Title:
NEURAL NETWORK MODEL AND NEURAL NETWORK DEVICE
Document Type and Number:
Japanese Patent JP3262857
Kind Code:
B2
Abstract:
PURPOSE: To provide the neural network model and neural network device which can be applied for various optimizing problems or recognizing problems and accelerate convergence to a solution without generating any vibrating solution.
CONSTITUTION: Concerning the mutual coupling neural network in which respective elements define saturated linear functions as input/output functions, the block sequential neural network model is provided for dividing the element into several blocks, synchronously updating the state of the element inside the block and successively (asynchronously) updating the state for each block and further, this neural network device is composed of a light emitting element array 201 for performing vector matrix arithmetic by using light, threshold value processing part 206 for operating threshold values, coefficient accumulation part 209 for operating a cutting width, saturated linear arithmetic part 207 for performing the saturated linear functions, and block operation part 208 for selecting the block to be updated and updating the state.
Inventors:
Ikutoshi Fukushima
Fumiyuki Shiratani
Takeshi Hashimoto
Fumiyuki Shiratani
Takeshi Hashimoto
Application Number:
JP26848592A
Publication Date:
March 04, 2002
Filing Date:
October 07, 1992
Export Citation:
Assignee:
Olympus Optical Co., Ltd.
International Classes:
G06E3/00; G06G7/60; (IPC1-7): G06G7/60; G06E3/00
Domestic Patent References:
| JP223324A | ||||
| JP2125362A | ||||
| JP2306365A |
Other References:
白谷文行、山本公明，ブロックシーケンシャルに動作する神経回路網を用いた組合せ最適化，電子情報通信学会論文誌，日本，ｖｏｌ．Ｊ７７−Ｄ１１ Ｎｏ．１，ｐ．２０４−２１０
Attorney, Agent or Firm:
Ryuyoshi Abe (7 outside)