Title:
CHARACTERISTICS ESTIMATION METHOD, CHARACTERISTICS ESTIMATION DEVICE, PROGRAM, AND RECORDING MEDIUM
Document Type and Number:
WIPO Patent Application WO/2020/188971
Kind Code:
A1
Abstract:
The present invention highly accurately estimates characteristics of an object using machine learning, regardless of the number of characteristics identified under targeted estimation conditions. Machine learning is performed using descriptive information relating to the configuration of an object, identification information about conditions for identifying characteristics, and characteristics of the object. The machine learning includes pre-learning and relearning. The pre-learning creates both an identification information calculation model, which receives identification information about each first condition and outputs calculated identification information, and a nonlinear estimation result output model, which receives the calculated identification information and either the descriptive information or information obtained from the descriptive information and outputs characteristics estimation results. The relearning updates the parameters in the identification information calculation model on the basis of identification information about a second condition and on the basis of characteristics that have been identified under the second condition, while maintaining constant the parameters in the estimation result output model, and creates an estimation model that estimates, from the descriptive information and the identification information about the second condition, characteristics identified under the second condition.
Inventors:
SHIBATA YOHEI (JP)
Application Number:
PCT/JP2020/000837
Publication Date:
September 24, 2020
Filing Date:
January 14, 2020
Export Citation:
Assignee:
FUJIFILM CORP (JP)
International Classes:
G06N20/00
Foreign References:
| JP2012226732A | 2012-11-15 |
Attorney, Agent or Firm:
ITOH Hideaki et al. (JP)
Download PDF: