Title:
FUZZY NEURAL NETWORK DEVICE AND LEARNING METHOD THEREFOR
Document Type and Number:
Japanese Patent JP3129932
Kind Code:
B2
Abstract:
PURPOSE: To provide a fuzzy neural network device which is capable of obtaining an output value even for incomplete input data containing an unknown value in an input parameter and performing learning by using even incomplete learning data.
CONSTITUTION: This device is provided with an input layer 1 outputting the value of an input parameter, membership layers 2 and 3 which is formed by dividing the ranges of the values which the input parameter can take, into plural areas defin the membership function of every area and outputs the membership value of each area in accordance with the output value from the input layer 1 for every input parameter, a rule layer 4 constructing a prescribed rule by the certain areas cooperating with each other between different input parameters and outputting the adaptability for the rule, an output layer 5 outputting the value of an output parameter in accordance with the output value from the rule layer 4 and a membership value setting means 6 setting the membership value corresponding to the unknown value to a prescribed value when a part of the input parameter has an unknown value.
Inventors:
Teruhiko Matsuoka
Takashi Aramaki
Takashi Aramaki
Application Number:
JP11721895A
Publication Date:
January 31, 2001
Filing Date:
May 16, 1995
Export Citation:
Assignee:
Sharp Corporation
International Classes:
G06G7/12; G06F15/18; G06G7/60; G06N3/00; G06N3/04; (IPC1-7): G06N3/00; G06G7/12; G06G7/60
Other References:
１．情報処理学会研究報告 ＶＯＬ．９２，ＮＯ．２５（ＩＳ−３８）ｐ１−９ １９９２
２．シャープ技報 ＮＯ．５１ ｐ２５−３０ １９９１
２．シャープ技報 ＮＯ．５１ ｐ２５−３０ １９９１
Attorney, Agent or Firm:
Eisuke Fujimoto