Master 2 — Sparsity and High Dimensions

Winter quarter (January–March 2027) · École Centrale Lyon · Monday 2:00–5:15pm

Audience: Master MeA, Master Maths avancées, Master GRAF, and third-year students of Centrale Lyon (MIR option).

Course Description

Sparsity is a ubiquitous concept in statistics, signal processing and machine learning. The course studies sparse recovery by convex relaxation. We start on a grid, with L1-minimisation, the null space property and dual certificates. We then move off the grid, to sparse measures and the Beurling LASSO (BLASSO). The focus is on estimation guarantees: exact recovery without noise, prediction and localisation bounds with noise, and stability of the support at low noise. The last lecture treats the generalised BLASSO with Fenchel-Young losses. Two examples run through the course: super-resolution from Fourier measurements, and the estimation of the means of a Gaussian mixture.

The 2027 lecture notes are available (version of 16 November 2026). The 2026 notes (first two lectures of the former syllabus) remain available.

Bibliography

  • [Foucart] A Mathematical Introduction to Compressive Sensing, Simon Foucart and Holger Rauhut, Applied and Numerical Harmonic Analysis (ANHA), Springer
  • [Blondel] The Elements of Differentiable Programming, Mathieu Blondel and Vincent Roulet.
  • [Wainwright] High-Dimensional Statistics: A Non-Asymptotic Viewpoint, Martin J. Wainwright, Cambridge University Press

Course Organizer

Projects

Guidelines: projects.pdf.
Please email Yohann your group names and 3 ranked reference papers by Monday, February 8, 2027.
Slides are due by email, as a PDF file, by Friday, March 19, 2027. Talks take place on Monday, March 22, 2027.

Suggested references:

  • [Super-Resolution] E. Candès and C. Fernandez-Granda, "Towards a Mathematical Theory of Super-Resolution". link
  • [Support Recovery] V. Duval and G. Peyré, "Exact Support Recovery for Sparse Spikes Deconvolution". link
  • [Off-the-Grid CS] C. Poon, N. Keriven, and G. Peyré, "The geometry of off-the-grid compressed sensing". link
  • [Representer] C. Boyer, A. Chambolle, Y. De Castro, V. Duval, F. De Gournay, and P. Weiss, "On Representer Theorems and Convex Regularization". link
  • [SR-Lasso] C. Poon and G. Peyré, "Super-Resolved Lasso". link
  • [GMM] R. Giard, Y. De Castro, and C. Marteau, "Gaussian Mixture Model with unknown diagonal covariances via continuous sparse regularization". link, notebook
  • [FW] Q. Denoyelle, V. Duval, G. Peyré, and E. Soubies, “The sliding Frank–Wolfe algorithm and its application to super-resolution microscopy,” Inverse problems, vol. 36, no. 1, p. 014001, Jan. 2020. link
  • [CPGD] L. Chizat, "Sparse Optimization on Measures with Over-parameterized Gradient Descent". link
  • [Mean-field] L. Chizat and F. Bach, "On the Global Convergence of Gradient Descent for Over-parameterized Models using Optimal Transport". link
  • [FY Losses] M. Blondel, A. F. T. Martins, and V. Niculae, "Learning with Fenchel-Young Losses". link
  • [Sparse iOT] F. Andrade, G. Peyré, and C. Poon, "Sparsistency for Inverse Optimal Transport", International Conference on Learning Representations (ICLR), 2024. link

Past exams

Clarifications (ECL only)

Il y a un seul cours, le cours « Parcimonie et Grande Dimension » (PGD), avec des publics et des notes différenciés. Nous proposons deux cours de niveau Master, « Optimisation et Représentation Parcimonieuse » (ORP/Master) et « Parcimonie et Grande Dimension » (PGD/Master), ainsi qu'un cours de 3e année Centrale, « Parcimonie et Grande Dimension » (PGD/Option). Les codes du portail de la scolarité sont indiqués dans le tableau ci-dessous. Le planning ci-dessous présente les cours de master. En cas de doute, les informations du portail officiel font foi.

Examen Date Qui est convoqué
Exam1 mercredi 24 mars 2027, 14h-16h (à confirmer) Tout le monde : centraliens, normaliens, Lyon 1, masters MeA, MA et GRAF, option ou non
Exam2 mercredi 24 mars 2027, le matin (horaire à préciser) Les centraliens de l'option seulement, en master ou non
Public Code Note finale
PGD/Option (3e année Centrale) I_G_S09_FO_MIR3_1 (Exam1 + Exam2)/2
PGD/Master (MeA) M_MAS_MEA_S3_OPT_01 (Exam1 + Projets)/2
ORP/Master (MeA) M_MAS_MEA_S4_ORP Exam2
Centraliens en master GRAF, Théorie M_ES_GRAF_S3_SML_1 (Exam1 + Projets)/2
Centraliens en master GRAF, Pratique M_ES_GRAF_S3_ETJ_1 Exam2

Schedule

Date Session Room
Mon 25 Jan 2027, 2–5:15pm Lecture 1: sparse recovery on a grid and dual certificates TBA
Mon 1 Feb 2027, 2–5:15pm Lecture 2: from the grid to measures, the BLASSO TBA
Mon 8 Feb 2027, 2–5:15pm Lecture 3: exact recovery and the construction of certificates TBA
Mon 1 Mar 2027, 2–5:15pm Lecture 4: noise, prediction, localisation and support stability TBA
Mon 15 Mar 2027, 2–5:15pm Lecture 5: the generalised BLASSO and Fenchel-Young losses TBA
Mon 22 Mar 2027, 2–5:15pm Projects: 15-minute talks (Guidelines) TBA
Wed 24 Mar 2027, morning (time to be confirmed) Exam 2: Centrale Lyon students of the MIR option only TBA
Wed 24 Mar 2027, 2–4pm (to be confirmed) Exam 1: all students TBA

Class time and location

Winter quarter (January–March 2027).
Lecture: Monday 2:00–5:15pm.
École Centrale Lyon.

Contact

Yohann: yohann.de-castro@ec-lyon.fr