The SigMA team’s skills are based on solid foundations in signal processing, statistics and machine learning. SigMA currently has 15 permanent staff (including 1 emeritus, 4 CNRS researchers, 3 in section 7 and 1 in section 17).
SigMA designs and studies new methods and algorithms for extracting useful information from physical measurements. These include natural images, astronomical observations and polarized light. Statistical approaches are at the heart of SigMA’s expertise, with a particular focus on Bayesian inference methods, and in particular Monte-Carlo sampling methods. The scientific objectives are, for example, to obtain error guarantees on predictions to support decision making processes, or to take into account data heterogeneity when modeling.
Statistical signal processing and modeling is a common denominator of SigMA. Among the team’s strengths are Markov chain sampling methods (MCMC) for inverse problems, stochastic point processes (particularly determinantal, DPP) and multimedia information security. This work involves statistical learning, both to feed Bayesian models for inverse problems and to optimize the performance of information protection methods.
The most common applications of our methodological activities range from astronomy to the optimization of catalytic chemical reactions, from polarimetric imaging to multimedia information security. Cosmology occupies a special place with the presence of an interdisciplinary CNRS researcher (INSU Section 17) at SigMA.
In terms of research training, the SIGMA team has strong links with the joint data science master’s program at Centrale Lille Institut, IMT Nord-Europe and the University of Lille, as well as with Centrale Lille’s data science and artificial intelligence program.
Illustrations of our research: * Bayesian inference and Monte-Carlo methods for inverse problems in radio astronomy, * applications to object tracking in video sequences and cosmology, * sampling of deterministic point processes, links with quantum physics, * information security: extraction and detection of weak signals for forensic analysis and steganalysis, protection by watermarking or selective encryption, * signal processing using tensorial approaches, * polarimetric imaging: detection of artifacts in materials.
Rémi Bardenet
Deep domain generalization for operational steganalysis
Algorithmes asynchrones de MCMC pour de l'inférence Bayésienne rapide
Apprentissage automatique de graphes volumineux basé sur des représentations en réseaux de tenseurs
Intelligence artificielle pour la conception de catalyseurs à haut débit
Stéganalyse et maîtrise des taux d'erreur
Développement de méthodologie de détection fiable pour la stéganalyse en exploitant l'apprentissage actif
Méthodes de Monte-Carlo avec processus de Coulomb.
Echantillonnage des sous-espaces à l’aide des processus ponctuels déterminantaux. 03/11/2020
Real-Time Sensor Management Strategties for Multi-Object Tracking 19/12/2018
Adaptive Bayesian inference for source reconstruction in atmospheric dispersion 21/11/2016
Segmentation d'images mammographiques pour l'aide au diagnostique 05/10/2016
Detection de personnes pour des systèmes de videosurveillance multi-caméra intelligents 28/09/2015
Modélisation probabilité d’imprimés à l’échelle micrométrique 18/05/2015