Since 2015 I've been a researcher in the
TAO/TAU team (machine learning and optimisation), at INRIA Saclay.
Bio: I defended my PhD thesis in computer vision within the Odyssee Team in December 2006;
my advisors were Olivier Faugeras and Renaud Keriven. Then I spent one year as a post-doc at the Max Planck Institute for Biological Cybernetics in Bernhard Schölkopf's team (statistical learning). In 2008 I joined the Pulsar/St☆rs team (video understanding), at INRIA Sophia-Antipolis, and in 2015 the TAO/TAU team (machine learning and optimisation), at INRIA Saclay. Administratively, I'm also the head of the Data Science department (of the LISN lab, Paris-Saclay University), since 2021.
Designing lighter architectures for generative models based on diffusion flows (same application as previous topic, but focusing on the generative architecture in general)
Links between explicability and formal proofs of neural networks (extraction of key concepts)
and lots of other very interesting topics, on demand
other offers from the TAU team (deep learning for population genetics, AutoML, causality, etc.), in particular:
architecture self-adaptation (with a functional analysis viewpoint),
guaranties (in some cases),
learning PDEs and numerical simulations,
generating distributions with right statistical properties
Main applications:
satellite imagery,
peoples' genetics,
protein conformations,
dynamical systems / fluid mechanics...
and also: weather forecast, brain imagery, etc.
Long term: Statistical learning, "artificial intelligence"
Unify statistical learning and "logical" models (e.g.: learn grammar, detect an algebric group operator)
How to learn possibly everything, without imposing any predefined structure ?
Past: Shapes, Vision, Optimization
Vision
Image / video segmentation
Shapes (object boundaries)
Structure of the space of shapes
Metrics (distance between shapes, cost of deformations), shape matching (point-to-point)
Learning functions from/to the space of shapes
Optimization
Graph cuts
Exploration strategies
Post-doc
Learning how to predict an image given another one with different modality
Automatic image colorization (color from greyscale texture)
Prediction CT images (computed tomography) from MR scans (magnetic resonnance)
Tools used: kernel methods, graph-cuts
Thesis
Aim: shape priors for image segmentation
Method: mean and statistics of images, of curves (and more generally surfaces)
Tools: non-rigid registration, shape gradient (choice of an inner product to perform gradient descents)
Energies, metrics considered: Hausdorff distance, geodesics, local cross-correlation.
Publications: (being updated... see Google Scholar for complete list) (until 2020, first author is underlined when not me) (I also have a special webpage dedicated to present my publications, with images and videos)
Chapter Kernel methods in medical imaging, with Matthias Hofmann and Bernhard Schölkopf, chapter of the book Biomedical Image Analysis: Methodologies and Applications, N. Paragios, J. Duncan & N. Ayache Editors, Springer, 2008. [bibtex]
Chapter Approximations of shape metrics and application to shape
warping and empirical shape statistics, with Olivier Faugeras, Renaud
Keriven and Pierre Maurel, chapter of the book Statistics and
Analysis of Shapes, H. Krim & A. Yezzi Editors, Birkaüser 2006. [bibtex]
An Implicit GNN Solver for Poisson-like Problems, by Matthieu Nastorg, Michele Alessandro Bucci, Thibault Faney, Jean-Marc Gratien, Guillaume Charpiat and Marc Schoenauer, dans Computers and Mathematics with Applications, CMA 2024. [bibtex]
Generalized Gradients: Priors on
Minimization Flows, with Pierre Maurel, Jean-Philippe Pons, Renaud
Keriven and Olivier Faugeras, in the International Journal of Computer Vision, Volume 73, Number 3, July 2007, IJCV 2007. [bibtex]
Variational, geometric, and statistical methods for modeling brain anatomy
and function, with Olivier Faugeras, Geoffray Adde,
Christophe Chefd'Hotel, Maureen Clerc, Thomas Deneux, Rachid Deriche,
Gerardo Hermosillo, Renaud Keriven, Pierre Kornprobst, Jan Kybic,
Christophe Lenglet, Lucero Lopez-Perez, Théo Papadopoulo, Jean-Philippe
Pons, Florent Ségonne, Bertrand Thirion, David Tschumperlé, Thierry
Viéville and Nicolas Wotawa, in the journal Neuroimage,
23S1:S46-S55, 2004. Note: Special issue: Mathematics in Brain Imaging -
Edited by P.M. Thompson, M.I. Miller, T. Ratnanather, R.A. Poldrack and T.E. Nichols.
