Title:
【発明の名称】自己学習システムの安定化方式
Document Type and Number:
Japanese Patent JP2856259
Kind Code:
B2
Abstract:
PURPOSE:To stably operate the system for reinforced learning by a network for tutored learning by alternating the execution and learning of a neural net work for action determination and a neural network for evaluated value learn ing. CONSTITUTION:In addition to an input pattern to the system, an external output to which a noise is added is inputted to the neural network 16 for evaluation learning at the time of the learning, and the output of the neural network 15 for action determination is inputted at the time of the execution. The output of the network 15, however, is obtained by learning the external output of the system and the evaluation when no noise is added to the output of the network 15 is accurately outputted by the network 16 at the time of the execution, so that noise control becomes accurate. In the learning, output-side tutor data to some input to the network 16 is determined unequivocally at the time of the learning to stabilize the operation of the network 16.
More Like This:
| JPH06131318 | NEURON ELEMENT |
| JPH06203005 | HIGH-SPEED DIVIDED NEURAL NETWORK AND METHOD FOR CONSTRUCTING IT |
| JPH06110924 | INFORMATION RETRIEVAL PROCESSING SYSTEM |
Inventors:
SAGA KAZUSHIGE
SUGASAKA TAMAMI
OSADA SHIGEMI
SEKIGUCHI MINORU
ASAKAWA ATSUKO
SUGASAKA TAMAMI
OSADA SHIGEMI
SEKIGUCHI MINORU
ASAKAWA ATSUKO
Application Number:
JP1922890A
Publication Date:
February 10, 1999
Filing Date:
January 31, 1990
Export Citation:
Assignee:
FUJITSU KK
International Classes:
G06G7/60; G06F15/18; G06N3/08; G06N99/00; (IPC1-7): G06F15/18
Other References:
佐賀一繁，菅坂玉美，関口実，長田茂美，「自己学習方式の一検討」１９８９年電子情報通信学会秋季全国大会講演論文集，ｐ．６−２７８〜６−２７９（１９８９）
Attorney, Agent or Firm:
Yoshiyuki Osuge (1 outside)