Webpage of Sebastien Loustau
I am Sebastien Loustau, researcher in the laboratory LMAP at UPPA and also involved in the I-Site E2S-UPPA, initiative d'excellence supported by UPPA, INRIA, INRAE and CNRS (see a french marketing video here).
I am also the founder of the non-profit organization IAPau and the founder of the AI startup LumenAI.
Prior to it, you might have met me in Marseille (Institut de Marseille, previously LATP) where I defended my phd in 2008 under the supervision of the late professor Laurent Cavalier, or in Centrale Marseille or Aix-Marseille 1, where I teached Probability, Statistics and Machine Learning, or more recently in LAREMA, where I defended my Habilitation thesis in 2014 about Online Learning and Inverse Statistical Learning.
My research interests are focused on online learning, mathematical statistics, information theory and more generally mathematical statistics for machine learning. I am interested by techniques that process data on the fly, that have mathematical motivations and theoretical guaranteed, such as excess risk or regret bounds. More recently, I am interested in the applications of these activities to deep learning techniques and environmental challenges.
Why deep learning ? Because many companies, researchers and engineers, have popularized this family of algorithms 10 years ago thanks to really good performances in computer vision and natural langage processing.
Why environmental challenges ? Because nowadays, there is a scientific evidence about the fact that the unprecedented current warming trend is extremely likely to be the result of human activity since the mid-20th century.
Feel free to send me any emails to discuss my work at sebastien[dot]loustau[at]univ-pau.fr.
☝♡ If you want to join the GreenAI UPPA team, contact me we have open positions. If you are at the second step of the 2021 Hiring Process, please, download this file to make your consumption report.
See me on various videos on the web to illustrate my daily work and scientific interests:
List of recent talks
- Bayesian Deep Active Learning - Theory and applications to Computer Vision and NLP, Intern seminar, Pau, march 2023, -slides here-
- Agriculture et sols vivants - Enjeu climatique, sociaux et environnementaux, Club Jardin du Lycée Louis Barthou, Pau, march 2023,-slides here-
- Comment entreprendre dans un contexte décroissant ? Macro et micro-économie de la décroissance, Degrowth Hacking meetup de Pau, december 2022, -slides here-
- Comment alléger l'empreinte environnementale des algorithmes de Deep Learning - Théorie et pratique, Summer School Green IT, invitation by Adel Noureddine, september 2022,
- Vers un apprentissage profondemment plus écologique, Journée MASS 2022, Rouen, invitation by Mohamed Hebiri, august 2022,
- Voyage sous surveillance vers une IA bienveillante, Sommet Benvivo, la bienveillance dans tous ses états, Table ronde avec Stanislas Mahieux, XXII, au Palais Beaumont, Pau, june 2022,
- Deep Learning theory for power-efficient algorithms, Team ApproxBayes of Riken Institute, invitation by Pierre Alquier, november 2021,
- Comment intégrer des contraintes environnementales dans l'apprentissage profond ?, PFIA'21, track Nouvelle-Aquitaine, invitation Nicolas Roussel, july 2021 program here,
- IA et réchauffement climatique, Conférence de vulgarisation au Lycée Louis Barthou, mai 2021 vidéo ici et un article ici.
- Power efficient Deep Learning, AI4Climate workshop, invitation Julien Brajard , oct 2020 download slides here,
Current main position
2020-, Researcher at LMAP of the UPPA, chairman of GreenAI UPPA
- Team presentation - Our research encompasses diverse projects and collaborations around the mathematical foundations of power-efficient deep/machine learning algorithms, and the applications of AI to build a more sustainable world. A website about the projects and the current team is available here.
Curent and past projects and activities
- 2023-, founder of Degrowth Hacking meetup, a dynamic group dealing with degrowth theory and regenerative agriculture,
- 2021-, founder of la-derniere-bibliotheque.org , a database of knowledge about the environmental emergency,
- 2021-, founder of the crazy PhiloML project, that mixes Philosophy and NLP,
- 2019-, Vice-President of the amazing basketball association ALC Basket,
- 2018-2022, founder and President of the non-profit organization IAPau,
- 2015-2020, founder and CEO of the french startup LumenAI,
- 2016-2020, founder and organizer of the Pau Machine Learning meetup,
- 2009-2015, assistant professor at Université d'Angers, researcher at LAREMA, UMR-CNRS 6093.
List of selected publications
- Chee, A. and Loustau, S. - Learning with BOT - Bregman and Optimal Transport divergences, 2021 HAL repository,
- Chee, A. and Loustau, S. - Sparsity regret bounds for XNOR-nets++, 2021 HAL repository,
- Li, L., Guedj, B. and Loustau, S. - A quasi-Bayesian perspective to online clustering, Electron. J. Statist., 12(2): 3071–3113. 2018,
- Darmaillac, Y., and Loustau, S. - MCMC Louvain for Online Community Detection, 2017 https://arxiv.org/abs/1612.01489
,
- Chichignoud, M. and Loustau, S. - Bandwidth selection in kernel empirical risk minimization via the gradient, Ann. Statist., 43(4): 1617-1646. 2015,
- Loustau, S. and Marteau, C. - Minimax fast rates for discriminant analysis with errors in variables, Bernoulli, 21(1): 176-208. 2015,
- Chichignoud, M. and Loustau, S. - Adaptive noisy clustering, IEEE Transactions on Information Theory, 60 (11), 7279-7292. 2014,
- Loustau, S. - Inverse statistical learning, Electronic Journal of Stats, 7: 2065-2097. 2013,
- Loustau, S. - Penalized empirical risk minimization over Besov spaces, Electronic Journal of Stats, 3: 824-850. 2009,
- Loustau, S. - Aggregation of SVM classifiers using Sobolev Spaces, Journal of Machine Learning Research, 9: 1559-1982, 2008.
My phd is also available here and my habilitation thesis here.