PhD Students
ARISTIDOU Petros (Starting date : October 2010)
Domain decomposition and parallel processing for the simulation of the large sets of Differential-Algebraic equations in power systems
Thesis advisor : Thierry Van Cutsem
BEGON Jean-Michel (Starting date : October 2014)
Feature learning with tree-based ensemble methods for topologically structured data
Thesis advisor : Pierre Geurts
The objective of this thesis is to explore topological and structural feature learning facilities in the context of tree-based ensemble methods for classification tasks. The general objective of this exploration is to provide new supervised learning algorithms for topologically structured data (images, times series, graphs, etc.) that realize practically relevant tradeoffs in terms of accuracy, scalability, interpretability and portability.
BESSONOV Kyrylo
Thesis advisor : Kristel Van Steen
BOUSSOUFI Badre
Power systems engineering, smart grids
CHAICHOOMPU Kridsadakorn
Thesis advisor : Kristel Van Steen
CIRILLO Ilario (Starting date : October 2014)
Nonlinear modal analysis
Thesis advisors : Rodolphe Sepulchre and Gaetan Kerschen
The objective of the research is to develop novel techniques for the analysis of nonlinear systems with oscillating behavior. Focusing on the framework of the nonlinear normal modes, new methods will be introduced with the aim of obtaining a better understanding of these type of systems.
DETHIER Julie (Starting date : September 2011)
Task-oriented neural information processing from multichannel neural recordings
Thesis advisor : Rodolphe Sepulchre
Multichannel recording of neural signals has been an ongoing technological breakthrough over the last fifty years that allows for an ever growing number of simultaneous recordings. This evolving technology offers a range of new possibilities in computational neuroscience, neurophysiology and neuroengineering. So far, the processing of those signals has been conventional and based on independent neuron coding. However, standard signal processing imposes severe limitations to the envisioned neuroengineering applications due to computational burden and power consumption. Likewise, modeling questions in computational neuroscience will be helped by multichannel recordings provided that the signal processing treats the recordings as a statistical ensemble of interconnected signals rather than as a collection of independent individual signals.
This research project aims at narrowing the gap between computational neuroscience and neuroengineering by developing a probabilistic framework to capture ensemble coding that integrates recent advances of computational neuroscience and neural coding modeling. This novel framework will be validated in specific neuroprosthetics and neurophysiological benchmark applications studied in collaboration with partner laboratories.
DIZIER Benjamin
Thesis advisor : Kristel Van Steen
More information soon.
FOULADI Ramouna
Thesis advisor : Kristel Van Steen
More information soon.
GEMINE Quentin (Starting date: September 2012)
Active network management
Thesis advisors: Damien Ernst, Thierry Van Cutsem
HIARD Samuel
Machine learning, preference learning, ranking
JOLY Arnaud (Starting date : October 2011)
Random projections and compressed sensing for supervised learning in very high-dimensional input and output spaces
Thesis advisors : L. Wehenkel, P. Geurts
The objective of this thesis is to confront and then to combine the theory and algorithmics of compressed sensing and supervised learning. The research project investigates the integration of compressed sensing with statistical learning in order to improve the treatment of high dimensional problem.
LIEGEOIS Raphaël (Starting date : October 2011)
Structured Dynamical modeling in neuroimaging
Thesis advisor : R. Sepulchre
The aim of this research is to study neuroimaging outputs as a dynamical model at a macroscopic level. Focus will be put on how to integrate structural information to the model and how to relate structural and functionnal connectivity in the brain.
MARCOS ALVAREZ Alejandro
Machine learning, optimization, learning for search
Thesis advisor : Louis Wehenkel
MATHEI Alex
Computational methods for industrial diagnostics
MATHIEU Sébastien (Starting date: July 2012)
Demand side management in power systems
Thesis advisors : Quentin Louveaux and Damien Ernst
OLIVIER Frédéric (Starting date: August 2013)
Smart grids
Thesis advisors : Damien Ernst, Thierry Van Cutsem
PAPANGELIS Lampros (Starting date : September 2013)
Robustness and defence plans in future AC-DC power systems
Thesis advisor : Thierry Van Cutsem
PINEDA Silvia
Thesis advisor : Kristel Van Steen
More information soon.
