발간년도 : [2024]
| 논문정보 |
|
| 논문명(한글) |
[Vol.19, No.3] Development of Object Detection Model Optimized for MLOps Based Kubeflow Environment |
|
| 논문투고자 |
Dong-Gil Kim, Tae-Yun Chung |
|
| 논문내용 |
AI(Artificial Intelligence) models tend to perform all steps manually, include set to up environment, extract feature, management resources, and learning models, as the service scale increases. This is a task that requires expertise and relies on professional manpower rather than automated , which can cause the reliability and consistency of the analysis results to deteriorate. Recently, the need for MLOps has become important to solve these problems, and research such as automated , deployment, and management of AI models is with actively conducted in environments using MLOps platform such as Kubeflow. Therefore, in this study, KFP(Kubeflow Pipeline) component code that can proceed with preprocessing and algorithm select required to implement AI models on the Kuberflow dashboard and described in depth the automated monitoring, distribute, and operation process of the YOLO object detection model. This provide in it was possible to increase efficiency by automated tasks that require a lot of time consume and iterative process of models, emphasize be placed on asynchronous deployment and execute of applications, enable adaptability to dynamically changed environment. In addition to, by efficiently manage parameter set be able to minimize the impact of various variables on the analysis result, reduce manpower and time to ensure consistent and superior performance, and check and track the status of how execute result are handled and how they affect them at each stage of the models. |
|
| 첨부논문 |
|
|
|
|
|
자료실 > 논문지 
