Title:
【発明の名称】データ処理装置およびその適正化方法
Document Type and Number:
Japanese Patent JP2784497
Kind Code:
B2
Abstract:
PURPOSE:To surely attain an escape from the local minimum value by applying a prescribed sloping deflection to the distribution of neuron threshold value for a prescribed period set when a fixed time elapses from the start of learning. CONSTITUTION:A data processor applies a smooth sloping deflection to the distribution of neuron threshold value for a prescribed period set when a fixed time elapses from the start of learning. At the same time, the threshold value of the neuron that produces a significant output at a certain time point is increased once up to the extreme value and then forcibly reduced down to the value set at a certain time point while a fixed input is applied to the data processor. The weight is controlled for a period when the threshold value is forcibly changed. Thus it is possible to surely secure an escape from the local minimum value with a mechanism subject to the function of an organism system. Then the data processor is effectively rationalized.
Inventors:
TAKATORI SUNAO
KUMAGAI RYOHEI
YAMAMOTO MAKOTO
MATSUMOTO KOJI
KUMAGAI RYOHEI
YAMAMOTO MAKOTO
MATSUMOTO KOJI
Application Number:
JP21393589A
Publication Date:
August 06, 1998
Filing Date:
August 19, 1989
Export Citation:
Assignee:
YOZAN KK
SHAAPU KK
SHAAPU KK
International Classes:
G06F15/18; G06N3/08; G06N99/00; (IPC1-7): G06F15/18
Domestic Patent References:
| JP318964A |
Other References:
【文献】甘利俊一,「神経回路網の数理」.産業図書(1978)
Attorney, Agent or Firm:
Yamamoto Makoto