectures and learning methods such as multilayer feedforward networks trained using the Backpropagation algorithm (see chapter 2) and logical networks (see chapter 3) were examined. The objective was to find networks which can be used as modules in a modular system as well as gaining knowledge of different architectures and their advantages.
In the next step, described in chapter 4, a further assessment of the nature of modularity was carried out. The aim was to find evidence for modular structures in the human nervous system and also to review recent developments in ANNs using the concept of modules.
Based on this knowledge a new modular artificial neural system was designed (see chapter 5), in chapter 6 the implementation is described and the evaluation of the model is given in chapter 7.
During the project a conference paper describing the most interesting parts of the architecture as well as some of results gained in the experiments was written, this is included in appendix A.