Title:
ディープラーニングモデルのチューニング方法およびシステム
Document Type and Number:
Japanese Patent JP7371188
Kind Code:
B2
Abstract:
To provide a method for tuning a deep learning model using a backbone model included in an information processing system to be executed by at least one processor of a user terminal.SOLUTION: This method includes the steps of: transmitting a service request including at least one of information related to input data of a deep learning model, information related to a tuning object layer and learning state information to an information processing system; receiving tuning data calculated by using a backbone model from the information processing system as a response to the service request; and using the tuning data to update the deep learning model. Also, the tuning data includes at least one of a parameter of the backbone model, an input value to a specific layer, an output value of the specific layer and a fluctuation portion of the parameter.SELECTED DRAWING: Figure 1
Inventors:
Lee Gi-chan
Ko Seok-hyun
Park Woo Myung
Ko Seok-hyun
Park Woo Myung
Application Number:
JP2022141858A
Publication Date:
October 30, 2023
Filing Date:
September 07, 2022
Export Citation:
Assignee:
NAVER Corporation
International Classes:
G06N3/082
Domestic Patent References:
| JP2020160764A | ||||
| JP2022523207A |
Foreign References:
| WO2020026741A1 |
Attorney, Agent or Firm:
Tadashige Ito
Tadahiko Ito
Osamu Miyazaki
Tadahiko Ito
Osamu Miyazaki