PURPOSE: To easily perform the learning of a recurrent type network for associative storage which includes a hidden unit.

CONSTITUTION: Neural networks corresponding to storage items one to one are prepared. The networks learn the relation between the corresponding storage items and other storage items. Those networks are feedforward type neural networks and an error inverse propagating method which is widely used by the feedforward type networks is employed for learning. The input of a feedforward type neural network after learning and the output of another feedforward type neural network are coupled through a feedback loop 103 to form one recurrent type network.