SYNTHETIC NERVE NETWORKS.

Most models of pattern recognition processes are conceived without regard to the difficulties which might be anticipated if one had to fabricate a full scale engineering model embodying the proposed principles. This report discusses recognition systems which take into account the fabrication limitations one would expect to encounter in the construction of extremely high density microelectronic pattern recognition systems. The resulting analog systems respond to transient patterns and the various artificial neurons within them exhibit functional behavior comparable to that found in biological prototypes. The application of these principles is first discussed in terms of a visual processing system which would exhibit many of the properties attributed to nerve fibres in the visual systems of frogs and cats, including those line sensing properties attributed by Hubel to fibres in the cats visual cortex. The manner in which these same principles can be applied to the problem of sound recognition is then considered. Methods of realization, and an important function of learning in such systems, are also discussed. Author