Yohann DE CASTRO, Researcher in Statistics and Machine Learning

Latest news

[Dec. 2022] New publication Neural Networks beyond explainability: Selective inference for sequence motifs with A. Villié, P. Veber, and L. Jacob is now available on ArXiv and BioRxiv.
[Oct. 2022] After Quentin DUCHEMIN (now at SDSC, EPFL) in June, two brilliant students, whom I have the pleasure to co-advise, will defend their Ph.D. in December. Check out Guillaume DALLE (machine learning and combinatorial optimization) and Romain PETIT (Inverse problems and Calculus of variations)'s works.
[Sept. 2022] Multiple Testing and Variable Selection along Least Angle Regression's path, with J-M Azaïs, has been published at Information & Inference: A Journal of the IMA.
[August 2022] Random Geometric Graph: Some recent developments and perspectives, with Q. Duchemin, has been accepted for publication at the proceedings of the High Dimensional Probability IX Conference, Progress in Probability series.
[June 2022] I have been nominated Junior member at Institut Universitaire de France (IUF), starting October 2022.
[June 2022] Three rates of convergence or separation via U-statistics in a dependent framework, with Q. Duchemin and C. Lacour, has been published at Journal of Machine Learning Research.
[May 2022] Concentration inequality for U-statistics of order two for uniformly ergodic Markov chains, with Q. Duchemin and C. Lacour, has been accepted for publication at Bernoulli Journal.

Quick Overview

Activities

Responsable élu du site École Centrale Lyon de l'Institut Camille Jordan; À ce titre, membre statutaire du conseil de laboratoire de l'Institut Camille Jordan et membre statutaire du comité de direction de l'École Centrale Lyon.
Chercheur associé au CERMICS, École Ponts ParisTech;
Membre extérieur de l'équipe DANTE, INRIA Lyon;
Membre élu du conseil scientifique de l'École Centrale Lyon;
Membre nommé du conseil scientifique de l'Institut Camille Jordan;
Membre élu du conseil du Département Maths-Info de l'École Centrale Lyon;
Membre du bureau SciDoLySE: Science des Données à Lyon et Saint-Etienne.
Membre nommé du conseil des thèses de l'École Centrale Lyon;
Membre des ANR JCJC "CIPRESSI", "PIECES" et "GrandMa";

Teaching Duties

Course on Machine Learning (2A @ELC);
Course on Convexity and Parsimony (3A @ECL joint with Master 2 MeA (Lyon 1))

Research

My publications can be found on this list of publications (type and year)

I provide below of list of publications by subjects (Sparse models & Mathematical Statistics)

Sparse models: High-Dimension (L1); Measures (TV); Variational (TV gradient)

