PURPOSE: To omit unnecessary learning to increase the learning speed in a neural network device.

CONSTITUTION: Upward correction of the synapse transfer efficiency between neurons 18H (16H), whose outputs are saturated in the high level, out of neurons 18 (16) of an intermediate layer (output layer) and each neuron 20 (18) of an input layer (intermediate layer) is inhibited, and downward correction of the synapse transfer efficiency between neurons 18L (16L), whose outputs are saturated in the low level, out of neurons 18 (16) of the intermediate later (output layer) and each neuron 20 (18) of the input layer (intermediate layer) is inhibited. Thus, learning in the direction where the output of each neuron is not changed is not performed in a neural network 12. That is unnecessary learning which does not have an influence upon success in learning is not performed to reduce the calculation volume.