Machine Learning Operations, conceitos e ferramentas

Abstract

The increasing adoption of Machine Learning (ML) in various fields has driven the need for efficient solutions to manage the lifecycle of models in production. Machine Learning Operations (MLOps) emerges as a response to this demand, combining software development practices (DevOps) with the specific characteristics of ML. MLOps aims to automate, integrate, and monitor the stages of model development, deployment, and maintenance, from data collection and preparation to continuous monitoring in production. This approach seeks to optimize model performance, ensure the reproducibility of experiments, and facilitate the scalability of ML solutions, allowing organizations to obtain maximum value from their investments in artificial intelligence. This study focuses on investigating the MLOps ecosystem, seeking to understand its fundamental concepts and analyze the main tools used in the construction and management of ML pipelines in production. The research, of a descriptive nature, combines a literature review with a comparative analysis of tools, addressing the origin and evolution of MLOps, its essential principles, and the challenges of its implementation. The study investigates the application of MLOps in sectors such as Industry 4.0, healthcare, and smart cities, and details the concept of MLOps pipelines, from data ingestion to continuous model monitoring, highlighting the importance of CI/CD practices. The research analyzes the main tools available on the market, describing their functionalities, architectures, use cases, advantages, and disadvantages, with the objective of providing a comprehensive view of the MLOps ecosystem and an overview of the available tools.