Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
MODEL LEARNING SYSTEM, MODEL LEARNING METHOD, AND MODEL LEARNING PROGRAM
Document Type and Number:
WIPO Patent Application WO/2018/066442
Kind Code:
A1
Abstract:
Provided is a model learning system which learns a partially observed Markov decision process model, in which a parameter estimation unit 81 estimates a parameter which maximizes a lower bound of a weighted marginalization likelihood with regard to observation value data in the model. A variation probability estimation unit 82 estimates a variation probability which maximizes the lower bound of the weighted marginalization likelihood. On the basis of the estimated variation probability, a hidden variation deletion assessment unit 83 assesses whether to delete a hidden variation in the model, and deletes the hidden variation which has been assessed to be subject to deletion. On the basis of the estimated variation probability or the estimated parameter, a convergence assessment unit 84 assesses the convergence of the model.

Inventors:
FUJIMAKI RYOHEI (JP)
IMAIZUMI MASAAKI (JP)
Application Number:
PCT/JP2017/035104
Publication Date:
April 12, 2018
Filing Date:
September 28, 2017
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
NEC CORP (JP)
International Classes:
G06N99/00; G06N7/00
Foreign References:
JP2016522458A2016-07-28
Other References:
UENO, TSUYOSHI ET AL.: "Weighted Likelihood Policy Search", IEICE TECHNICAL REPORT, vol. 112, no. 279, 31 October 2012 (2012-10-31), pages 165 - 170, XP055498438, ISSN: 0913-5685
YOSHIMOTO, JUNICHIRO ET AL.: "Identification of Partially Observable Environment Based on On-Line Variational Bayes Method and Its Application to Reinforcement Learning", IEICE TECHNICAL REPORT, vol. 102, no. 731, 19 March 2003 (2003-03-19), pages 131 - 136, XP055498442, ISSN: 0913-5685
Attorney, Agent or Firm:
IWAKABE Fuyuki et al. (JP)
Download PDF: