Mathilde Mougeot
Mathilde Mougeot is an affiliate professor in mathematics at ENS Paris-Saclay and a university professor in data science at ENSIIE (École Nationale Supérieure d'Informatique pour l'Industrie et l'Entreprise).
She has held the "Industrial Data Analytics & Machine Learning - IdaML" chair (Atos/Bertin Technologies/CEA/ENS Paris-Saclay/ENSIIE) since 2018. She teaches statistics and machine learning at master's level, supervises PhD students on machine learning issues and is interested in their applications in industry.
Mathilde Mougeot is also in charge of promoting research at the Insmi/CNRS (Institut national des sciences mathématiques et de leurs interactions), raising awareness and supporting mathematical researchers in projects to promote their work.
Biography
- Since 2016, she has been an affiliated professor at ENS Paris-Saclay, holder of the "Industrial Data Analytics & Machine Learning" (IdAML) chair, professor of data science at ENSIIE and project manager for the development of the Insmi/CNRS.
- 2015: Habilitation to direct research "Contribution to statistics and Data Science for industrial applications: from neural networks to spare linear models".
- 2005-2009: Senior lecturer at Université Paris La Défense and Université Paris Diderot
- 1999-2005: statistical expert and consultant for Miriad Technologies
- 1992-1999: Senior lecturer at Université Paris La Défense
- 1992: doctorate in mathematical science "Connectionist methods for image compression and self-organisation of the mammalian visual system", supervised by Robert Azencott
- 1988-1991: Thomson-Csf engineer
- 1987: graduated from AgroParisTech
Research areas
Mathilde Mougeot's area of research is in the field of machine learning for the development of artificial intelligence.
It revolves around the development of predictive models for monitoring, diagnosis and decision-making purposes, in a variety of application contexts.
In particular, she is working on model aggregation issues: combining different predictive models in a context where data and appropriate models change rapidly.
She is also working on transfer learning methods, to capitalise on a high-performance model already calibrated in a source domain and transfer the knowledge to a new target domain. "When data is scarce, artificial intelligence cannot properly calibrate models. Transfer learning solves this problem.
This research in applied mathematics is a response to major industrial challenges.
In 1999, Mathilde Mougeot took a turning point in her career and obtained a leave of absence to help, with her thesis supervisor Robert Azencott, to set up the company Miriad Technologies, which develops decision support algorithms for the industrial sector. "This was a very unusual career move at the time, because bridges between the worlds of business and academia were not yet part of university culture. Not to mention the fact that I was taking a certain amount of risk in terms of managing my professional career.
The researcher nevertheless ventured down this path for six years, during which time she was responsible for carrying out feasibility studies for a number of industries. From chemicals to aerospace, via pharmaceuticals, she carried out numerous proofs of concept to develop software using statistical learning methods. For example, she designed an automatic detection model for over-consumption of compressors for Air Liquide, which has been deployed in one of its plants in Texas. Mathilde Mougeot adds: "During this start-up experience, I also took on roles for which I had no experience. I had to carry out tasks related to sales, pre-sales and marketing. This multifunctionality, added to the variety of scientific projects, gave me a good dose of adrenalin!
Une nouvelle manière d’enseigner et de mener des travaux
La recherche est aussi une source d’exaltation pour Mathilde Mougeot, qui retrouve son poste à l’université de Paris Nanterre en 2006. Elle aborde ses travaux de recherche en statistique et méthode d’apprentissage avec un nouveau regard, fortement motivé par des objectifs d’applications concrètes. "J’ai réalisé qu’en axant mes recherches en fonction d’usages réels, je me les appropriais beaucoup plus, car je m’inspirais de problématiques de terrain. Ce qui a grandement changé ma manière d’aborder mon métier."
Cette double compétence académique et industrielle, et cette nouvelle perspective lui valent de nombreuses sollicitations. Notamment pour la mise en place du master PRO ISIFAR au sein de l’Université Paris-Nanterre, d’un cours de data mining à CentraleSupélec et d’un cours d’outils statistiques à destination des étudiants littéraires de l’ENS de Paris. "Mon mode d’enseignement intègre désormais systématiquement des perspectives d’applications réelles que j’adapte à mon public et aux concepts que j’expose."
En 2009, elle poursuit ses recherches académiques au Laboratoire de probabilités, statistique et modélisation de l’Université Paris Diderot, où elle enseigne et obtient son habilitation à diriger des recherches en 2015.
En 2017, la chercheuse obtient un poste de professeure en science des données à l'ENSIIE, puis en 2018 de professeure affiliée à l'ENS Paris-Saclay où elle devient titulaire de la chaire "Industrial Data Analytics & Machine Learning" (IdAML) du centre Borelli (Université Paris-Saclay, CNRS, ENS Paris-Saclay, Université de Paris, SSA).
Dans ce cadre, elle poursuit ses travaux en apprentissage par transfert et travaille notamment en partenariat avec la multinationale Tarkett, spécialisée dans les revêtements de sols et surfaces de sports.
Elle supervise, en collaboration avec Nicolas Vayatis, directeur du Centre Borelli, des modèles de détection automatique de chute de personnes âgées, grâce à des données collectées sur des expériences faites avec des personnes jeunes. "Recueillir des données de chutes pour des personnes vulnérables est très difficile et très long. Nous utilisons donc des connaissances sources issues de données relatives aux chutes de personnes bien portantes et les transférons à un nouveau domaine."
En collaboration avec le CEA et l’entreprise Michelin, la chercheuse commence aujourd’hui à travailler sur des modèles hybrides, c’est-à-dire des modèles d’apprentissage qui intègrent également des connaissances modélisées issues de la science physique (mécanique des fluides et des solides, ou phénomènes météorologiques).
Bridging the gap between academia and industry: a vocation
From 2016 to 2019, Mathilde Mougeot was in charge of promoting the CNRS's Institut national des sciences mathématiques et de leurs interactions (INSMI). During this time, she was involved in setting up the first joint mathematics research structure as part of the "LabCom" programme initiated by the French National Research Agency (ANR): the LabCom I3M, which brings together the CNRS, the University and University Hospital of Poitiers and Siemens.
Since 2019, the researcher has been deputy director of the Fondation Mathématiques Jacques Hadamard at the Université Paris-Saclay and deputy director of the university's Graduate School of Mathematics. There, she leads and builds relationships to encourage exchanges between students and the business world. "After an atypical career during which I moved away from the University, only to return to it again, I'm now working to bring these two worlds together. Bringing these two worlds together and promoting all their benefits is exactly in my DNA", concludes the researcher.