Title:
NEURAL NETWORK CIRCUIT
Document Type and Number:
Japanese Patent JP3172278
Kind Code:
B2
Abstract:
PURPOSE: To recognize an object body at a high speed by the multilayered neural network circuit which recognizes feature data on a given image, etc., by the arithmetic of a network.
CONSTITUTION: Feature data memories 107-109 of respective layers are stored with feature data on the respective layers. Coefficient memories 101-103 of the respective layers are stored with coupling coefficients of neurons of the respective layers except the final layer. A weight memory 118 is stored with the weight of neurons of the final layer. Address converters 104-106 calculate only the addresses of networks whose coupling coefficients are effective among the networks. A table memory 112 outputs the total coupling coefficient obtained by multiplication by effective coupling coefficients read out of the coefficient memories 101-103 of the respective layers. An accumulating arithmetic unit 113 cumulatively adds the total coupling coefficient in the table memory 112 and the weight in the weight memory 118. Therefore, the arithmetic is performed for only the networks whose coupling coefficients are effective to speed up the arithmetic.
Inventors:
Hiroyuki Nakahira
Masakatsu Maruyama
Shiro Sakiyama
Susumu Maruno
Masakatsu Maruyama
Shiro Sakiyama
Susumu Maruno
Application Number:
JP24287892A
Publication Date:
June 04, 2001
Filing Date:
September 11, 1992
Export Citation:
Assignee:
Matsushita Electric Industrial Co., Ltd
International Classes:
G06F15/18; G06G7/60; G06N3/00; G06N3/063; G06N99/00; G06T1/40; G06T7/00; (IPC1-7): G06N3/00; G06G7/60; G06T1/40
Other References:
Ｎａｔｉｏｎａｌ Ｔｅｃｈｎｉｃａｌ Ｒｅｐｏｒｔ 第４２巻 第６号（平成８年１２月）ｐ７３６−７４２ 中平博幸 他「適応増殖量子化ニューロチップ」
電子情報通信学会技術研究報告 Ｖｏｌ．９３ Ｎｏ．２３３ １７．０９．１９９３ Ｐ３９−４６ 中平博幸 他「量子化ニューロンを用いたディジタルニューロプロセッサ」
電子情報通信学会技術研究報告 Ｖｏｌ．９３ Ｎｏ．２３３ １７．０９．１９９３ Ｐ３９−４６ 中平博幸 他「量子化ニューロンを用いたディジタルニューロプロセッサ」
Attorney, Agent or Firm:
Hiroshi Maeda (2 outside)