PURPOSE: To increase the processing sped of the input information using the analog value, to improve the learning accuracy, and to increase the synapse connection weight converging speed by using a self-learning function to change and converge the synapse connection weights contained in each neuron circuit.

CONSTITUTION: A forward processing means A receives plural forward input analog signals xi and performs the product sum processing operations between these signals xi and their corresponding synapse connection weight wi for production of the forward output analog signals (y). An internal processing means B receives plural adverse input error signals Δyi and performs the due processing between these signals Δyi and the forward output analog signals (y) for production of the internal correction signal. An updating means C performs the processing between the internal correction signal and each signal xi to calculate the correction value of each synapse connection weight. Then the means C corrects and updates the corresponding weight wi based on the calculated correction value. Then an adverse processing means d performs the processing between the internal correction signal and each weight wi for production of plural adverse output error signals Δxi. Thus it is possible to increase the processing speed of the input information using the analog value, to improve the learning accuracy, and to increase the synapse connection weight converging speed respectively.