Laureate 2020 :
The 2020 laureate of the PGMO Thesis Award are:
Tristan Garrec
Tristan Garrec a effectué son doctorat en mathématiques à l’Université de Toulouse Capitole sous la direction de Jérôme Renault. Il est l'auteur d'une thèse élégante de théorie des jeux dynamiques, portant sur l’existence de l’équilibre uniforme dans les jeux stochastiques produits communicants, la caractérisation de la valeur et des stratégies optimales dans les jeux de recherche et de dissimulation dans des espaces continus ou des graphes aléatoires, et enfin, la caractérisation de la martingale optimale dans un jeu répété de persuasion bayesienne, les problématiques de ces deux dernières thématiques étant empruntées respectivement aux problèmes de sécurité d’un côté et d'économie et finance de l’autre.
Tristan Garrec utilise un large spectre de techniques différentes, combinant optimisation, probabilités, processus stochastiques, dans des cadres discrets et continus dans l’espace ou dans le temps. Cette thèse impressionne par la diversité des sujets et d'outils, ainsi que par la finesse et profondeur des résultats obtenus.
Tristan Garrec a publié trois articles significatifs dans Journal of Mathematical Analysis and Applications et European Journal of Operational Research, dont deux comme auteur unique.
Maturin Massias
Mathurin Massias a préparé sa thèse à Telecom Paris / Inria Saclay sous la direction d'Alexandre Gramfort et Joseph Salmon, sur le développement de méthodes variationnelles en neuro-imagerie (MEG - EEG), pour des problèmes d'identification de sources associés à une physique linéaire bruitée. Le sujet central est la mise au point d'algorithmes efficaces d'optimisation convexe dans un contexte de régression parcimonieuse (problème Lasso).
Les travaux de Mathurin Massias apportent des contributions significatives et originales à ce type de problème dans le cadre de la neuro-imagerie, avec un potentiel de répercussion plus large en optimisation.
En exploitant la structure de parcimonie de la solution et des techniques de dualité, les solveurs itératifs proposés par Mathurin Massias accélèrent deux types de méthodes classiques en optimisation : la descente de gradient proximale et l’optimisation coordonnée par coordonnée, tout en permettant une estimation automatique des paramètres de régularisation.
Mathurin Massias a écrit deux packages Python d'un niveau remarquable : l'un, pour traiter la régression parcimonieuse linéaire (en complément de scikit-learn), l'autre, à destination des praticiens, interactif pour l’analyse de signaux M/EEG.
Laureate 2019 :
The 2019 laureate of the PGMO Thesis Award are:
Charles Bertucci
Contributions à la théorie des jeux à champ moyen"
"The PGMO Prize 2019 is awarded to Charles BERTUCCI for his PhD thesis entitled "Contributions à la théorie des jeux à champ moyen ». His PhD, completed at the Université Paris Dauphine under the supervision of Pierre-Louis Lions, is concerned with the study of several problems of significant interest arising in the theory of mean field games. His work has been recognized as an extremely important contribution to the theory with the introduction of deep and original techniques."
Cécile Rottner
Aspects combinatoires du Unit Commitment Problem"
"The PGMO Prize 2019 is awarded to Cécile ROTTNER for her PhD thesis entitled « Aspects combinatoires du Unit Commitment Problem ». Cécile ROTTNER obtained her PhD from Sorbonne Université under the supervision of Pascale Bendotti of EDF R&D and Pierre Fouihoux of the computer science lab LIP6. In her thesis, she investigates the Unit Commitment Problem from the point of view of combinatorial and discrete optimization. She is awarded the PGMO Prize 2019 in recognition of her impressive treatment of the Unit Commitment Problem which contains a broad spectrum of novel and relevant insights and approaches."
