Applications of MLOps in the Cognitive Cloud Continuum

Abstract

Background. Since the rise of Machine Learning, the automation of software development has been a desired feature. MLOps is targeted to have the same impact on software development as DevOps had in the last decade. Objectives. The goal of the research is threefold: (RQ1) to analyze which MLOps tools and platforms can be used in the Cognitive Cloud Continuum, (RQ2) to investigate which combination of such tools and platforms is more beneficial, and (RQ3) to define how to distribute MLOps to nodes across the Cognitive Cloud Continuum. Methods. The work can be divided into three main blocks: analysis, proposal and identification, and application. The first part builds the foundations of the work, the second proposes a vision on the evolution of MLOps then identifies the key concepts while the third validates the previous steps through practical applications. Contribution. The thesis’s contribution is a set of MLOps pipelines that practitioners could adopt in different contexts and a practical implementation of an MLOps system in the Cognitive Cloud Continuum.