Aprovisionamiento de recursos para la ejecución de experimentos basados en Machine Learning a través de las guías MLops

Abstract

Within scientific research, beyond the specific knowledge required to conduct research, some knowledge is needed in other areas such as machine learning, statistics, programming, or databases, to name a few. On the other hand, manual machine learning workflows can present various technical issues due to the variety of tools and professionals involved in these processes, especially when considering the machine learning development lifecycle. This implies that during experimentation, there may be issues starting with the preparation of necessary resources and services to carry out such procedures. Due to this, areas such as MLOps seek to reduce technical debt in the development of machine learning-based applications by promoting process automation through what is known as MLOps pipelines. In this context, this paper proposes the creation of a software prototype that facilitates the provisioning of necessary tools for the creation, configuration, execution, and versioning of scientific experiments based on machine learning using MLOps techniques and methodologies.