PURPOSE: To make fast the learning processing by providing a unit for threshold to output a fixed value to each layer of an input layer and an intermediate layer one by one and a unit providing mechanism for threshold to set arbitrarily the fixed output value of the unit for the threshold.

CONSTITUTION: A hierarchical neural network is constituted of three layers of an input layer, an intermediate layer and an output layer, and at the input layer and the intermediate layer, each unit for threshold is provided one by one. By providing each unit for the threshold one by one, the number of the weight of an internal coupling is decreased and the calculating quantity is made small. The weight of the internal coupling between the input layer and the intermediate layer is Wih and the weight of the internal coupling between the internal layer and the output layer is Who. The input/output of the unit for the threshold is controlled by the unit setting mechanism for threshold. Here, by the control of the unit setting mechanism for the threshold, the arbitrary fixed output value can be obtained.