Tam Le

Tam Le

Postdoc at LJK, Université Grenoble Alpes

Hello! I am a postdoctoral researcher in the team DAO, at LJK, Université Grenoble Alpes. I work on statistical aspects of Wasserstein distributionally robust models.

Previously, I completed my PhD (October 2023) at Toulouse School of Economics (TSE) and ANITI, supervized by Jérôme Bolte and Edouard Pauwels. I worked on stochastic, nonsmooth and nonconvex optimization in machine learning. Some questions I studied there included convergence guarantees when using automatic differentiation in stochastic algorithms (e.g. SGD in deep learning), and nonsmooth implicit differentiation applied to optimization layers and hyperparameter selection.

[CV (fr)].

Pre-prints

  • Universal Generalization Guarantees for Wasserstein Distributionally Robust Models, T. Le and J. Malick, under review [preprint].
  • Inexact subgradient methods for semialgebraic functions, J. Bolte, T. Le, E. Moulines and E. Pauwels, [preprint].

Publications

  • Nonsmooth Implicit Differentiation for Machine Learning and Optimization J. Bolte, T. Le, E. Pauwels, A. Silveti-Falls, NeurIPS 2021. [paper]
  • Subgradient sampling for nonsmooth nonconvex minimization, J. Bolte, T. Le, E. Pauwels, SIAM Journal on Optimization 2023 [paper]
  • Nonsmooth nonconvex stochastic heavy ball, to appear in Journal of Optimization Theory and Applications 2024 [paper]

Thesis

Nonsmooth calculus and optimization for machine learning: first-order sampling and implicit differentiation, T. Le, PhD Thesis, 2023. Advised by Jérôme Bolte and Edouard Pauwels. [manuscript] [slides]

Communications

** Generalization guarantees of Wasserstein robust models

  • Journées SMAI-MODE Lyon (talk), 2024

** Nonsmooth nonconvex stochastic heavy ball, Mathematical Optimization research seminar, University of Tübingen (online talk), 2024.

** Nonsmooth implicit differentiation in machine learning and optimization

  • ANITI-PRAIRIE workshop, Toulouse (poster), 2023.
  • Neurips (online poster), Neurips Paris event (poster), 2021.
  • Stat-Eco-ML seminar, CREST (talk), 2021.

** Subgradient sampling in nonconvex minimization (talks).

  • EUROPT, Budapest 2023.
  • SIAM Conference on optimization, Seattle 2023.
  • PGMO Days, Paris 2022.
  • GdR MOA Days, Nice 2022.
  • Mathematical Optimization research seminar, University of Tübingen (online), 2022.
  • ICCOPT, Bethlehem (Pennsylvania) 2022.
  • French-German days Inria, Le Chesnay-Rocquencourt 2021.
  • Toulouse School of Economics, PhD students seminar.

Teaching

I gave several tutorials at Université Toulouse 1 Capitole and TSE (64 H / year):

** 2022

  • R for data science and statistics (M1 Data science for social sciences)
  • Optimization for big data (M1 Data science for social sciences)
  • PyTorch tutorial for Deep Learning (M2 Data science for social sciences)
  • Optimization (L3 Economics)

** 2021

  • Mathematics for Economics and Management, (L1 Economics and Management)
  • Mathematics, Undergraduate (L1 Economics and Mathematics)
  • Analysis and Optimization, (L3 Economics and Mathematics)

** 2020

  • Support course in mathematics (L1)
  • Descriptive statistics (L1)
  • Mathematics for Management(L1)

Reviewer

I served as a reviewer for AISTATS (2023), SIAM Journal on optimization and Mathematical programming.

Education

  • Ph.D. in Applied Mathematics, Toulouse School of Economics, 2020 - 2023
  • MSc in Machine Learning and Computer Vision, ENS Paris-Saclay, 2019 - 2020
  • MSc in Statistics and Machine Learning, ENSAE Paris 2017 - 2020