Gauthier Gidel

Assistant Professor
Université de Montréal - DIRO
Core member of Mila Send an email


I am an assistant professor at Université de Montréal (UdeM) at DIRO and a core faculty member of Mila. I graduated my Ph.D. under the supervision of Simon Lacoste-Julien. During my Ph.D., I have been an intern at Sierra, ElementAI and DeepMind.

Link to my Google scholar.

Prospective Students

Please read this page before sending me any email (otherwise you may get no answer).


[NEW!]


  • I am teaching the data science class in french this fall.
  • Danilo, Zichu and Damien are starting

Research interests

My research aims at getting a better understanding of adversarial formulations for machine learning (ML). Particularly, I am interested in the following questions:
  • What are the fundamental reasons behind the great successes of adversarial formulations?
  • What are the new difficulties and issues arising when training models through differentiable games and how to tackle them?
  • How can we design more efficient training methods for differentiable games?
  • What is generalization and how is it impacted by the choice of the training method?

I identify to the fields of ML (JMLR, NeurIPS, ICML, AISTATS, COLT, and ICLR) and optimization (SIAM OP)

Students

Alumni

  • Chiara Régniez, Junior Data Scientist, Owkin
  • Leonardo Cunha, Software Engineer, Muvraline
  • Sunand Raghupathi, CEO of Seven Seas Strategies
  • Eduard Gorbunov, Research scientist at MBZUAI
  • Aleksandr Beznosikov Ph.D. at MIPT.
  • Michael Przystupa Ph.D. at University of Alberta.
  • Alexandre Duplessis M.Sc. at Oxford.
  • Chris Junchi Li (Visiting Postdoc co-supervised with Michael I. Jordan)

  • Preprints - Publications - Manuscripts


    With my currents responsabilitities, I have no time to update this website frequently. Thus, this List is not up-to-date, for my latest publications please check my Google Scholar

    • 37
      Beyond L1: Faster and Better Sparse Models with skglm
      By Quentin Bertrand, Quentin Klopfenstein, Pierre-Antoine Bannier, Gauthier Gidel, and Mathurin Massias.
      Preprint 2022

    • 36
      Last-Iterate Convergence of Optimistic Gradient Method for Monotone Variational Inequalities
      By Eduard Gorbunov, Adrien Taylor, and Gauthier Gidel.
      Preprint 2022

    • 35
      Variance Reduction is an Antidote to Byzantines: Better Rates, Weaker Assumptions and Communication Compression as a Cherry on the Top
      By Eduard Gorbunov, Samuel Horváth, Peter Richtárik, and Gauthier Gidel.
      Preprint 2022

    • 34
      Clipped Stochastic Methods for Variational Inequalities with Heavy-Tailed Noise
      By Eduard Gorbunov, Marina Danilova, David Dobre, Pavel Dvurechensky, Alexander Gasnikov, and Gauthier Gidel.
      Preprint 2022

    • 33
      A General Framework For Proving The Equivariant Strong Lottery Ticket Hypothesis
      By Damien Ferbach, Christos Tsirigotis, Gauthier Gidel, and Joey Bose.
      Preprint 2022

    • 32
      On the Limitations of Elo: Real-World Games, are Transitive, not Additive
      By Quentin Bertrand, Wojciech Marian Czarnecki, and Gauthier Gidel.
      Preprint 2022

    • 31
      Optimal Extragradient-Based Bilinearly-Coupled Saddle-Point Optimization
      By Simon S Du, Gauthier Gidel, Michael I Jordan, Chris Junchi Li.
      preprint 2022

    • 30
      Only Tails Matter: Average-Case Universality and Robustness in the Convex Regime
      By Leonardo Cunha, Gauthier Gidel, Fabian Pedregosa, Damien Scieur, and Courtney Paquette.
      ICML 2022

    • 28
      Extragradient method: O(1/K) last-iterate convergence for monotone variational inequalities and connections with cocoercivity
      By Eduard Gorbunov, Nicolas Loizou, and Gauthier Gidel.
      AISTATS 2022

    • 27
      Convergence Analysis and Implicit Regularization of Feedback Alignment for Deep Linear Networks
      By Manuela Girotti, Ioannis Mitliagkas and Gauthier Gidel.
      Unpublished. Presented at the Beyond first-order methods in ML systems ICML2021 workshop.

