Context
The selected candidate will be based in Lille in the MAGNET team.
The main objective of the team is to develop ethically acceptable machine learning algorithms focusing on fairness, privacy, and decentralized learning and to empower end users of artificial intelligence. The candidate will be under the supervision of Michaël Perrot.
Assignment
Machine learning is nowadays used in various applications, such as medical diagnosis and speech recognition. Its success stems from the performance of learned models, sometimes reaching human-level capabilities. However, deploying these models on a larger scale requires more than just accuracy, and it is imperative to consider fairness and privacy when human lives are affected. For instance, a model used for medical diagnosis or an automated hiring process should not be biased against subgroups of the population. Similarly, released models should not leak information about the data that was used to train them. Several approaches have been proposed in the literature to enforce fairness and privacy (either individually or jointly) in machine learning models.
The goal of this engineer position will be to make fairness and privacy preserving approaches developed in the team more accessible to the machine learning community. On the one hand, the FairGrad library was developed a few years ago as an easy to use approach to enforce fairness. A first objective will thus be to expand it, benchmark it, and integrate it to existing fairness libraries (such as AIF360 and Fairlearn) as well as general purpose machine learning libraries (such as scikit-learn). On the other hand, recent work in the team showed that it is possible to learn from locally private data with a controlled bias [1]. A second goal will thus be to propose an efficient implementation of the proposed approach and to benchmark it on a variety of tasks. Finally, a third objective will be to integrate both methods with each other to propose a well founded pipeline to learn privacy preserving and fair machine learning models.
[1] Learning with Locally Private Examples by Inverse Weierstrass Private Stochastic Gradient Descent , Jean Dufraiche, Paul Mangold, Michael Perrot, Marc Tommasi, 2026.
Main activities
Review and follow the existing libraries on Fair and Privacy preserving Machine Learning. Maintain, benchmark and expand the scope of the FairGrad library. Implement, maintain, and benchmark a machine learning pipeline for machine learning from locally private data.
Skills
- A good command of English
- A good command of Python
- A strong background in implementation and experimentation
- A strong background in mathematics
- A good knowledge of machine learning, statistics and algorithms
- Some knowledge on fairness and/or privacy would be a plus
#J-18808-Ljbffr
Engineer Position on Fairness and Privacy in Trustworthy Machine Learning (F/M) employeur : Inria Villeneuve d'Ascq
En tant qu'employeur, notre entreprise se distingue par son engagement envers l'innovation et la recherche de pointe dans le domaine de l'optimisation combinatoire. Nous offrons un environnement de travail collaboratif où les chercheurs peuvent développer leurs compétences tout en contribuant à des projets significatifs sur des plateformes de calcul haute performance. De plus, notre culture d'entreprise valorise la croissance personnelle et professionnelle, avec des opportunités de publication et de collaboration internationale.
Coordonnées :
Équipe de recrutement de Inria Villeneuve d'Ascq
