Feed-Forward Threshold Logic Nets for Digital Switching and Pattern Recognition

A computer-programmable algorithm is presented for the design of general feed-forward nets of threshold logic gates, which realize arbitrary digital switching functions. The algorithm is proved to be convergent for all switching functions. A simplified version of the algorithm is presented for the case of symmetric switching functions. Both algorithms are proved minimal for single-gate nets; a necessary condition is given for the symmetric algorithm to produce minimal two-gate nets when they exist; and two-gate minimality is also demonstrated for the general algorithm, for a certain class of two-gate switching functions. However, general minimality is not proved. The case of partially defined switching functions is treated. Minor alterations in the general algorithm allow its use as an adaptive pattern recognition procedure, where the input variables represent real numbers rather than Boolean variables. Experimental restults are presented for both digital switching-function realization and analog pattern recognition.