    • 25
      Generalized Natural Gradient Flows in Hidden Convex-Concave Games and GANs
      By Andjela Mladenovic*, Iosif Sakos, Georgios Piliouras, and Gauthier Gidel.
      ICLR 2022

    • 20
      Multi-player games in the era of machine learning

      Gauthier Gidel.
      Ph.D. thesis.

    • 5
      Adaptive Three Operator Splitting

      Fabian Pedregosa, Gauthier Gidel.
      Proceedings of the 35th International Conference on Machine Learning (ICML), 2018.

      Paper Poster Slides bibtex arXiv
      @inproceedings{pedregosa2018adaptive,
        author      = {Pedregosa, Fabian and Gidel, Gauthier},
        title       = {Adaptive Three Operator Splitting},
        booktitle   = {ICML},
        year        = {2018} 
                      }

    • 4
      Frank-Wolfe Splitting via Augmented Lagrangian Method

      Gauthier Gidel, Fabian Pedregosa and Simon Lacoste-Julien.
      Proceedings of the 21st International Conference on Artificial Intelligence and Statistics (AISTATS), 2018. (ORAL, TOP 5% of submitted papers)

      Paper Webpage Poster Slides bibtex
      @inproceedings{gidel2018fwal,
        author      = {Gidel, Gauthier and Pedregosa, Fabian and Lacoste-Julien, Simon},
        title       = {Frank-Wolfe Splitting via Augmented Lagrangian Method},
        booktitle   = {AISTATS},
        year        = {2018} 
                    }

    • 2
      Frank-Wolfe Algorithms for Saddle Point Problems

      Gauthier Gidel, Tony Jebara and Simon Lacoste-Julien.
      Proceedings of the 20th International Conference on Artificial Intelligence and Statistics (AISTATS), 2017.

      Paper bibtex arXiv
      @inproceedings{gidel2017saddle,
        author      = {Gidel, Gauthier and Jebara, Tony and Lacoste-Julien, Simon},
        title       = {{F}rank-{W}olfe Algorithms for Saddle Point Problems},
        booktitle     = {Proceedings of the 20th International Conference on Artificial Intelligence and Statistics (AISTATS)},
        year        = {2017} 
                      }
      HAL Code Project Poster

      Accepted as an oral presentation at the NeurIPS OPT2016 workshop
      (TOP 10% of accepted submissions).
    • 1
      Extensions de l’algorithme de Frank-Wolfe pour la recherche de points selles. (In French)

      Gauthier Gidel.
      Introduction au domaine de recherche (Equivalent of Master's thesis) au DMA, ENS 2016.


    Talks


    2021
    2020
    • Job talk, UdeM 2020. Two-player Games in the Era of Machine Learning. Slides
    2019
    2018
    2017 2016

    Reviewing


  • NeurIPS 2018 Top 200 reviewer.
  • ICLR 2019.
  • AISTATS 2019.
  • ICML 2019 (top 5% reviewer).

  • Teaching Experience


    Mathematics Examiner in CPGE (most competitive undergrad program in France)


    Third Person Bio


    Gauthier Gidel is an assistant professor at Université de Montréal (UdeM), a core faculty member of Mila, and a recipient of a CIFAR AI Chair. His research aims to build a better understanding of adversarial formulations for machine learning. He’s well known for his work in games, including theoretical analyses of GANs and the introduction of the extragradient method to the deep learning community. Gauthier co-organized a popular series of workshops on smooth games during NeurIPS. More recently, he also organized two editions of the ICLR blog post track.

    Contact


    Gauthier Gidel
    Pavillon André-Aisenstadt
    2920 Chemin de la Tour, office 3373
    Montreal, QC
    H3T 1J4 CANADA

    © 2019