Title:
NEURAL NETWORK
Document Type and Number:
Japanese Patent JP3389684
Kind Code:
B2
Abstract:
PURPOSE: To improve recognition capacity for a complicated recognition target.
CONSTITUTION: In a hierarchical neural network having an input layer S, an intermediate layer A and an output layer R, the output layer R has multiplex constitution provided with three units corresponding to respective recognition categories A to C. In the case of back propagation type learning, correct answer teacher data are applied only one unit having a maximum output value out of output units belonging to the correct answer category A e.g. and teacher data whose learning is delayed are applied to other units belonging to the category A. Consequently learning clearing a boundary between categories can be attained.
More Like This:
WO/2017/070255 | AUTOMATED PATIENT CHART REVIEW SYSTEM AND METHOD |
JP3792882 | FEELING GENERATOR AND FEELING GENERATION METHOD |
JPH02206867 | SORTING LEARNING DEVICE |
Inventors:
Shizuo Nagashima
Yoshihiro Matsuura
Yoshihiro Matsuura
Application Number:
JP16047094A
Publication Date:
March 24, 2003
Filing Date:
July 13, 1994
Export Citation:
Assignee:
KABUSHIKI KAISHA MEIDENSHA
International Classes:
G06F15/18; G06N3/04; G06N3/08; (IPC1-7): G06N3/08; G06N3/04
Domestic Patent References:
JP5135000A | ||||
JP4337406A | ||||
JP581226A |
Attorney, Agent or Firm:
Fujiya Shiga (1 person outside)