PURPOSE: To accelerate the converging speed of an error function by performing smoothing by adding or deleting a prescribed signal processing element on an existing network system.

CONSTITUTION: An instruction to configure a neural network is issued from a user to a neural net part 10 corresponding to each learning target, and the neural net part 10 is configured corresponding to a configuration instruction. Also, the requirement of smoothing is judged by a smoothing requirement decision part 80 based on the initial value of coupling load set by a neural net setting part and the factor of the configuration instruction, etc., before the learning is performed. Furthermore, when it is judged that it is required to perform the smoothing by the smoothing requirement decision part 80, the addition or the deletion of the prescribed signal processing element is performed by a prescribed signal processing element change part 60.