Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
NEURAL NETWORK LEARNING METHOD
Document Type and Number:
Japanese Patent JPH04323763
Kind Code:
A
Abstract:

PURPOSE: To increase the calculating speed of a double loop by comparing the node numbers with each other between the layers adjacent to each other and defining the layer having the larger number of nodes as an inner loop.

CONSTITUTION: A forward loop control part 106 decides the inner and outer sides of the double loop calculation every layers adjacent to each other in a forward calculation state. Meanwhile a backward loop control part 107 decides the inner and outer sides of the double loop calculation every layers adjacent to each other in a backward calculation state. Then a weight setting control part 109 sets the weight so that the weight values are referred to or updated in sequence in both forward and backward calculation states. In such a constitution, one of both adjacent layers that has the larger number of nodes is defined as an inner loop. Thus the calculating speed of a double loop is increased. Furthermore the weight values are set so that the weight values of the connection lines are referred to or updated in sequence for calculation of the double loop.


Inventors:
Hara, Miyuki
Tono, Junichi
Application Number:
JP1991000094121
Publication Date:
November 12, 1992
Filing Date:
April 24, 1991
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
HITACHI LTD
International Classes:
G06F15/18; G06G7/60; G06N3/08; G06N99/00; (IPC1-7): G06F15/18; G06G7/60



 
Previous Patent: JPS4323762

Next Patent: DOCUMENT INPUT DEVICE