Research Interests

  • Graphical models learning methods for
    • Bayesian networks,
    • Markovian and semi-markovian models,
    • Conditional random fields,
    • Influence diagrams and Markov decision process,
    • Chain graphs,
    • Multi-entity bayesian networks,
  • Automated knowledge discovery as
    • Causal discovery,
    • Identification of latent variables,
    • Temporal/dynamic influences,
    • Hierarchical dependences,
  • Machine Learning algorithms as
    • Discriminative methods,
    • Clustering algorithms,
    • Exponential family, kernel methods and CRF,
  • Modelisation and Simulation of complex systems.
top

Temporal Bayesian Network Modeling Maintenance Strategies

Prevention of Broken Rails

I am actually assistant researcher on a project with the Bombardier transport
This work is done at the french National Institute for Transport and Safety Research in the GRETTIA and in the team of Diagnosis and Maintenance.

top

Temporal Bayesian Network Modeling Maintenance Strategies

Prevention of Broken Rails

I have been assistant researcher on a project with the Autonomous control center of Parisian transport to evaluate the security level of the future implementation of the new train positionning system to evaluate the security level of new maintenance strategies that come with.
Our work consist in building a model of the rail degradation together with a modelisation of the maintenance strategies to simulate scenarios to evaluate the quality of the maintenance.
We have chose to use Partialy Observed Markov Decision Processes and learning from datasets and learning from experts knowledge together with experience return to build the model.
This work is done at the french National Institute for Transport and Safety Research (INRETS of Arcueil) in the New Technologies Laboratory (LTN) and in the team of Diagnosis and Maintenance of Guided Transport Systems.

top

Energy consumption Sociotechnical modelisation

Modeling Belief Network for Carbon Reduction

CaRB is a major research project funded by the EPSRC and Carbon Trust as part of Carbon Vision Buildings (CVB) - a research programme aimed at reducing carbon emissions from the UK building stock. CVB has been formed out of the Carbon Vision Partnership of the EPSRC and the Carbon Trust, and with the support of the ESRC.
My work in this project could be seen here.

top

PhD Thesis

From Bayesian Network Structure Learning to Pattern Recognition from Incomplete Knowledge.

This work was done at the Computer Science, Information Processing, and Systems Laboratory (LITIS, Rouen, Normandie, France) in the Perception and Machine Learning team.
My supervisor was Pr. Philippe LERAY.

The Thesis page. top

Structure Learning Package

All the function codes I realize during my PhD are freely available for use.
I am not responsible of trouble using those function.
They are encoded using Matlab 6.1.
One could download this package (which need the installation of the Bayes Net toolbox first) in a stable version on the project page or an unstable version here.
One could find support on the BNT mailing list or in french on the French BNT site or simply by mailing me.

The Structure Learning Package page.

top

Master Thesis

Utilisation des réseaux bayésiens pour l'aide au diagnostique médical

To achieve my master's degree, I realized a thesis on the use of bayesian networks to perform medical diagnosis.
I explored the performances of many classifiers based on bayesian networks, from the discrete naive bayes structure, mixte discrete and continous bayesian networks, the use of mixture of experts or the use of structure learning to obtain models.
My application were the Heart disease datasets from UCI repository. The paper is available in french.
This work was selected to appear in the french Artificial Intelligence Journals [draft]. top