Laureate 2018 :
The 2018 laureate of the PGMO Thesis Award are:
Nicolas BONIFAS
Geometric and Dual Approaches to Cumulative Scheduling
" The work of Nicolas Bonifas falls in the scope of constraint-based scheduling. In this framework, the most frequently encountered resource constraint is the cumulative, which enables the modeling of parallel processes. In his thesis, Nicolas studies the cumulative constraint with the help of tools rarely used in constraint programming (polyhedral analysis, linear programming duality, projective geometry duality) and propose two contributions for the domain. Cumulative strengthening is a means of generating tighter redundant cumulative constraints, analogous to the generation of cuts in integer linear programming. This is one of the first examples of a redundant global constraint. Energy Reasoning is an extremely powerful propagation for cumulative constraint, with hitherto a high complexity of O(n^3). Nicolas proposes an algorithm that computes this propagation with a O(n^2 log n) complexity, which is a significant improvement of this algorithm known for more than 25 years. Nicolas Bonifas, Geometric and Dual Approaches to Cumulative Scheduling . Université Paris-Saclay, 19/12/2017. http://www.theses.fr/2017SACLX119."
Nicolas Flammarion
L'approximation stochastique et régression par moindres carrés avec applications en apprentissage automatique.
" Many problems in machine learning are naturally cast as the minimization of a smooth function defined on a Euclidean space. While small problems are efficiently solved by classical optimization algorithms, large-scale problems are typically solved with first-order techniques based on gradient descent. Nicolas Flammarion considers, in his thesis, the particular case of the quadratic loss. He addresses its minimization when gradients are only accessible through a stochastic oracle and proposes optimal algorithms in different cases. His work offers many perspectives of applications of the quadratic loss in machine learning. Clustering and estimation with shape constraints are the two first applications already considered. Nicolas Flammarion. Stochastic approximation and least-squares regression, with applications to machine learning. Paris Sciences et Lettres, 24/07/2017. http://www.theses.fr/2017PSLEE056. "
The jury for the 2018 edition was chaired by Mathilde Mougeot and composed of :
Members appointed by the PGMO Scientific Council :
Mathilde Mougeot, ENSIIE et Université Paris-Diderot, Présidente du Jury
Gabriel Peyré, ENS, CNRS
Nicolas Vieille, HEC
Members appointed by the ROADEF
Céline Gicquel, Université Paris-Sud
Sandra Ulrich Ngueveu, LAAS, CNRS
Michael Poss, Université de Montpellier
Members appointed by the group MODE of SMAI
Jean Baptiste Caillau, Université de la Côte d’Azur
Thierry Champion, Université de Toulon.
Aude Rondepierre, INSA de Toulouse
Laureate 2017 :
The 2017 laureate of the PGMO Thesis Award are:
Vincent Cohen-Addad
From Practice to Theory : Approximation schemes for clustering and network design under the direction of Claire Mathieu
"This thesis contains pathbreaking and practically very important results concerning local search heuristics for clustering (k-means, k-median) and network design (traveling salesman, Steiner tree). It establishes some structural properties under which these local search heuristics perform very well and even yield polynomial time approximation schemes for these problems."
Joon Kwon
Stratégies de descente miroir pour la minimisation du regret et l'approchabilité sous la direction de Rida Laraki & Sylvain Sorin
"The thesis begins with a remarkably clear presentation of the basics of online linear optimization, regret minimization, mirror descent and approachability. The author develops the analysis of the classical problem of prediction with expert advice in which the outcome vector is assumed to be sparse, and design of optimal approachability strategies for the problem of prediction under partial monitoring. The author also shows how a continuous mirror descent motivates a large set of minimization algorithms in discrete time and the thesis ends with an elegant result bounding variations of convex functions."
The jury for the 2017 edition was chaired by Guillaume Carlier and composed of :
Members appointed by the ROADEF
Clarisse Dhaenens, CRIStAL, Université de Lille
Marcel Mongeau, ENAC
Sourour Elloumi, ENSTA
Members appointed by the group MODE of SMAI
Guillaume Carlier, CEREMADE (Président)
Anatoli Juditsky, Laboratoire Jean Kuntzmann, Grenoble
Jalal Fadili, ENSICAEN
Members appointed by the PGMO Scientific Council :
Luce Brotcorne, INRIA Lille
Julien Mairal, INRIA Grenoble
Jérôme Renault, Toulouse School of Economics
Laureate 2016 :
The 2016 laureate of the PGMO Thesis Award are:
Pauline SARRABEZOLLES
Pauline Sarrabezolles obtained her PhD thesis in Applied Mathematics at Université Paris-Est and ENPC ParisTech under the supervision of Frédéric Meunier.
