PURPOSE: To reduce errors and to prevent the reliability of an output value from deteriorating by changing the configuration of learning data at the time of the study by means of a back propagation method.
CONSTITUTION: Excitement output Pi of a neuron (i) becomes the input to the next neuron, and n-pair of the pattern of an input signal and the pattern of a teacher signal are prepared as learning data for the study of a neural network propagating the stimulation turned to be a learning data set. The learning data is taken out one by one from this learning data set to adopt the back propagation method. In this case, the learning data as an auxiliary processing for the application of the back propagation method is changed. Thus, the error between the output signal and the teacher signal generated at the time of the learning can be reduced, and the deterioration of the reliability of the output value can be prevented.