Title:
MODEL MANAGEMENT SUPPORT SYSTEM AND METHOD
Document Type and Number:
WIPO Patent Application WO/2021/256171
Kind Code:
A1
Abstract:
This system calculates the distance between target data distribution and base data distribution for each of one or more second periods that are from among the entire period from the time point at which a first period ends to the time at which a learning model is retrained. For each of the second periods, the target data is, from among the entirety of data for the entire period, data for that second period. The base data is for the first period and is involved in creation of the learning model. The system specifies, as a variation point, the time point attributed to the second period in which the calculated distribution distance is not less than a threshold. The system determines one or more training data candidates from some or all of the entirety of data on the basis of one or more variation points and outputs the one or more training data candidates or meta information thereof for retraining of the learning model.
Inventors:
TAJIMA YOSHIYUKI (JP)
INABA DAISUKE (JP)
MOCHIZUKI YOSHINORI (JP)
INABA DAISUKE (JP)
MOCHIZUKI YOSHINORI (JP)
Application Number:
PCT/JP2021/019404
Publication Date:
December 23, 2021
Filing Date:
May 21, 2021
Export Citation:
Assignee:
HITACHI LTD (JP)
International Classes:
G06N20/00
Foreign References:
| JP2018513490A | 2018-05-24 |
Other References:
ISHIDA, TSUTOMU ET AL.: "Evaluation of Concept Drift Detection Methods for Unlabeled Data in Operation", THE 34TH ANNUAL CONFERENCE OF JSAI, 12 June 2020 (2020-06-12)
Attorney, Agent or Firm:
SUNNEXT INTERNATIONAL PATENT OFFICE (JP)
Download PDF:
Previous Patent: ABNORMALITY DETERMINATION SYSTEM AND POWER GENERATION SYSTEM
Next Patent: SNZN SOLDER AND PRODUCTION METHOD THEREFOR
Next Patent: SNZN SOLDER AND PRODUCTION METHOD THEREFOR