The title of the thesis is "colorful linear programming" and it stands at the intersection of discrete mathematics, combinatorics, optimization, graph theory and algorithmics.
Colorful linear programming is an extension of linear programming where the variables are assigned to different categories (colors) and their number in each category is bounded.
It has many applications in geometry and complex optimization problems.
She studied the complexity of some algorithms like a colorful version of the simplex algorithm and proved a combinatorial conjecture in connection with the colorful Carathéodory theorem.
The jury was impressed by the unique combination of various skills used by Pauline Sarrabezolles to solve these problems.
Bruno ZILIOTTO
Bruno Ziliotto obtained his PhD thesis in Applied Mathematics at Université de Toulouse under the supervision of Jérôme Renault. Its title is "Stratégies et paiements de long terme dans les jeux répétés à deux joueurs" and it is concerned with asymptotics of repeated zero-sum games, possibly with stochastic aspects. In particular it disproves a long-standing conjecture on the existence of a limit value and of a limit optimal strategy for the player. By the dynamic programming approach, this result has a link with the homogenization of stochastic Hamilton-Jacobi equations. More precisely, Bruno Ziliotto also found a striking counter-example of a non-convex Hamiltonian for which no stochastic homogenization occurs. The jury particularly appreciated the various and deep results obtained in different areas of mathematics and optimization.
The jury for the 2017 edition was chaired by Grégoire ALLAIRE and composed of :
Members appointed by the ROADEF
Eric Gourdin, Orange Labs
Christelle Jussien, Université d'Angers
Aziz Moukrim, UTC
Members appointed by the group MODE of SMAI
Francis Bach, INRIA Paris--Rocquencourt
Maitine Bergounioux, Université d'Orléans
Michel de Lara, ENPC
Members appointed by the PGMO Scientific Council :
Marianne Akian, INRIA Saclay--Ile-de-France
Grégoire Allaire, École polytechnique
Marie-Christine Costa, ENSTA
Laureate 2015 :
Benjamin MARTIN
Benjamin Martin prepared his PhD thesis in Nantes in Computer Science after a Bachelor in Mathematics at the University of Nantes and a Master Degree in Computer Science at the University of Nantes too. The thesis directed and co-directed by Laurent Granvilliers, Alexandre Goldsztejn, Christophe Jermann is titled « Rigorous Algorithms for non-linear biobjective optimization ». The thesis deals with the interval based rigorous algorithm, i.e. with guaranteed results, to solve biobjective problems. The candidate proposes a certified continuation method that tracks locally a connected manifold of optimal solutions, which supplements other techniques from the literature. The proposed method adapts finely to the shape of manifolds and deals with singularities resulting from inequality constraints in biobjective problems. Moreover, the candidate develops an interval Branch & Bound (B&B) algorithm that globally computes a verified enclosure of the optimal solutions. This method integrates constraint propagation techniques, noticeably exploiting bounds on the objectives, in order to enhance the solving process. The jury particularly appreciated the fact that the thesis presents both strong theoretical and applied results.
Samuel VAITER
Samuel Vaiter did is PhD thesis in Mathematics at the Univerity Paris Dauphine under the direction of Gabriel Peyré. He studied Computer Science and Mathematics at the ENS Lyon (Bachelor) and ENS Cachan (Master) respectively. The thesis is titled « Low Complexity Regularization of Inverse Problems ». This thesis is concerned with recovery guarantees and sensitivity analysis of variational regularization for noisy linear inverse problems. This is cast as a convex optimization problem by combining a data fidelity and a regularizing functional promoting solutions conforming to some notion of low complexity related to their non-smoothness points. This thesis makes a very nice contribution to the field of linear inverse problems, convex geometry and analysis. The candidate has provided a unified framework for analyzing the robustness (vis a vis noise) and sensitivity of solutions to the inverse problem. The results are sharp enough to recover some of the known results for special instances. At the same time, the framework is general enough to accommodate most regularizers used in practice. The jury particularly appreciated the fact that Samuel Vaiter was able to put under a single umbrella a series of techniques and results for treating a variety of problems.
