PURPOSE: To obtain the error signal between a tutor signal and an output by a simple logic circuit by using a neural network composed of a digital circuit which uses pulse density as a signal.
CONSTITUTION: The AND of the NOT of the tutor signal Kj of a (j)th neuron and the output signal Oj of the (j)th neuron is regarded as the positive component (+)j of the error signal of the (j)th neuron. Further, the AND of the tutor signal Kj and the NOT of the output signal Oj regarded as the negative component (-)j of the error signal. Consequently, the error signal can be processed by the simple gate circuit and the method and device can easily be applied to hardware which hardly represents negative values, thereby improving the learning ability for coupling coefficients.
FURUTA TOSHIYUKI
EGUCHI HIROTOSHI