I am currently a postdoctoral researcher in the SIERRA group with Francis Bach.
Before that, I was a postdoctoral researcher at Université Catholique de Louvain, in the department of mathematical engineering (part of the ICTEAM institute), where I was holding a F.R.S.-FNRS FRIA scholarship for my PhD under the supervision of François Glineur and Julien Hendrickx.
Broad research interests; current focus on theory and algorithms for convex optimization.
Research news (02/2019) - In a joint work (presented at COLT 2019) with Francis Bach, we developed a computer-aided technique for designing sequences of potential (or Lyapunov) functions, aiming at proving convergence results for (deterministic, randomized, and stochastic) first-order methods. The methodology also goes one step further toward automatic generation of first-order methods [PMLR, arXiv, codes].
News (01/2019) - 2017 best paper award in Optimization Letters, for a joint work with Etienne de Klerk and François Glineur (for this paper).
Research news (12/2018) - In a joint work with Ernest Ryu, Carolina Bergeling, and Pontus Giselsson, we developed the performance estimation approach for operator splitting methods [arXiv, codes].
News (09/2018) - My thesis (under the supervision of François Glineur and Julien Hendrickx) had the chance to be awarded the ICTEAM thesis award for 2018, the IBM-FNRS innovation award for 2018, and to be a finalist for the AW Tucker prize for 2018.
Research news (07/2018) - Performance estimation problems and integral quadratic constraints, two competing methodologies for automated worst-case analyses of optimization algorithms, were recently linked in a new work (joint with Bryan Van Scoy and Laurent Lessard) that was presented at the international conference on machine learning (ICML) 2018 [PMRL, arXiv, codes].
Research news (06/2018) - In a joint work with Yoel Drori, we propose a novel (constructive) approach to the design of optimal first-order methods for convex optimization [Springer, arXiv, codes].
Software news (12/2017) - The current version of the Performance EStimation TOolbox (PESTO) is available from here (full manual here). The numerical worst-case analyses from PEP can now be performed just by writting the algorithms just as you would implement them. This work will be presented at CDC2017 in Melbourne; the proceeding offers a nice introduction to the topic of automated worst-case analyses using performance estimation problems.