The jury for the 2015 edition was chaired by Roberto Wolfler (Paris Nord LIPN) and composed of :
- Frédéric BONNANS, Jean-Baptiste HIRIART-URRUTY et Marc QUINCAMPOIX pour la SMAI
- Luce BROTCORNE, Jean-Charles BILLAUT et Arnaud RENAUD pour la ROADEF
- Michel DE LARA, Laurent DUMAS et Roberto WOLFLER pour le PGMO
Laureate 2014
Daniel HOEHENER
Daniel HOEHENER prepared his PhD thesis at Université Pierre et Marie Curie (Paris 6) on "Conditions d’optimalité pour des problèmes de contrôle optimal avec contraintes d’états'' (Optimality conditions for some optimal control problems with state constaints), supervised by Hélène Frankowska. His thesis includes several original results about second order conditions, proven under more general assumptions than the results already existing in literature, expressed in primal form, and involving both state and control constraints. The jury committee was impressed by the novelty of the results in a very technical domain, which also gave rise to excellent publications.
Miquel OLIU BARTON
Miquel OLIU BARTON also prepared his PhD thesis in Paris 6, under the supervision of Sylvain Sorin. The dissertation, on "Jeux dynamiques à information incomplète en temps discret et continu" (Dynamical games with incomplete information in dicrete and continuous time) is mainly devoted to the long-time behavior in differential and repeated games and control problems, and to games with lack of information. The jury committee appreciated the impressive amount of works and the novelty of the results, and in particular the new proof that M. Oliu Barton gave of the existence of a limit for the value in discounted games, with new and self contained techniques.
The jury for the 2014 edition was chaired by Filippo SANTAMBROGIO and composed of :
- Jerome Bolte, Pierre Cardaliaguet et Jean Bernard Lasserre pour la SMAI
- Pierre Lemaire, Frédéric Meunier et Safia Kedad-Sidhoum pour la ROADEF
- Alexandre d'ASPREMONT, Roberto WOLFER et Filippo SANTAMBROGIO pour le PGMO
Laureate 2013
Olivier FERCOQ
Olivier Fercoq a fait sa thèse sous la direction de Stéphane Gaubert, Marianne Akian (X, CMAP +INRIA) et Mustapha Bouhtou (Orange Labs).
Le sujet était: "Optimisation de vecteurs propres de Perron et applications: du référencement de pages web à la chonothérapie."
Cette thèse constitue une "contribution majeure dans le domaine de l'optimisation de fonctions d'utilité sur l'ensemble des matrices positives" (selon l'un des rapporteurs). Elle présente à la fois un ensemble de résultats théoriques (propriétés, analyse de complexité,...) et des applications intéressantes.
Thibaut VIDAL
Thibaut Vidal a fait sa thèse sous la direction de Michel Gendreau, Theodor Crainic et Christian Prins, avec une co-tutelle entre les universités de Montréal et Troyes (CIRRELT+LOSI).
Le sujet était: "General solution approaches for multi-attribute vehicle routing and scheduling problems".
Cette thèse peut être vue comme "un recueil des résultats les plus intéressants et les plus marquants des dernières décades dans le domaine des recherche de solutions heuristiques pour les tournées de véhicules"(selon l'un des rapporteurs) et "un pas en avant dans la quête d'une approche générale de résolution d'une large palette de problèmes de tournées" (selon l'autre). Elle présente des résultats génériques concernant les méta-heuristiques pour les problèmes riches de tournées et pouvant être étendus à d'autres problèmes, une analyse complète de ces problèmes et s'appuie sur de lourdes réalisations informatiques.
