What Are The Key Areas Of ML-Ops / DL-Ops In Business Problems For Company Growth Using Cloud Environment?

Abstract

This research delves into the exploration and comprehension of Machine Learning (ML) operations within industrial environments, highlighting the growing necessity for data-driven organizations to adopt AI and ML. Operating and maintaining ML models in industrial production settings pose significant challenges. The integration of DevOps principles has revolutionized how software engineers release products, fostering efficiency and creativity. A parallel trend is observed in the machine learning sphere, where data science teams are beginning to integrate these principles into ML operations, termed MLOps. This literature review aims to shed light on the current obstacles encountered in the productionization of machine learning, drawing upon academic sources to scrutinize the prevailing challenges in MLOps. The focus is twofold: firstly, on the critical role of MLOps principles in industrial contexts, and secondly, on the application of these DevOps principles to enhance the operationalization of machine learning projects.
However, in recent years, the use of Machine Learning (ML) has witnessed a significant increase, but many organizations still face challenges when operationalizing ML. This thesis investigates the current best practices, challenges, and potential solutions associated with developing an MLOps process in the cloud from a RE perspective by exploring the intersection between machine learning operations (MLOps) and Requirements engineering (RE). This thesis aimed to create an artifact that would guide MLOps implementation in the cloud from an RE perspective, thus offering a more systematic approach to managing ML models in production by establishing goals and attitudes toward their development in the future. Three research questions were investigated using the Design Science Research methodology during design artifact creation. The study examined existing barriers to the MLOps process's establishment, found possible ways to overcome these difficulties, and assessed the efficiency. The study followed three cycles, each answering all the research questions but mainly concentrating on one question, allowing initial artifact creation and subsequent refining depending on data collected during each process. By establishing a better in-depth understanding of how these two spheres interact and providing some practical guidance for implementing MLOps processes from the RE perspective, this study advances both MLOps and RE fields. In terms of theoretical evaluations, quality feedback was collected about the artifact. In this regard, one major limitation is that an assessment of the artifact's efficiency in real-life situations needs to be made. As a result, future research should evaluate this artifact's effectiveness by conducting case studies in real-world settings and enhancing its limitations.