SELF-LEARNABLE MULTILAYER NEURONETWORK AND LEARNING METHOD

PURPOSE: To enable self-learning by eliminating necessity for a user to input a learnt weighted value by regulating the weighted value of each synapse and performing learning while moving to the next layer. CONSTITUTION: A system is initialized, the weighted value of the synapse becomes '0' on this stage, and the input and output patterns composed of (m) pieces of pattern elements and the inputted maximum number of times of repetition are applied. Then, a repetition time counter 30 counts the number of times of repetition, a comparing means 40 performs comparison for deciding whether the maximum number of times of repetition is equal with the output signal of the repetition time counter 30 and when the output signal of the comparing means 40 expresses that both the signals to be compared are not equal, a pattern counter 50 counts the number of patterns. Thus, the input data of N bits and the desired output data of M bits are received, learning is executed while regulating the weighted value of each synapse so as to generate the output data corresponding to the input data and when any error is generated, the number of times of learning is increased but when any error is generated even after learning is executed prescribed times, self-learning is performed by moving to the next layer.