Applications of MLOps in the Cognitive Cloud Continuum

In the last decade, the development of software based on artificial intelligence has increased exponentially. The adoption of techniques based on the use of machine learning and deep learning in particular, has made it possible to develop applications and create previously unthinkable solutions. The evolution led by the use of machine learning over classical software development has required new guidelines for the software development lifecycle. While DevOps (a blend of the words "DEVelopment" and "it OPerationS") has traditionally been the standard guideline for software development, it lacks certain steps required by machine learning. This led to the rise of its natural evolution: MLOps (a combination of the words "Machine Learning" and "it OPerationS"). This thesis investigates the application of MLOps in the Cognitive Cloud Continuum by exploiting both empirical methodologies and practical applications in the field of machine learning for software engineering. To do so the primary objective is to identify the risk of embedding open source libraries when developing machine learning-based applications, in particular when following the MLOps principles. Secondly, we aim at investigating what is the Cognitive Cloud Continuum and what are the future implications. Finally, by showing different use cases we compare the differences between the development of software not based on machine learning and the development of software following the MLOps guidelines. The results show that it is possible to compute the risk of embedding open-source software (and therefore libraries) when developing machine learning-based code. The analysis of the literature has identified an increased interest in the concept of Cognitive Cloud Continuum, which has resulted in the development of tools aimed at increasing the level of automation in order to comply with the new requirements of this concept. The results achieved have been showing that the concepts provide promising results. The thesis has successfully demonstrated how to develop a model based on an MLOps pipeline. This thesis contributes to the state of the research by increasing awareness of the concept of Cognitive Cloud Continuum and emphasizing the importance of following MLOps guidelines when developing ML-based software. In the future, devices will be capable of exploiting the computational power of other entities being part of the same environment to carry out multiple tasks such as continuous training and deployment. The required level of automation in the software development lifecycle will be based on the guidelines defined by MLOps