Current Result Document :
| 한글제목(Korean Title) |
FADA 및 CORAL을 사용한 연합 도메인 적응 기반 인간 동작 인식 |
| 영문제목(English Title) |
Federated Domain Adaptation based Human Activity Recognition using FADA and CORAL |
| 저자(Author) |
이그나티우스 이완
베르나르도 누그로호 야흐야
이석룡
Ignatius Iwan
Bernardo Nugroho Yahya
Seok-Lyong Lee
|
| 원문수록처(Citation) |
VOL 49 NO. 01 PP. 1014 ~ 1016 (2022. 06) |
한글내용
(Korean Abstract) |
|
영문내용
(English Abstract) |
Federated learning has provided a way to collaboratively learn between multiple instances. However, trained machine learning models may not be suitable to new devices. The different activities and devices cause domain shift problem which is referred to different distribution between target and source data. In this work, we identify the problem of federated domain adaptation in classifying human activities through wearable devices and aim to enhance the model performance when the model transferred from source domain to target domain. We leverage adversarial approach for learning domain difference and discrepancy loss through correlation alignment (CORAL) between two domains. Experiment is conducted using different datasets that contain different setting on the same activities. |
| 키워드(Keyword) |
|
| 파일첨부 |
PDF 다운로드
|
