AI 워크플로우의 효율적인 관리를 위한 MLOps 개발 환경에 관한 연구 = A Study on MLOps Development Environments for Effective Management of AI Workflow

다국어 초록 (Multilingual Abstract)

Most AI(Artificial Intelligence) models had difficulty entering the production environment beyond the research environment, and it is very important to adapt quickly to the environment and maintain performance consistently depended on the situation. ML(Machine Learning) or DL(Deep Learning) the lifecycle consists of many complex component such as data ingest, data prep, model train, model tune, model monitoring, and does not end in service distribution, but can update the model through continuous learned to reflect data change or new pattern, and as the trained process of the model becomes complicated and diverse, the need for MLOps(Machine Learning Operations) to manages and standardize it is becomes important. However, because the approach is relatively early, there is a lack of structured literature or guidance, and expertise is required because it include a wide range of technology for commercialize such as container, kubernetes, data preprocess, and model distribution. Therefore, in this study, the MLOps system for open-source computing resource integrated specialized in AI model and high-performance computing research is described in depth based on multiple node, and this provided a system that could reproduce performance under the same condition in different environment to automate time consumed and iterative model of workflow and track how hyper-parameter used in data affect model performance.

참고문헌 (Reference)

• 1 이준희 ; 남재현 ; 김진우, "컨테이너 환경에서의 호스트 자원 고갈 공격 영향 분석" 한국정보보호학회 33 (33): 87-97, 2023
• 2 김정집 ; 임재윤 ; 김성환 ; 정회경, "인공지능 기반 구조물 안전관리 예측 시스템" 한국지식정보기술학회 18 (18): 247-257, 2023
• 3 홍승후 ; 이강윤, "연합학습시스템에서의 MLOps 구현 방안 연구" 한국인터넷정보학회 23 (23): 97-110, 2022
• 4 유지현, "악성 이메일 공격의 사전 탐지 및 차단을 통한이메일 보안에 관한 연구" 한국전기전자학회 25 (25): 672-678, 2021
• 5 손정모 ; 김수민, "시계열 데이터 이상탐지를 위한 AI MLOps Workflow 언어와 플랫폼" 한국컴퓨터정보학회 27 (27): 19-27, 2022
• 6 레 반끄엉 ; 유명식, "사물인터넷 환경을 위한 경량 K3S 도구" 한국통신학회 46 (46): 1958-1964, 2021
• 7 김동길 ; 박용순 ; 박래정 ; 정태윤, "머신러닝 자동화를 위한 개발 환경에 관한 연구" 대한임베디드공학회 15 (15): 307-316, 2020
• 8 김동길 ; 박용순 ; 정태윤, "도커 기반의 실시간 데이터 연계 및 처리 환경을 고려한 빅데이터 관리 플랫폼 개발" 대한임베디드공학회 16 (16): 153-161, 2021
• 9 "https://www.kubeflow.org/docs/components/centr al-dash/overview/"
• 10 "https://www.kubeflow.org/docs/components/cen tral-dash/overview/"
• 11 "https://minikube.sigs.k8s.io/docs/start"
• 12 "https://kubernetes.io/ko/docs/tasks/tools/"
• 13 "https://kubernetes.io/ko/docs/reference/setup-too ls/kubeadm/"
• 14 "https://gitlab-gemscrc.gwnu.ac.kr/dgkim1108/m y-projects/-/tree/main/MLOps"
• 15 "https://github.com/kubeflow/manifests"
• 16 "https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html#docker"
• 17 "https://docs.k3s.io/installation"
• 18 "https://docs.docker.com/engine/install/ubuntu/"
• 19 "https://docs.docker.com/compose/install/"
• 20 D. S. Son, "What is 'MLOps', an innovative tow truck for AI development?" 7 : 1-7, 2021
• 21 이승욱 ; 이재우, "STRIDE 위협 모델링에 기반한 클라우드 컴퓨팅의 쿠버네티스(Kubernetes)의 보안 요구사항에 관한 연구" 한국정보통신학회 26 (26): 1047-1059, 2022
• 22 K. Salama, "Practitioners guide to MLOps: A framework for continuous delivery and automation of machine learning" Google Cloud White 1-36, 2021
• 23 G. Symeonidis, "MLOps - definitions, tools and challenges" 453-460, 2022
• 24 D. Sculley, "Hidden technical debt in machine learning systems" Neural Information Processing Systems 2503-2511, 2015
• 25 D. G. Kim, "Development of processing missing value for the code using correlation" 9-10, 2019
• 26 송지호 ; 김판영 ; 최인빈 ; 정회경, "BERT를 활용한 전자정부표준프레임워크 코드 생성 모델 구축" 한국지식정보기술학회 18 (18): 775-784, 2023