QIU Aaron (Starting date: August 2013)
Smart Grids
Thesis advisor : Damien Ernst
SAFADI Firas (Starting date : September 1, 2010)
Thesis advisor : Damien Ernst
More information to follow soon
SCHRYNEMACKERS Marie (Starting date : October 2009)
Supervised inference of biological networks - application to the prediction of genetic interactions in yeast
Thesis advisor : P. Geurts
The general objective of this project is to develop new methods for supervised graph inference or improve existing ones and to apply them to different biological networks, starting with the genetic interaction network of the yeast. The expected contributions will be in the field of machine learning as well as in the field of biology.
SOLEIMANI Hamid (Starting date: July 2013)
Real-time control of active distribution systems
Thesis advisors: Thierry Van Cutsem and Damien Ernst
SUTERA Antonio (Starting date: October 1, 2013)
Characterization and extension of variable importance measures derived from forests of randomized trees
Thesis advisors : Pierre Geurts, Louis Wehenkel
The first objective of this project is to characterize as precisely as possible tree-based importance measures that have been proposed in the literature, from a theoretical and an empirical point of view. Our goal with this characterization is to assess the practical relevance of these measures and possibly to improve them. The second objective is then to propose new measures or algorithms to extract more complete information about an input-output relationship from tree-based ensemble models. In particular, the project will focus on questions of redundancy, interactions, and causality between input variables, three properties that are not well taken into account by existing importance measures.
VAN LISHOUT François
Thesis advisor : Krystel Van Steen
More information soon.
WEHENKEL Marie (Starting date : October 2014)
Spatiotemporal modeling of the dopaminergic neuron
Thesis advisors : Rodolphe Sepulchre and Vincent Seutin
Dopaminergic (DA) neurons have been studied a lot to date but some questions have not yet been resolved. Firstly, only few quantitative data are available on the exact density of various ion channel species throughout the various compartments of these neurons, which is a major obstacle to the building of a realistic computational model. Secondly, DA neurons have one very intriguing structural peculiarity. In a majority of them, the axon hillock is not appended to the soma (contrary to what is the case generally), but arises from a primary dendrite. To date, the impact of this organization and which role it plays in the electrical behaviour of DA neurons are not exactly known.
Thus, this thesis notably involves to improve semi-realistic (multicompartmental and conductance-based) models of DA neurons and obtain a more realistic and biophysically-based model of DA neurons, including spatiotemporal phenomena and experimental data from rat brains. Moreover, this work focuses on the prominent role played by the somatic and dendritic SK channels in the firing pattern of DA neurons.
Post-Docs
CARLI Francesca
Nonsmooth optimization on manifolds
Host : Rodolphe Sepulchre
CORNÉLUSSE Bertrand (Starting date: January 2013)
Smart Grids
Host : Damien Ernst
DRION Guillaume
Neurodynamics, modelling of dopaminergic neurons
Host : Rodolphe Sepulchre
FORNI Fulvio (Starting date : October 2011)
Contraction theory and applications in nonlinear control
Host : Rodolphe Sepulchre
The objective is to develop methods and algorithms based on contraction theory for the analysis of nonlinear systems, with the aim of casting several control problems (e.g. observers design, output regulation, antiwindup design) within a unifying framework based on contraction theory.
FONTENEAU Raphael
Evaluation of performances and identification of informative variables in the context of the inference from clinical data of dynamic treatment regimes.
Hosts : D. Ernst, L. Wehenkel
Nowadays, many diseases as for example HIV/AIDS, cancer, inflammatory or neurological diseases are seen by the medical community as being chronic-like diseases, resulting in medical treatments that can last over very long periods. For treating such diseases, physicians often adopt explicit, operationalized series of decision rules specifying how drug types and treatment levels should vary over time, called Dynamic Treatment Regimes (DTRs). While typically DTRs are based on clinical judgment and medical insight, since a few years the biostatistics community is investigating a new research field addressing specifically the problem of inferring in a well principled way DTRs directly from clinical data gathered from patients under treatment. This research project aims at studying two open problems related to this well pincipled way of designing DTRs : first, the prediction of the performances of DTRs using only clinical data, and second the development of methods to select the most relevant clinical indicators in order to build convenient DTRs.
FRANCI Alessio
Singularity theory applied to neuronal behaviors analysis
Host : Rodolphe Sepulchre
GADALETA Francesco
Host : Kristel Van Steen
GUSAVERA Elena
Host : Kristel Van Steen
KARANGELOS Efthymios
Probabilistic Methods for Power System Reliability Management
Hosts : Louis Wehenkel, Damien Ernst
MISHRA Bamdev
Low-rank factorizations for large-scale optimization algorithms
Host : Rodolphe Sepulchre