PURPOSE: To obtain the optimum solution at a high speed and with high accuracy.

CONSTITUTION: An NN(neural network) setting device 11 is provided to calculate the output data based on the given input data together with a state setting device 12 which sets the initial state of a parameter, an inputting/setting device 14 which inputs and sets the input data stored in an NN learning data base 13, and an evaluation value computing device 15 which calculates the weight evaluation value of an NN from the output data given from the device 11 and the teacher data and then decides the learning state of the NN. Furthermore an initial learning device 16 is added to surely search the optimum solution like a GA(genetic algorithm), etc., as a learning device which learns the weight of the NN based on the evaluation value together with a detail learning device 17 which has a higher learning speed than the device 16 for the BP(back propagation), etc. Both devices 16 and 17 are switched to each other by the device 15.