Français Anglais
Accueil Annuaire Plan du site
Home > Research results > Dissertations & habilitations
Research results
Ph.D de

Group : Learning and Optimization

Introduction of Statistics in Optimization

Starts on 01/09/2008
Advisor : SCHOENAUER, Marc

Funding : A
Affiliation : Université Paris-Sud
Laboratory : LRI AO

Defended on 08/12/2011, committee :
- Damien Ernst, Université de Liège, rapporteur
- Abdel Lisser, Université Paris Sud, examinateur
- Martin Müller, Université d'Alberta, rapporteur (présent par vidéo-conférence)
- Liva Ralaivola, Université de Marseille, examinateur
- Frédéric Saubion, Université d'Angers, examinateur
- Marc Schoenauer, INRIA, directeur de thèse
- Olivier Teytaud, INRIA, co-directeur de thèse (présent par vidéo-conférence)

Research activities :

Abstract :
In this thesis we study two optimization fields. In a first part, we study the use of evolutionary algorithms for solving derivative-free optimization problems in continuous space. In a second part we are interested in multistage optimization. In that case, we have to make decisions in a discrete environment with finite horizon and a large number of states. In this part we use in particular Monte-Carlo Tree Search algorithms.

In the first part, we work on evolutionary algorithms in a parallel context, when a large number of processors are available. We start by presenting some state of the art evolutionary algorithms, and then, show that these algorithms are not well designed for parallel optimization. Because these algorithms are population based, they should be we well suitable for parallelization, but the experiments show that the results are far from the theoretical bounds. In order to solve this discrepancy, we propose some rules (such as a new selection ratio or a faster decrease of the step-size) to improve the evolutionary algorithms. Experiments are done on some evolutionary algorithms and show that these algorithms reach the theoretical speedup with the help of these new rules.
Concerning the work on multistage optimization, we start by presenting some of the state of the art algorithms (Min-Max, Alpha-Beta, Monte-Carlo Tree Search, Nested Monte-Carlo). After that, we show the generality of the Monte-Carlo Tree Search algorithm by successfully applying it to the game of Havannah. The application has been a real success, because today, every Havannah program uses Monte-Carlo Tree Search algorithms instead of the classical Alpha-Beta. Next, we study more precisely the Monte-Carlo part of the Monte-Carlo Tree Search algorithm. 3 generic rules are proposed in order to improve this Monte-Carlo policy. Experiments are done in order to show the efficiency of these rules.

Ph.D. dissertations & Faculty habilitations
The original manuscript conceptualizes the recent rise of digital platforms along three main dimensions: their nature of coordination devices fueled by data, the ensuing transformations of labor, and the accompanying promises of societal innovation. The overall ambition is to unpack the coordination role of the platform and where it stands in the horizon of the classical firm – market duality. It is also to precisely understand how it uses data to do so, where it drives labor, and how it accommodates socially innovative projects. I extend this analysis to show continuity between today’s society dominated by platforms and the “organizational society”, claiming that platforms are organized structures that distribute resources, produce asymmetries of wealth and power, and push social innovation to the periphery of the system. I discuss the policy implications of these tendencies and propose avenues for follow-up research.