발간년도 : [2023]
| 논문정보 |
|
| 논문명(한글) |
[Vol.18, No.2] A Study on the Federated Learning System Architecture in Hybrid Cloud Environment |
|
| 논문투고자 |
Hyeju Shin, Sangwon Oh, Seungmin Oh, Kwangki Kim, Jinsul Kim |
|
| 논문내용 |
Recently, with the explosive increase in data and services, steady growth in cloud computing technology is expected. In particular, hybrid cloud technology is attracting attention as an efficient artificial intelligence (AI) learning method. However, concerns have arisen regarding the severity of security and server overload issues caused by data movement in traditional centralized AI learning methods. To address these challenges, various distributed learning methods have been studied, and a federated learning algorithm capable of generating high-performance models without data sharing has been proposed. Nonetheless, the federated learning algorithm has limitations in terms of the amount of resources required on local devices. Therefore, this paper proposes a system architecture that applies federated learning within an AWS-based hybrid cloud environment, enabling the generation of stable performance models with limited local device resources. To validate the feasibility of the proposed architecture, we evaluated the federated learning model performance in various scenarios considering data distribution and quantity uniformity for image classification tasks. As a result, we confirmed that accuracies of over 98% were achieved for both local and global models without sharing local data. This not only offers a solution to server overload issues due to future data and service growth, but also contributes to providing a wider range of AI services by generating high-performance learning models without sharing sensitive data such as personal information. |
|
| 첨부논문 |
|
|
|
|
|
자료실 > 논문지 
