Harnessing the Structure of some Optimization Problems


Speciality : Mathématiques Appliquées

15/12/2021 - 15:00 Franck Iutzeler (Université Grenoble Alpes) Auditorium, IMAG building

Mathematical optimization is becoming more and more important in data science. This is partly due to the increasing difficulty of learning tasks but also to the particular structure of the associated minimization problems which makes them often tractable, sometimes distributable, but always interesting. This is the central thread of this habilitation.

In the first part, I study the mathematical characterization of the underlying structure of the solutions of regularized problems (\eg when a sparsity prior is added to the problem) as well as in the algorithmic exploitation of this phenomenon. The second part of this work deals with the resolution of minimization problems by several machines coordinated asynchronously by a central entity; this type of computation is again made possible by the particular structure of data science problems. Finally, some perspectives conclude this work.


Adeline Leclerc-Samson (Univ. Grenoble Alpes)


  • Alexandre d'Aspremont (ENS Paris )
  • Jérôme Bolte (Université Toulouse Capitole & TSE )
  • Adrian Lewis (Samuel B. Eckert Professor of Engineering, Cornell )


  • Jalal Fadili (ENSI Caen )
  • Julien Mairal (INRIA Grenoble )