
Guillaume Garrigos
My research essentially focuses on dynamical systems associated to optimization problems. In particular, I am interested in algorithms and regularization methods for nonsmooth (eventually nonconvex) problems arising in signal/image processing and machine learning. I am particularly interested in the interplay between the modelization of a learning/inverse problem as an optimization problem, and its resolution. My papers are here.
I am a CNRS postdoc researcher at the École Normale Supérieure (DMA), member of the NORIA (Numerical Optimal tRansport for ImAging) research group.
Formerly, I used to be a postdoc researcher at the Laboratory for Computational and Statistical Learning, a joint lab between the IIT and MIT. Before that I did a PhD in Optimization between the Université de Montpellier and the Universidad Santa Maria.
Upcoming news
-
Upcoming conferences: Workshop "Parcimonie et applications" (May 03, Bordeaux), Inverse Problems: Modeling and Simulation (May 21, Malta), SIAM Conference on Imaging Science (Jun 05, Bologna), International Symposium on Mathematical Programming (Jul 02, Bordeaux), European Signal Processing Conference (EUSIPCO) (Sep 03, Roma).
Others: I organize, together with Robert Gower, a session at ISMP titled "A sketchy session: Dimensionality reduction tools for learning", featuring Nicolas Keriven, Florent Krzakala, Arthur Mensch and Alessandro Rudi.
News
-
Mar302018
Best talk award
I received the Dodu Award for my talk about consitency of low-complexity models in inverse problems and learning, that I gave the Journées SMAI-MODE.
-
Mar232018
Papers release
In the last months I submitted a few papers discussing the consitency of low-complexity models (sparsity, low-rank, ...), their implications from an optimisation perspective, and applications to learning problems:
Thresholding gradient methods in Hilbert spaces: support identification and linear convergence ,
with L. Rosasco and S. Villa, Preprint on arXiv:1712.00357.
Sparse Multiple Kernel Learning: Support Identification via Mirror Stratifiability,
with L. Rosasco and S. Villa, Preprint on arXiv:1803.00783.
Model Consistency for Learning with Mirror-Stratifiable Regularizers,
with J. Fadili, J. Malick and G. Peyré , Preprint on arXiv:1803.08381.
-
Feb212018
Seminars : a Winter Tour
In the last three months I have been presenting my work on Iterative regularization for inverse problems in a few places. A poster at the MIA conference (Berlin, Jan 15), and talks at the Séminaire LJK-Probabilités & Statistique (Grenoble, Dec 7), V-Coloquio (Valparaison, Chile, Dec 11), Séminaire CVN (École Centrale Paris, Jan 20), LCSL Machine Learning Seminar (Universita di Genova, Italy, Jan 29), Séminaire Images Optimisation et Probabilités (Université de Bordeaux, Feb 07), Séminaire commun de Statistique P6-P7 (Paris, Feb 20). Slides of my talk can be found here.
-
Dec142017
Workshop
I spoke at the Workshop on Dynamics and Algorithms in Convex Optimization in Santiago de Chile, about optimization and inverse problems.
-
Nov182017
Workshop
I was at the workshop "Gradient dynamics, convergence and related topics" organized by Jérome Bolte and Sylvain Sorin at Université Pierre et Marie Curie, Paris. I gave a talk intitled "Quadratic functions in Hilbert spaces: Lojasiewicz inequality and Sobolev regularity", based on the work done in this paper (see Section 5.1).
-
Oct172017
Conference
I was participating to the annual conference of the GdR MOA (Mathématiques de l'Optimisation et Applications), organized this year in Bordeaux, together with the GdR MIA. I gave a talk on the rates of optimization methods and their interplay with inverse problems.
-
Oct012017
Post-doc position
I am starting a post-doc at the École Normale Supérieure in Paris, where I am part of the new NORIA (Numerical Optimal tRansport for ImAging) research group. I will collaborate with Gabriel Peyré on inverse problems in imaging and optimal transport.
-
Sep82017
Conference
I was invited to give a talk by Clément Royer, who were organizing a session at Optimization 2017, a conference organized this year in Lisboa.
-
Aug12017
Paper accepted
Our paper "Iterative regularization via dual diagonal descent" has been accepted in Journal of Mathematical Imaging and Vision. It is available on the editor website and arXiv. See also some recent slides.