PURPOSE: To accelerate learning of a neural network and to simplify the hardware structure by applying an accumulation signal and a high frequency signal component to a neural network after mixing them together and controlling the parameter of the neural network.
CONSTITUTION: A neural network 7 is provided together with an input pattern generator 8, an error detector 9, a parameter control part 10, and a teacher pattern generator 11. In a learning state of the network 7, the output signal of the network 7 is compared with a teacher signal and an error signal 104 is produced. Then the correlation is obtained between the signal 104 and the component of a prescribed high frequency signal 102. The correlation detecting signals 103 are sequentially accumulated in terms of time. At the same time, the accumulation signal 100 is mixed with a high frequency signal component 101 and applied to the network 7. Thus the parameter of the network 7 is controlled. As a result, the network 7 is learnt at a high speed and the hardware structure is simplified.