Title:
TRAINING DEVICE, TRAINING METHOD, AND TRAINING PROGRAM
Document Type and Number:
WIPO Patent Application WO/2023/214585
Kind Code:
A1
Abstract:
This training device trains an agent's learning model, and comprises: a reinforcement learning unit for training a learning model so that the remuneration that the agent is granted in a prescribed environment is maximized; an evaluation index value calculation unit for calculating the first and second index values of the learning model; and a model extraction unit for extracting a learning model the number of learning steps of which is greater than or equal to a prescribed value, as a trained model. The model extraction unit selects a trained model among the trained models, the first and second index values of which satisfy a prescribed condition, respectively, as the trained model to be evaluated.
Inventors:
KATAOKA YUJIRO (JP)
ITO MASAYUKI (JP)
MATSUNAMI NATSUKI (JP)
ITO MASAYUKI (JP)
MATSUNAMI NATSUKI (JP)
Application Number:
PCT/JP2023/017150
Publication Date:
November 09, 2023
Filing Date:
May 02, 2023
Export Citation:
Assignee:
MITSUBISHI HEAVY IND LTD (JP)
International Classes:
G06N20/00
Domestic Patent References:
| WO2021064767A1 | 2021-04-08 |
Foreign References:
| JP2022035686A | 2022-03-04 | |||
| JP2019219741A | 2019-12-26 | |||
| CN112016704A | 2020-12-01 |
Attorney, Agent or Firm:
SAKAI INTERNATIONAL PATENT OFFICE (JP)
Download PDF: