PURPOSE: To efficiently bring learning to its end by multiplying the correction quantity of coupling intensity between neurones by a learning factor, and investigating recognition rate for input intended to learn, and determining the learning factor to be used for the following learning by using the recognition rate.

CONSTITUTION: The characters A to E are inputted successively to a network, and each outputted result is investigated, and the value of the learning factor ηis determined for the recognition rate at that time. Subsequently, the learning is executed on time for each character A to E by using this value of η. After that, the characters A to E are inputted again to the network, and the recognition rate is investigated, and the value of η is determined newly. The learning is executed again one time for each of A to E by using the newly determined value of η. Henceforward, the investigation of the recognition rate, the setting of η, and the learning of every one time are repeated alternately, and at the point of time when the recognition rate becomes 100%, the learning is finished. Thus, the learning can be carried forward efficiently.