JANGIN TRAINING METHOD AND DEVICE FOR SUPERVISED LEARNING WITH SMALL DATA SET

Disclosed are an artisan training method and device for supervised learning based on the small amount of data. According to an embodiment of the present invention, the artisan training method for supervised learning based on the small amount of data includes: a first step of extracting label data ha...

Full description

Saved in:
Bibliographic Details
Main Authors KIM HYE MEE, BAE HYE RIM, SIM SUNG HYUN
Format Patent
LanguageEnglish
Korean
Published 18.05.2022
Subjects
CALCULATING
COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
COMPUTING
COUNTING
PHYSICS
Online AccessGet full text

Cover

More Information
Summary:Disclosed are an artisan training method and device for supervised learning based on the small amount of data. According to an embodiment of the present invention, the artisan training method for supervised learning based on the small amount of data includes: a first step of extracting label data having the same label from all label groups; a second step of learning a craftsman network related to the label using the extracted label data; and a step of adding the learning data output from the learned craftsman network as novel label data. The present invention learns the craftsman network based on WGAN-GP and generates/secures the learning data. 소량 데이터 기반의 지도학습을 위한 장인 트레이닝 방법 및 장치가 개시된다. 본 발명의 일실시예에 따른, 소량 데이터 기반의 지도학습을 위한 장인 트레이닝 방법은, 전체 라벨 그룹으로부터, 동일한 라벨을 갖는 라벨 데이터를 추출하는 제1 단계; 상기 추출된 라벨 데이터를 이용하여, 상기 라벨에 관한 크래프트맨 네트워크(Craftsman Network)를 학습하는 제2 단계; 및 상기 학습된 크래프트맨 네트워크로부터 출력되는 학습 데이터를, 새로운 라벨 데이터로서 추가하는 단계를 포함하여 구성한다.
Bibliography:Application Number: KR20200150343