Neural activation ratio based fuzzy reasoning

Presents a class of binary neural nets which seem to be similar to the natural neural nets concerning topological and functional issues. It is the medium activity among the neurons in a given neural area which represents the “amplitude” of the associated variable, making the net insensitive to individual errors. It is proved that fuzzy reasoning is an emergent property of such nets, if predefined membership functions are considered. Macroscopic (net level) fuzzy reasoning emerges from microscopic (neuron level) Boolean operations. Strategies for teaching with real experiments are proposed resulting in a global learning of the net from local neural operations