Learning processing apparatus and method

The invention provides a learning processing device and method capable of completing learning of a lightweight model in a short time. A learning processing device that obtains a new second learning model from an existing first learning model is provided with: an input unit that obtains a first learning model, which is generated beforehand by learning a first learning data set, and an unpruned neural network (hereinafter referred to as a neural network); an important parameter determination unit (304) that uses the first learning model and the NN to initialize the NN to be learned, and uses the second learning data set and the initialized NN to determine the degree of importance of the parameter in the recognition process of the initialized NN; a new model generation unit (306) that generates a second NN by performing pruning processing for deleting an unimportant parameter from the initialized NN using the degree of importance of the parameter; a learning unit that learns the second NN using the second learni