To suppress a probability that produces erroneous recall by generating a random number or a series that is similar to it and converting each component of an input pattern to be given a neural network in accordance with its value.

An input pattern converting device 6 consists of a processor, a memory which stores programs that are executed by the processor or an electric circuit, etc., and has each function of a random number generating means 61 and an input pattern component converting means 62. And, the means 61 receives an input pattern, it generates mutually independent random numbers which are as many as the number of input pattern components, i.e., the number of neurons of an association storage neural network. The means 62 converts the value of an input pattern component in accordance with the value of each random number produced by the means 61. Further, the means 62 sets the value of each component of the converted input pattern as an initial value of an internal state of each neuron of the association storage neural network 1.