Abstract
MLOps has emerged in the past decade, aiming to define and implement best practices for developing and deploying machine learning (ML) models. Researchers have since been exploring and applying MLOps to various use cases, thus contributing to a better understanding and definition of its requirements and practical implementation. This poster examines the challenges faced by MLOps research and investigates the advancements and challenges addressed by ML paradigms in general. By merging these two areas of work, our preliminary idea is to introduce Continuous Learning as a new pillar for MLOps. Indeed, Continuous Learning will enhance MLOps with the capability to learn from past choices, reduce the need for constant retraining of models, and mitigate issues related to both data and concept drifts.