To provide a device which starts with a non-learning state and progressively bootstraps itself to higher level of adaptive or creative capability through cumulative cycles of experiment and learning.
There is provided a discovery system using a neural network which executes training in the discovery system to be stimulated so as to generate a new output pattern through perturbations in various forms applied to itself, and an evaluating function neural network which performs training on the spot in the system similarly and can relate availability or value thereof to useful or more important new pattern among those patterns in the former network while actuating reinforced learning of such patterns. A feedback mechanism as an option between the former and latter self-learning artificial neural networks is used to accelerate convergence of this system on a useful concept or action plan.
| JPH10507553A | 1998-07-21 | |||
| JPH06259542A | 1994-09-16 | |||
| JPH08225258A | 1996-09-03 | |||
| JPH10507553A | 1998-07-21 |
CSNG200000975004; 高秀 誠 Ｍａｋｏｔｏ ＴＡＫＡＨＩＤＥ: '不完全／不確実情報処理利用技術' 電気学会研究会資料 第９９-１／１０巻, 19990319, 第２３頁, 社団法人電気学会
Toshifumi Onuki