A new procedure for Selective Inference with the Generalized Linear Lasso, with Q. Duchemin, arXiv, 2022.
Towards off-the-grid algorithms for total variation regularized inverse problems, with R. Petit & V. Duval, Proceedings of the eighth international conference on Scale Space and Variational Methods in computer vision (SSVM), Pages 553--564, 2021; and published in Journal of Mathematical Imaging and Vision, 2022 (extended version). GitHub Repository
Dual optimal design and the Christoffel-Darboux polynomial with F. Gamboa & D. Henrion & J.-B. Lasserre, Optimization Letters, 2021, Vol. 15, 3-8.
SuperMix: Sparse Regularization for Mixtures, with S. Gadat & C. Marteau & C. Maugis-Rabusseau, Annals of Statistics, 2021, Vol. 49, No. 3, 1779-1809.
Multiple Testing and Variable Selection along Least Angle Regression's path, with J-M Azaïs, Information & Inference: A Journal of the IMA, 2022. GitHub Repository
A Short Introduction to “Moment-SoS Hierarchies”, Deuxième Congès National de la SMF, 2018.
On Representer Theorems and Convex Regularization, with C. Boyer & A. Chambolle & V. Duval & F. de Gournay & P. Weiss, SIAM Journal on Optimization, 29(2), 1260–1281. A related conference paper published at iTWIST'18 is Convex Regularization and Representer Theorems, with C. Boyer & A. Chambolle & V. Duval & F. de Gournay & P. Weiss.
Testing Gaussian Process with Applications to Super-Resolution, with J.-M. Azaïs & S. Mourareau, Applied and Computational Harmonic Analysis, Volume 48, Pages 445-481, 2020. GitHub Repository
Approximate Optimal Designs for Multivariate Polynomial Regression with F. Gamboa & D. Henrion & R. Hess & J.-B. Lasserre, Annals of Statistics, Volume 47, Issue 1, Pages 127–155, 2019. Matlab Code
Extended version of D-Optimal Design for Multivariate Polynomial Regression via the Christoffel function and Semidefinite Relaxations, with F. Gamboa & D. Henrion & R. Hess & J.-B. Lasserre, arXiv 1703.01777.
Sparse Recovery from Extreme Eigenvalues Deviation Inequalities, with S. Dallaporta, ESAIM P&S, Volume 23, Pages 430-463, 2019.
Adapting to Unknown Noise Level in Sparse Deconvolution, with C. Boyer & J. Salmon, Information & Inference: A Journal of the IMA, Volume 6, Issue 3, 1 September 2017, Pages 310–348.
Exact solutions to Super Resolution on semi-algebraic domains in higher dimensions, with F. Gamboa & D. Henrion & J.-B. Lasserre, IEEE Transactions on Information Theory, Volume 63, Issue 1, Jan. 2017, Pages 621-630. Matlab Code
Power of the Spacing test for Least-Angle Regression, with J.-M. Azaïs & S. Mourareau, Bernoulli Journal, Volume 24 n°1 (2018), Pages 465-492.
Randomized pick-freeze for sparse Sobol indices estimation in high dimension, with A. Janon, ESAIM P&S, Volume 19, 2015, Pages 725-745.
Non-uniform spline recovery from small degree polynomial approximation, with G. Mijoule, Journal of Mathematical Analysis and Applications, Volume 430, Issue 2, 15 October 2015, Pages 971–992.
A Rice method proof of the Null-Space Property over the Grassmannian, with J.-M. Azaïs & S. Mourareau, Annales de l’Institut Henri Poincaré, Probabilités et Statistiques, Volume 53, Number 4 (2017), Pages 1821-1838.
Spike Detection from Inaccurate Samplings, with J.-M. Azaïs & F. Gamboa, Applied and Computational Harmonic Analysis, Volume 38, Issue 2, March 2015, Pages 177–195.
Exact Reconstruction using Beurling Minimal Extrapolation, with F. Gamboa, Journal of Mathematical Analysis and Applications, Volume 395, Issue 1, Nov. 2012, Pages 336-354.
A remark on the lasso and the Dantzig selector, Statistics & Probability Letters, Volume 83, Issue 1, January 2013, Pages 304-314.
Optimal designs for Lasso and Dantzig selector using Expander Codes, IEEE Transactions on Information Theory, Volume 60, Issue 11, Nov. 2014, Pages 7293-7299.

Mathematical Statistics, Learning, and Applications

Neural Networks beyond explainability: Selective inference for sequence motifs, with A. Villié, P. Veber, and L. Jacob, arXiv:, submitted, 2023.
Minimax Estimation of Partially-Observed Vector AutoRegressions, with G. Dalle, arXiv:2106.09327, submitted, 2021.
Forecasting Nonnegative Time Series via Sliding Mask Method (SMM) and Latent Clustered Forecast (LCF), with L. Mencarelli, arXiv:2102.05314, 2021.
Random Geometric Graph: Some recent developments and perspectives, with Q. Duchemin, accepted for publication at the proceedings of the High Dimensional Probability IX Conference, Progress in Probability series.
Three rates of convergence or separation via U-statistics in a dependent framework, with Q. Duchemin and C. Lacour, Journal of Machine Learning Research, Volume 23, Issue 201, Pages 1-59, 2022.
Concentration inequality for U-statistics of order two for uniformly ergodic Markov chains, with Q. Duchemin and C. Lacour, Bernoulli Journal, 2022 (to appear).
Markov Random Geometric Graph (MRGG): A Growth Model for Temporal Dynamic Networks, with Q. Duchemin, Electronic Journal of Statistics, Volume 16, Pages 671-699, 2022.

