Ecole
Centrale de Lille Cité
Scientifique BP 48 59651
Villeneuve d'Ascq Cedex, France
emmanuel(dot)duflos(at)ec-lille(dot)fr
From 1987 to 1991, I was in Lille (France) as undergraduate in
the Institut Supérieur
d'Electronique du Nord
an engineer school in the north of
France. Then I went for one year in Paris in the University of PARIS XI
(Orsay) to prepare a Diplôme d'Etude Approdondie (DEA) in Automatic
Control and Signal Processing. From 1992 to 1995 I had PhD position at
the University of Toulon where I defended, in September 1995,
my Ph.D. thesis in Signal Processing. From September 1995 to August
2003 I had had a position of teacher at the Institut Supérieur
d'Electronique du Nord where I was, from september 1999 to august 2003
the Head of the Signal, System and Communication Departement. In
September 1999 I also joined the laboratory LAIL (Laboratoire
d'Automatique et d'Informatique Industrielle de Lille) as a researcher.
The LAIL have become the
LAGIS (Laboratoire d'Automatique, Génie Informatique et
Signal) since january 2006. On December 16th, 2002, I obtained the
"Habilitation a Diriger des Recherches (HDR)", from the "Université des
Sciences et Technologies de Lille" (manuscript now downloadable on
Publications page). In september 2003, I obtained a position of
Professor at the Ecole
Centrale de Lille
where I am also the IT Manager and a member of the board of Directors.
In April 2006, I've participated to the creation of the INRIA project
team SequeL devoted
Sequential Learning. SequeL hosted by the INRIA Lille Nord
Europe.
Research
My fields of research are Signal Processing and
Data Fusion :
Bayesian
Analysis of Time series,
Monte-Carlo and Sequential Monte Carlo
Methods
Random Measures and Non Parametrical
Bayesian Analysis
Functional Regression
Sensors Management
Bayesian fusion / Belief functions
Applications :
Global Navigation Satellite System : GPS, Gallileo
Management of Electronically Scanned Array
Noise
estimation for speech recognition
Modelling
of the propagation channel for 60 GHz communication
"Optimal
Policies Search for Sensor Management : Application to the ESA Radar",
Thomas Bréhard, Pierre-Arnaud Coquelin, Emmanuel Duflos, Philippe
Vanheeghe, FUSION
2008 Conference, Cologne, Germany, June 30th, July 3rd 2008 (Download
pdf preview)
"Reception
State Estimation of GNSS satellites in urban environment using particle
filtering", Fleury Donnay Nahimana, Emmanuel Duflos, Juliette
Marais, FUSION 2008 Conference, Cologne, Germany, June
30th, July 3rd 2008 (Download pdf preview)
Optimal Sensor Management - Review and New Prospects
Context : A significative part of the SequeL research
activity concerns the so-called "Sensor Management Problem". In an
increasing number of applications the number of the sensors and the
amount of information they provide lead to a complexity that is
untractable by human operators when we need to answer the question :
how manage all these sensors so as to optimize the quality and the
quantity of the information needed to achieve the goal the system(s)
which embends the sensors. Today sensors aren't anymore limited to the
"measurement provider" role as they used to be several years ago.
Intelligent sensors are a able to determine when, where and during how
many time they must "look for" specific information.
Description of the work : Sensor Management is often
associated with the
tasking of sensors like radars and infrared cameras to detect, track
and identify targets in a complex environment. The main questions
arising being how long should each sensor observe the space, in which
direction should it observe, how do the sensors collaborate, which
amount of information must they transmit to the others, which strategy
of communication must they use ... The underlying question is therefore
the choice of an action (see as a set of specific actions) for each
sensor. The answer isn't straight forward because it mainly depends on
the goals of the systems which embends the sensors and on the limited
capabilities of each of its component; all of these components being in
interaction. Partially Obserbable Markov Decision Processes (POMDP) are
well adapted to model such a problem and it has been used for many
years. Unfortunately, the optimal solution, although existing, is
untractable in practice and sub-optimal solutions must be found. The
aims of this post-doctorate position consists in reveiwing the existing
modellings of Sensor Management Problem, in order to propose new
prospects to work on in terms of methodology but also in terms of
potential original application.
Deadline for application : not
known at the moment
Education
I have (or I've had!) classes in the following fields : Computer
Sciences, Signal Processing, Automatic Control and Statistics. See my education page for more
details.