1 minute read

On August 2, 2024, the FATES-MLOps project was awarded €600,000 by the Agence Nationale de la Recherche (ANR) to work on Incorporating FATES Principles in Continuous Development of ML-Integrated Systems

The key issue of thje project is: “Can we systematically and progressively include justifications related to the FATES (Fairness, Accountability, Transparency, Ethics, and Security) building and operating of a software system integrating Machine Learning?”

Abstract

The MLOps movement adopts the DevOps objective of reducing the gaps between development and operations teams by integrating data scientist teams and Machine Learning (ML) models. In this project, we aim to apply and adapt good software engineering practices to strengthen both the overall quality of the ML model construction processes and the quality of the produced software systems, particularly in terms of extra-functional properties that will become crucial issues: Fairness, Accountability, Transparency, Ethics, and Security (FATES). In this project, we will focus on formalization, measurement, and management of these properties throughout the MLOps process. In order to evaluate models’ robustness in practical applications, we require more than traditional Key Performance Indicators (KPIs), such as precision and recall. Our project aims to study the FATES properties and, by refining proven software engineering concepts and tools, propose a systematic and tailored approach for considering those properties, particularly from the lens of ML Scientists or ML Engineers, throughout the lifecycle of the software developed following an MLOps approach.

Consortium

  • Institut de Recherche en Informatique de Toulouse (IRIT)
    • Jean-Michel Bruel
    • Olivier Teste
    • Marc Pantel
  • Laboratoire d’Informatique, Signaux et Systèmes de Sophia Antipolis (I3S)
    • Mireille Blay-Fornarino
    • Philippe Collet
  • Institut National de la Recherche en Sciences et Technologies du Numérique (Inria)
    • Fréderic Precioso
    • Michel Riveill
  • McMaster Centre for Software Certification (McSCert)
    • Sébastien Mosser
    • Richard Paige

Updated: