Knowledge acquisition from networks of abstract bio-neurons

The acquisition of knowledge from trained artificial neural networks and its representation in a logical formalism is a fast-growing subject that has attracted the attention of numerous researchers. This study focuses on the extraction of rules from artificial neural networks incorporating abstract bio-neurons. A model of an abstract bio-neuron represents a bridge between the wide applicability and the higher biological plausibility of neural networks. The issue of rule extraction is addressed at the theoretical level. The presented theoretical material introduces conditions which are generally applicable to rule extraction from nonlinear mappings. The results of the theoretical analysis are applied to three-layer network structures containing abstract bio-neuron computational elements. Appropriate conditions enabling the representation of the network's classification in the form of rules are formulated.