The state of MLOps

This paper describes the current state of DevOps for Machine Learning (ML). We analyze the relationship between software engineering and data science in the ML development life cycle which has brought the necessity for specialized DevOps for ML. We review some of the recently emerged pipeline-like architectures and the most common components. Furthermore, we examine the most common toolchains used in DevOps and how they correlate with the tools used in MLOps. It is important to note that MLOps community has developed a vast landscape of tools to support these processes. We show many examples of how the tools tackle (or not) problems for the ML domain. We present some proposed possible solution that we found with the adoption of DevOps in the ML world. We conclude that to ensure the correct development and deployment of the ever-growing quantity of models used in the real world it is necessarily to use...

• Autores: Varón Maya, Andrés Felipe
• Tipo de recurso: Trabajo de grado de pregrado
• Fecha de publicación: 2021
• Institución: Universidad de los Andes
• Repositorio: Séneca: repositorio Uniandes
• Idioma: eng
• OAI Identifier: oai:repositorio.uniandes.edu.co:1992/51495
• Acceso en línea: http://hdl.handle.net/1992/51495
• Palabra clave: Aprendizaje automático (Inteligencia artificial); Ingeniería de software; Ciencia de datos; Ingeniería
• Rights: openAccess
• License: http://creativecommons.org/licenses/by-nc-nd/4.0/