Model Reports, a Supervision Tool for Machine Learning Engineers and Users

Abstract

This article investigates a methodology to design an automated supervision report, ensuring the suitability between the designers and the users of an algorithm. For this purpose, we built a super-vision tool, focused on error diagnosis. The argumentation of the article relies first on the exposition of the reasons to use model reports as a supervision artefact, with a prototype of implementation at an organization level, describing the necessary tooling to industrialize its production. Finally, we propose a method for supervising machine learning algorithms in a responsible and sustainable way, starting from the conception of the algorithm, along its development and dur-ing its operating phase.