PURPOSE: To improve the learning efficiency in a self-learning system using a neural network.
CONSTITUTION: When a self-learning system starts, a self-evaluating part 21 evaluates by itself whether the control output of its own output unit is ideal or not based on the input information given from the outside end the input/ output information obtained at decision of an action. A learning control pert 22 controls the learning of the control data performed in a neural network 24 based on an evaluation signal (d) of an entire system and a self-evaluation signal (g) of the part 21. If the self-evaluating result of the part 21 is satisfactory, the control data obtained at decision of an action are learnt as they are in the network 24. Meanwhile these control data are not learnt when the self- evaluating result of the part 21 is not satisfactory.
SUGASAKA TAMAMI
OSADA SHIGEMI
