HIGHLY PARALLEL COMPUTATION NETWORK EMPLOYING A BINARY-VALUED T MATRIX AND SINGLE OUT PUT AMPLIFIERS

摘要

A HIGHLY PARALLEL COMPUTATIONNETWORK EMPLOYING A BINARY-VALUEDT MATRIX AND SINGLE OUTPUT AMPLIFIERSAbstractAdvantageous neural network realizations areachieved by employing only negative gain amplifiers anda clipped T matrix having conductances Tij which haveonly two values. Preferably, one of these values is apreselected value set by the value of a fixed resistor,and the other value is zero, created simply with an opencircuit. Values for the Tij terms of the clipped Tmatrix are obtained through an iterative process whichoperates on the clipped and nonclipped matrices andminimizes the error resulting from the use of theclipped T matrix.