LEARNING METHOD FOR NEURAL NETWORK

PURPOSE:To provide the learning method for a neural network, which can easily and certainly learn the network even when the number of categories K to be identified are many. CONSTITUTION:The learning method has the neural network consisting of an input layer consisting of not less than one input units inputting a learning pattern and an output layer consisting of K pieces of output units provided in accordance with the K(K>=2, but 'k' is an integer)-number of categories to be identified, a teacher signal generation part 8 geneating a teacher signal Tk ('k'=1, 2,...,K) being the output ideal value of the output unit, a scale factor dicision part 14 deciding a coefficient N in accordance with the number of the categories K and a network learning part calculating a differential Kc betweena teacher signal Tc given to the output unit corresponding to a category (c) (1