Title:
NEURAL NETWORK LEARNING METHOD AND SIGNAL PROCESSOR USING THE SAME
Document Type and Number:
Japanese Patent JP3102918
Kind Code:
B2
Abstract:
PURPOSE: To apply this method even to the case that the output signal of a neuron unit(NU) is expressed in two or large number of variables and to improve learning accuracy.
CONSTITUTION: A coupling coefficient storage means 101 stores the coupling coefficient of each coupling between the NUs. An input/output sensitivity coefficient generating means 107 generates an input/output sensitivity coefficient 106 for each coupling between the NUs by using the coupling coefficients of all the coupling receiving an input signal and an input signal inputted to the NUs via the coupling. An internal error generating means 111 generates an internal error 109 by taking the arithmetic sum of error signals 108 received by the NUs. A coupling coefficient changing means 114 changes the coupling coefficient 103 based on the input/output sensitivity coefficient 106. the internal error 109, and the input signal 102.
Inventors:
Shuji Motomura
Hirotoshi Eguchi
Toshiyuki Furuta
Hirotoshi Eguchi
Toshiyuki Furuta
Application Number:
JP20342091A
Publication Date:
October 23, 2000
Filing Date:
July 19, 1991
Export Citation:
Assignee:
株式会社リコー
International Classes:
G06F15/18; G06G7/60; G06N3/04; G06N3/063; G06N3/08; G06N99/00; (IPC1-7): G06G7/60; G06F15/18
Domestic Patent References:
| JP1318180A | ||||
| JP1193982A |
Attorney, Agent or Firm:
Makoto Suzuki (1 person outside)