PURPOSE: To easily calculate a synapse load and to easily make a neural network(NN) into hardware by performing learning while adding a prescribed limitation on the synapse load.
CONSTITUTION: During the reproduction of an optical disk, outputs (a) to (d) from a photodetector 3 are stored at a prescribed sampling period for a prescribed period. the prescribed operation is executed based upon the data of each sampling point and its operation result is stored in a memory as a teacher signal. On the sampling point, a synapse load is calculated by back propagation algorithm. In this learning, each synapse load is compensated based upon a difference between an output SO obtained at the time of inputting the outputs (a) to (d) to the NN 10 and the teacher signal by means of simulation. In this case, the synapse load connecting between units is determined with a prescribed limitation by using a non-linear function for compressing a change in the output SO against the change of the synapse load. Thereby the synapse load is frequently converged.