Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
LEARNING DATA SET OPTIMIZATION METHOD FOR SIGNAL IDENTIFICATION DEVICE AND SIGNAL IDENTIFICATION DEVICE CAPABLE OF OPTIMIZING THE LEARNING DATA SET
Document Type and Number:
WIPO Patent Application WO/2008/016109
Kind Code:
A1
Abstract:
Provided is a method processed by a selector. The selector selects respective elements constituting a learning data set from a data set source. Each of the elements of the source is feature data extracted via a converter in advance and assigned to one of the categories. The selector calculates an alienation degree of each element of the source so as to obtain an average alienation degree. When output neurons of an out put layer of a neural network are correlated to different categories in all the categories indicated by the output layer, the selector includes all the element of the source corresponding to the category of the minimum average alienation degree in the selection from the source to the learning data set. Moreover, the selector removes all the elements of the source corresponding to all the remaining different categories.

Inventors:
HASHIMOTO, Yoshihito (1048, Oaza-Kadom, Kadoma-shi Osaka 86, 5718686, JP)
Application Number:
JP2007/065172
Publication Date:
February 07, 2008
Filing Date:
August 02, 2007
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
MATSUSHITA ELECTRIC WORKS, LTD. (1048, Oaza-Kadoma Kadoma-shi Osaka, 86, 5718686, JP)
松下電工株式会社 (〒86 大阪府門真市大字門真1048番地 Osaka, 5718686, JP)
International Classes:
G06N3/00; G06N3/08; G10L11/00; G06N3/00; G10L11/00
Attorney, Agent or Firm:
NISHIKAWA, Yoshikiyo et al. (Hokuto Patent Attorneys Office, Umeda-Daiichiseimei Bldg. 5F 12-17, Umeda 1-chome, Kita-ku, Osaka-sh, Osaka 01, 5300001, JP)
Download PDF: