Title:
INFORMATION PROCESSOR AND METHOD, LEARNING DEVICE AND METHOD, AND PROGRAM
Document Type and Number:
Japanese Patent JP2007265345
Kind Code:
A
Abstract:
To learn or generate a long sequence in an RNN.
In the RNN (recurrent type neutral network) 41, the next input to an input node 61-i is generated by adding output of an output node 64-i to input to the input node 61-i just before it by prescribed ratio and the next input to a context input node 62-k is generated by adding output of a context output node 65-k to input to the context input node 62-k just before it by prescribed ratio. The present invention is applicable, for example, to an information processor using the recurrent type neural network.
COPYRIGHT: (C)2008,JPO&INPIT
Inventors:
NISHIMOTO TAKANOSUKE
TANI ATSUSHI
ITO MASATO
TANI ATSUSHI
ITO MASATO
Application Number:
JP2006093108A
Publication Date:
October 11, 2007
Filing Date:
March 30, 2006
Export Citation:
Assignee:
SONY CORP
RIKAGAKU KENKYUSHO
RIKAGAKU KENKYUSHO
International Classes:
G06N3/08
Attorney, Agent or Firm:
Yoshio Inamoto
Previous Patent: SYSTEM, METHOD AND PROGRAM FOR MANAGING DOCUMENT
Next Patent: PROCESSOR, PROGRAM, AND FORM FOR ELECTRONIC PEN
Next Patent: PROCESSOR, PROGRAM, AND FORM FOR ELECTRONIC PEN