Leader: Marc Tommasi



Magnet is a joint research team between INRIA and CRIStAL.

A primary objective of Magnet is in making artificial intelligence more acceptable to society by solving some ethical issues of Machine Learning (ML) and on empowering end users of artificial intelligence. From a scientific perspective we focus on privacy, fairness, (data) sobriety. Our approaches are typically based on the common theme of leveraging the relationships between data and between learning objectives. We study graph-based machine learning methods which are the common foundations of the research group and we rely on methods coming from statistical and computational learning theory, graph theory, representation learning, (distributed) optimization and statistics.

We are mainly interested in provable properties for machine learning algorithms but we also consider more empirical work. Our application domains cover health, mobility, social sciences, voice technologies.

Research themes

Our research is organized along three main axes:

- Mining and learning in graphs: we study the trade-off between predictive accuracy, computational complexity and verification of ethical properties for graph-based learning algorithms.

- Machine Learning for Natural Language Processing: we study how to enhance NLP methods with graph-based learning algorithms and how to develop ethical learning algorithms in NLP. The objective is to go beyond vectorial classification to solve task like coreference resolution and entity linking, temporal structure prediction, and discourse parsing.

- Decentralized Machine Learning: we address the problem of learning in a private, fair and energy efficient way when users and data are naturally distributed in a network. From an algorithmic perspective, we study federated learning and fully decentralized learning and optimization. We also consider research on a global and holistic level, in complex pipelines that involve learning.



Antoine Barczewski

Apprentissage automatique décentralisé respectueux de la vie privée

Moitree Basu

Integrated privacy-preserving AI

Ioan-tudor Cebere

Privacy-Preserving Machine Learning

Edwige Cyffers

Apprentissage automatique décentralisé respectueux de la vie privée

Marc Damie

Secure protocols for a verifiable decentralized machine learning

Aleksei Korneev

Technologies fiables et multi-sites renforçant le respect de la vie privée

Dinh-viet-toan Le

Approches de traitement automatique du langage naturel dans le domaine musical : adaptabilité, performance et limites

Bastien Lietard

Modèles Computationnels pour le Changement Sémantique Lexical

Gaurav Maheshwari

Apprendre des représentations fiables pour le traitement du langage naturel

Paul Mangold

Optimisation décentralisée et respectueuse de la confidentialité des données pour l'apprentissage statistique

Les autres équipes du groupe thématique ' DatInG '