[bibtex]
Pre-prints (submitted to journals/conferences or about to be):
Growth strategies for arbitrary DAG neural architectures, by Stella Douka, Manon Verbockhaven, Théo Rudkiewicz, Stéphane Rivaud, François P. Landes, Sylvain Chevallier and Guillaume Charpiat, European Symposium on Artificial Neural NetworksESANN 2025. [bibtex]
Multi-Level GNN Preconditioner for Solving Large Scale Problems, by Matthieu Nastorg, Jean-Marc Gratien, Thibault Faney, Michele Alessandro Bucci, Guillaume Charpiat and Marc Schoenauer, IEEE International Parallel and Distributed Processing Symposium WorkshopsIPDPSW 2024. [bibtex]
DS-GPS: A Deep Statistical Graph Poisson Solver, by Matthieu Nastorg, Marc Schoenauer, Guillaume Charpiat, Thibault Faney, Jean-Marc Gratien and Michele Alessandro Bucci, Workshop Machine Learning and the Physical Sciences, at NeurIPS 2022. [bibtex]
Shape Metrics, Warping and Statistics, with Olivier Faugeras
and Renaud Keriven, Proceedings of the International
Conference on Image Processing, ICIP 2003. IEEE Signal Processing Society. [bibtex]
Other International Conferences:
Distance-Based Shape
Statistics, with Pierre Maurel, Renaud Keriven and Olivier
Faugeras, IEEE International Conference on Acoustics, Speech, and
Signal Processing, Special Session: Statistical
Inferences on Nonlinear Manifolds with Applications in Signal and Image
Processing (This article mainly summarizes some previous articles but also
briefly introduces the graph Laplacian applied to shapes). ICASSP 2006. [bibtex]
Invited Poster
Poster at the Designing Tomorrow's
Category-Level 3D Object Recognition Systems: An International Workshop (september 2003)
Poster at the Designing Tomorrow's
Category-Level 3D Object Recognition Systems: An International Workshop (september 2003)
Animations:
Miscellaneous animations (image segmentation with/without shape prior;
evolution obtained by gradient descent with the standard inner product vs
one which favors rigid transformations; first modes of deformations of a
database of faces) presented in my PhD defense.
Faces: characteristical modes associated to a face
database (intensity variations and warping at the same time)
Antoine Szatkownik (designing neural networks tailored to population genetics) [2022-], co-supervised with Cyril Furtlehner, Flora Jay and Burak Yelmen
Francesco Pezzicoli (Graph Neural Networks to tackle amorphous materials: glass transition) [2022-], co-supervised with François Landes
Thibault Monsel (learning dynamical systems such as delay differential equations) [2022-], co-supervised with Lionel Mathelin and Onofrio Semeraro
Matthieu Nastorg (machine learning enhanced resolution of Navier Stokes equations on general unstructured grids) [2021-], co-supervised with Alessandro Bucci, Thibault Faney, Jean-Marc Gratien and Marc Schoenauer
Loris Felardos (neural networks for molecular dynamics simulation), co-supervised with Jérôme Hénin and Bruno Raffin [2018-2022]
Julien Girard (formal proof of neural networks), co-supervised with Zakaria Chihani and Marc Schoenauer [2018-2021]
Nicolas Girard (satellite image vectorization using neural networks), co-supervised with Yuliya Tarabalka and Pierre Alliez [2017-2020]
Théophile Sanchez (flexibility of neural networks with application to peoples' genetics), co-supervised with Flora Jay and Marc Schoenauer [2017-2022]
Pierre Wolinski (learning the structure of neural networks), co-supervised with Yann Ollivier [2016-2020]
Emmanuel Maggiori (segmentation of satellite images with shape prior), co-supervised with Yuliya Tarabalka and Pierre Alliez [2015-2017]
Ratnesh Kumar (fiber-based segmentation of videos for activity recognition), co-supervised with Monique Thonnat [2011-2014]
Master 2 students
María Belén Guaranda Cabezas (hierarchical deep learning generative architectures for turbulent systems), co-supervised with Sergio Chibbaro and Lionel Mathelin [2023]
Louis Dumont (capturing molecular conformations with graph neural networks), co-supervised with Loris Felardos and Jérôme Hénin [2020]
Pierre Jobic (permutation-invariant deep learning for peoples' genetics), co-supervised with Théophile Sanchez and Flora Jay [2020]
Andrew Khalel (pan-sharpening using neural networks (fusion of images of different resolutions and modalities)), co-supervised with Yuliya Tarabalka [2018]
Mo Yang (weather forecast: prediction of the trajectory of storms), co-supervised with Claire Monteleoni and Sophie Giffard-Roisin [2018]
Hugo Richard (video generation and analysis, with neural networks, with application to brain imaging), co-supervised with Bertand Thirion [2017-2018]
Armand Zampieri (registration of satellite images with the cadastre), co-supervised with Yuliya Tarabalka [2017]
Théophile Sanchez (peoples' genetics), co-supervised with Flora Jay [2017]
Priyanka Mandikal (registration of 3D medical images), collaboration with Therapixel [2017]
Emmanuel Maggiori (shape features in partition trees of images), co-supervised with Yuliya Tarabalka [2014]
Kandan Ramakrishnan (tracking dust particles in a fusion reactor), co-supervised with Vincent Martin [2011]
Ezequiel Cura (strategies for automatic model construction) [2010]
Anja Schnaars (texture-based segmentation) [2010]
Master 1 students, L3 students or equivalent
Martin Toth (explanation of the decision taken by a neural network), collaboration with Hossein Khonsari [2017-2019]
Louis Bethune (tracking paramecia with a motorized microscope using reinforcement learning), collaboration with Romain Brette [2017]
Raphaël Guegan (crowd dynamics estimation with neural networks), co-supervised with Emanuel Aldea [2017]
Etienne Desbois (skin disease classification), collaboration with Hossein Khonsari [2016]
Sorana Capalnean (classification of gestures obtained by a depth camera) [2012]
Bertrand Simon (dynamics of an articulated movement and gesture recognition) [2012]
Engineers
Raphaël Jaiswal (driving scenario classification), collaboration with Renault [2017-2018]
Etienne Brame (multi-class classification of a big database of images), collaboration with Armadillo within the Adamme project [2017-2018]
/
en français ?)
