Initial Insights on MLOps: Perception and Adoption by Practitioners

TL;DR

The paper investigates how practitioners perceive and adopt MLOps guidelines through a qualitative study of three experts from diverse organizational contexts. It finds heterogeneous adoption, with some practitioners using MLOps implicitly, while others rely on offline processes and face high tool costs and a steep learning curve. The results suggest Generative AI and Explainable AI hold promise to enhance coding support, monitoring, and explainability, potentially accelerating reliable ML deployments. Overall, the work provides empirical insights to improve MLOps guidelines, tooling, and organizational training for scalable and ethical AI applications.

Abstract

The accelerated adoption of AI-based software demands precise development guidelines to guarantee reliability, scalability, and ethical compliance. MLOps (Machine Learning and Operations) guidelines have emerged as the principal reference in this field, paving the way for the development of high-level automated tools and applications. Despite the introduction of MLOps guidelines, there is still a degree of skepticism surrounding their implementation, with a gradual adoption rate across many companies. In certain instances, a lack of awareness about MLOps has resulted in organizations adopting similar approaches unintentionally, frequently without a comprehensive understanding of the associated best practices and principles. The objective of this study is to gain insight into the actual adoption of MLOps (or comparable) guidelines in different business contexts. To this end, we surveyed practitioners representing a range of business environments to understand how MLOps is adopted and perceived in their companies. The results of this survey also shed light on other pertinent aspects related to the advantages and challenges of these guidelines, the learning curve associated with them, and the future trends that can be derived from this information. This study aims to provide deeper insight into MLOps and its impact on the next phase of innovation in machine learning. By doing so, we aim to lay the foundation for more efficient, reliable, and creative AI applications in the future.