Systems Methodology for Classical Neural Nets at Various Levels

Abstract

The different aspects of classical neural nets are treated here at the light of Systems Theory. First, we consider McCulloch-Pitts-Blum formalisms, which are regarded as biological counterparts of logical machines (automata). Systems Theory inspired the introduction of state transition and functional matrices, to treat problems of analysis, synthesis, stability and oscillations. The so called McCulloch Program I is completed by the development of systems technics to allow the neural synthesis of arbitrary probabilistic automata. Next, nets to integrate sensorial functions, as well as intermodal integration and effector action are considered, which corresponds to a level higher than the “molecular” neuron-like computing element. Finally, at the highest level, modules for the so called generalized robotic-like behavioral systems are presented, in line with the known McCulloch-Problem II. This includes a diagram for a behavioral system under a command and control subsystem of the type of the reticular formation of vertebrates.