GitHub Repository

Latent distance estimation for random geometric graphs, with E. Araya, NeurIPS 2019.
Adaptive Estimation of Nonparametric Geometric Graphs, with C. Lacour & T. M. Pham Ngoc, Mathematical Statistics and Learning, 2020 (to appear).
Nonnegative Matrix Factorization with Side Information for Time Series Recovery and Prediction, with J-M Azaïs & Y. Goude & G. Hébrail & J. Mei, IEEE Transactions on Knowledge and Data Engineering, Volume 31, Issue 3, March 1, Pages 493-506, 2019.
Nonnegative Matrix Factorization for Time Series Recovery From a Few Temporal Aggregates, with J-M Azaïs & Y. Goude & G. Hébrail & J. Mei, Proceedings of the 34th International Conference on Machine Learning (ICML 2017), PMLR 70:2382-2390, 2017.
Reconstructing undirected graphs from eigenspaces, with T. Espinasse & P. Rochet, Journal of Machine Learning Research, Volume 18, Issue 51, Pages 1−24, 2017.
Consistent estimation of the filtering and marginal smoothing distributions in nonparametric hidden Markov models, with É. Gassiat & S. Le Corff, IEEE Transactions on Information Theory, Volume 63, Issue 8, Aug. 2017, Pages 4758-4777.
Minimax adaptive estimation of non-parametric Hidden Markov Models, with É. Gassiat & C. Lacour, Journal of Machine Learning Research, Volume 17, Issue 111, Pages 1-43, 2016. Matlab Codes
Estimating the transition matrix of a Markov chain observed at random times, with F. Barsotti & T. Espinasse & P. Rochet, Statistics & Probability Letters, Volume 94, November 2014, Pages 98-105.
Quantitative Isoperimetric Inequalities on the Real Line, Annales Mathématiques Blaise Pascal, Volume 18 n°2 (2011), Pages 311-331.

Ph.D. Sutdents

Antoine SIMOES (CDSN ENS Lyon, Starting September 2022) with Joseph Salmon; High-Dimensional Statistics and MCP methods;
Martin GJORGJEVSKI (ANR GrandMa, Starting September 2022) with Nicolas Keriven and Simon Barthelmé; Signal on Graphs;
Ahmed EL RAFY (CIFRE, Starting September 2022) with Marc Jacob; ML for Acoustic and Fluids Mechanics modeling;
Guillaume DALLE (IPEF, 2019-2022) with Axel Parmentier; OR and ML; Now Post-Doc at EPFL;
Quentin DUCHEMIN (DIM Math Innov, 2019-2022) with Claire Lacour; Random Graphs and Post-Selection Inference; Now Post-Doc at SDSC (EPFL);
Romain PETIT (DIM Math Innov, 2019-2022) with Vincent Duval; Off-the-grid methods; Now Post-Doc at ML Genova Center (MaLGa);
Antoine VILLIÉ (LBBE, 2019-2022) with Laurent Jacob; Kernel Post-Selection Inference;
Ernesto ARAYA (Orsay grant, 2017-2020); Random Graphs; Now Post-Doc at INRIA Lille;
Jiali MEI (ÉDF CIFRE, 2014-2017) with Jean-Marc Azaïs, Yannig Goude and Georges Hébrail; "Time series recovery and prediction with regression-enhanced nonnegative matrix factorization applied to electricity consumption"; Prix Paul Caseau de l'académie des sciences, 2018; Now Senior Researcher at Shift Technology, a french unicorn.

Post-Docs

Nicolas Jouvin (2021), Sparse regularization on measures. Now Chargé de Recherche @INRAE;
Luca Mencarelli (2018-2020), Matrix factorization for Time Series;
Guillaume Mijoule (2018-2019), Semi-dicreste Optimal Transport. Now Data Scientist in Paris Greater Area.

Teaching

My office hours are on Monday 1pm-3pm room E6203.

Machine Learning (2nd year -- Master 1), more detail at machine-learning-centrale.github.io .
Convexity and Parsimony (3rd year -- Master 2) more detail at statistical-learning-centrale.github.io .

Vitae

2022 - 2027	Junior member of Institut universitaire de France
Since 2019	Full Professor at Institut Camille Jordan, École Centrale Lyon
2018-2019	Permanent Senior Researcher at CERMICS, École Ponts ParisTech
2017-2018	Invited Researcher at Inria Paris, MOKAPLAN Team
2012-2018	Maître de Conférences at Institut de Mathématique d'Orsay	H.D.R (Nov 2016)
2008-2012	Ph.D. at Institut de Mathématiques de Toulouse with J.-M. Azaïs and F. Barthe	Ph.D. Thesis (Dec 2011)
2004-2009	Student at École Normale Supérieure

Contact

yohann.de-castro@ec-lyon.fr
+33 (0)4 72 18 67 90
Institut Camille Jordan
École Centrale de Lyon
36 Avenue Guy de Collongue
69134 Écully, France