Colloquium d’Informatique de Sorbonne Université
Mathieu Feuillet, Camille Couprie, Mathilde Noual, Lauréats du prix de thèse Gilles Kahn 2012

Tuesday 25 June 2013 18:00
Amphi 15 Sorbonne University - Faculté des Sciences

Espoirs : Winners of the 2012 Gilles Kahn prize

Mathieu Feuillet, Camille Couprie, Mathilde Noual

Le prix Gilles Kahn récompense chaque année une excellente thèse en informatique. Il est donné par la Société Informatique de France et est patronné par l'Académie des Sciences. Ce prix est destiné à promouvoir toutes les facettes de l'informatique, des travaux fondamentaux aux travaux appliqués ayant donné lieu à transfert industriel, de ceux réalisés dans les grands centres à ceux réalisés dans des centres plus modestes. L'objectif de ce prix est de dynamiser et de motiver de jeunes chercheurs en les récompensant, et de faire connaître à l'ensemble de la communauté informatique d'excellents travaux de recherche.
The Gilles Kahn Prize is awarded yearly to an excellent PhD thesis in informatics. It was created by the Société informatique de France (SIF, the French learned society in Informatics) and sponsored by the Académie des Sciences.

Abstract

<h3>(Main Prize) Mathieu Feuillet: Algorithm design via stochastic modelling: an example from wireless networks
As wireless networking continues to develop, efficient use of the frequency spectrum, a rare shared resource, is crucial. We show that CSMA, the random-access dynamic bandwidth allocation protocol currently used by WiFi, is inefficient, and we present a simple, fully distributed algorithm that fixes it.
In this talk, we will give a brief introduction on network stochastic modelling. In particular, we will give an intuition of scaling analysis and stochastic averaging, which are used to analyse stochastic processes resulting from this kind of modelling. In the second part of the talk, we will illustrate these tools on the example of CSMA. In particular, we will explain how this approach allows us to give a precise definition of "efficient" and prove that an algorithm is efficient or not.

(Second prize) Camille Couprie: Graph-based variational optimization and applications in computer vision

Image processing and analysis techniques are prevalent in various fields, such as medical imaging, materials analysis, or photography. They are based on techniques such as image segmentation, i.e., the extraction of patterns or objects, for the purpose of quantification or visualization. The classical “Watershed” segmentation algorithm “floods” the image to extract contours in the areas where different catchment basins meet. This technique is widely used in image processing, due to its linear complexity and to its ability to segment images in an arbitrary large number of regions.
The main result of my PhD shows how Watershed can be used to optimize a cost function that is very popular in computer vision and image processing. This work has two main impacts:

It provides a unified model for variational segmentation algorithms.
It interprets the Watershed algorithm as a general optimization method, allowing it to be used in a variety of computer vision tasks, such as filtering, surface reconstruction, stereo vision, object class segmentation.

(Second prize) Mathilde Noual: About time and the structure of interaction systems

From genetic regulation systems to computation models and also social networks, many systems encountered daily can be seen as networks of interacting entities in which all events that can take place result from interactions between entities. To better understand these systems, one can consider the characteristics of each specific system and aim at modelling it as faithfully as possible. But one can also consider instead a much cruder modelling of their common mechanisms. In these lines, I propose to explore some properties that are intrinsic to being and interaction system.