PURPOSE: To attain early convergence in learning and highly precise operation in recollection by evaluating the precision of the learning, and deciding a pattern good in the learning precision to be an optimum teacher pattern.
CONSTITUTION: A teacher pattern defining means 1 generates plural groups of input teacher data and corresponding output teacher data from data in the past, and stores them. A neural network defining means 2 constitutes, at least, one neural network whose one element, at least, among structure, a learning parameter and a transfer function is stipulated. Besides, a learning means 5 executes the learning about this neural network by using the generated teacher pattern. Then, a learning precision evaluating means 3 compares the recollected result recollected as the result of the learning with the output teacher data, and determines precision of the learning, and when evaluation exceeds a predetermined standard, a teacher pattern determining means 4 decides this teacher pattern to be the optimum teacher pattern.
Haruki Inoue
Masahiro Yahiro
Noboru Abe
日立エンジニアリング株式会社