Mathieu Dagréou

Mathieu Dagréou

Ph.D. student in Machine Learning

Inria Saclay - Mind team

Biography

I am a Ph.D. student at Inria in the Mind team working under the supervision of Pierre Ablin , Thomas Moreau and Samuel Vaiter . I am currently working on bilevel optimization for Machine Learning.

Interests
  • Optimization
  • Machine Learning
Education

  • MSc in Mathematics, Vision and Machine Learning (MVA), 2021

    École Normale Supérieure Paris-Saclay

  • Engineering Diploma, 2020

    École Centrale de Nantes

Papers

Quickly discover relevant content by filtering publications.
Mathieu Dagréou, Thomas Moreau, Samuel Vaiter, Pierre Ablin (2024). A Lower Bound and a Near-Optimal Algorithm for Bilevel Empirical Risk Minimization. In AISTATS.

PDF Cite arXiv PMLR

Mathieu Dagréou, Thomas Moreau, Samuel Vaiter, Pierre Ablin (2023). Borne inférieure de compléxité et algorithme quasi-optimal pour la minimisation de risque empirique bi-niveaux. XXIXème Colloque Francophone de Traitement du Signal et des Images GRETSI.

PDF Cite

Mathieu Dagréou, Pierre Ablin, Samuel Vaiter, Thomas Moreau (2022). A framework for bilevel optimization that enables stochastic and global variance reduction algorithms. In NeurIPS (Selected for an oral - Top 2% among 10,411 submissions).

PDF Cite Code Poster Slides NeurIPS arXiv

Thomas Moreau, Mathurin Massias, Alexandre Gramfort, Pierre Ablin, Pierre-Antoine Bannier, Benjamin Charlier, Mathieu Dagréou, Tom Dupré La Tour, Ghislain Durif, Cassio F. Dantas, Quentin Klopfenstein, Johan Larsson, En Lai, Tanguy Lefort, Benoit Malézieux, Badr Moufad, Binh T. Nguyen, Alain Rakotomamonjy, Zaccharie Ramzi, Joseph Salmon, Samuel Vaiter (2022). Benchopt: Reproducible, efficient and collaborative optimization benchmarks. In Neurips.

PDF Cite arXiv Benchopt

Mathieu Dagréou, Pierre Ablin, Samuel Vaiter, Thomas Moreau (2022). Algorithmes stochastiques et réduction de variance grâce à un nouveau cadre pour l’optimisation bi-niveaux. In XXVIIIème Colloque Francophone de Traitement du Signal et des Images GRETSI.

PDF Cite

Recent Posts

How to compute Hessian-vector products?
With Thomas Moreau, Samuel Vaiter and Pierre Ablin we wrote a blogpost that got accepted in the blogpost track of ICLR2024. It explains the different ways to compute Hessian-Vector products with automatic differentiation and provides a benchmark comparing Jax and PyTorch in HVP computation.
Apr 4, 2024 1 min read

Teaching

I am Teacher’s Assistant for the following course

Talks & events

  • 2024-05-02 Poster Session at AISTATS (Valencia): A lower bound a near-optimal algorithm for bilevel empirical risk minimization
  • 2023-08-31 Poster Session at GRETSI (Grenoble): A lower bound a near-optimal algorithm for bilevel empirical risk minimization
  • 2023-06-26 Poster Session at the workshop Optimization and machine learning (Toulouse): A lower bound a near-optimal algorithm for bilevel empirical risk minimization
  • 2023-02-09: Talk at Center of Data Science (ENS): A framework for bilevel optimization that enables stochastic and global variance reduction algorithms
  • 2022-12-01: Poster session at NeurIPS (New-Orleans): A framework for bilevel optimization that enables stochastic and global variance reduction algorithms
  • 2022-10-12: Poster session at GDR MOA days (Nice): A framework for bilevel optimization that enables stochastic and global variance reduction algorithms
  • 2022-09-07: Poster session at GRETSI (Nancy): A framework for bilevel optimization that enables stochastic and global variance reduction algorithms
  • 2022-06-21: Poster session at Curves and Surfaces (Arcachon): A framework for bilevel optimization that enables stochastic and global variance reduction algorithms
  • 2022-04-05: Talk at the Parietal seminar: A framework for bilevel optimization that enables stochastic and global variance reduction algorithms
  • 2022-03-15: Talk at Proba-Stat seminar (LJAD Nice): A framework for bilevel optimization that enables stochastic and global variance reduction algorithms
  • 2022-03-03: Talk at the Miles team seminar (LAMSADE): A framework for bilevel optimization that enables stochastic and global variance reduction algorithms

Contact