page d'accueil   sommaire   faculté  

Machine Learning Group [Machine Learning Group] (MLG)
Faculté des Sciences / faculty of Sciences - Informatique (unité ULB741)

Le Laboratoire en apprentissage automatique a été fondé en 2004 par Gianluca Bontempi. En Octobre 2008 Tom Lenaerts a rejoint le groupe. Les activités de recherche concernent l'apprentissage automatique, la fouille intelligente des données et leur applications à la bioinformatique et la biologie computationnelle. [The ULB Machine Learning Group (MLG) was founded within the Computer Science Department of the Faculty of Sciences in 2004 by Gianluca Bontempi. The activity of the group covers the areas of machine learning, computational modelling and statistics and their applications in data mining, simulation and time series prediction. In October 2008, Tom Lenaerts joined MLG as a new academic, extending the group's expertise towards computational biology, evolutionary dynamics and complex systems research. Currently we focus on:Data mining, modeling and predictionBioinformatics and computational biologyDynamics of cooperation and competition]



coordonnées / contact details


Machine Learning Group [Machine Learning Group]
tel +32-2-650.55.91
http://mlg.ulb.ac.be
Campus de la Plaine, NO8
CP212, boulevard du Triomphe, 1050 Bruxelles

Pour en savoir plus, consultez le site web de l'unité.



responsables / head


Gianluca BONTEMPI Tom LENAERTS


composition / members


Martin BIZET Fabrizio CARCILLO Reggiani CLAUDIO Antonio COLAPRICO Matthieu DEFRANCE Jacopo DE STEFANI Elias FERNANDEZ Andrea GAZZO Yann-Aël LE BORGNE Liran LERMAN Catharina OLSEN Gabriele ORLANDO Sofia PAPADIMITRIOU Daniele RAIMONDI


projets / projects


ICT4REHAB - "Advanced ICT Platform for Rehabilitation" [ICT4REHAB - ''Advanced ICT Platform for Rehabilitation'']
The Advanced ICT Platform for Rehabilitation (ICT4Rehab) Project is a 3 years strategic platform project funded by IWOIB.The ICT4Rehab project addresses a new rehabilitation paradigm in the domain of muscle spasticity rehabilitation. It will require the integration of various data sources, the development of novel algorithms and putting into practice several ICT tools supporting 2D/3D user interaction. Two important aspects of clinical spasticity management will be combined: maintaining optimal patient motivation during her/his physical therapy data handling and processing of the spasticity-related clinical data.A portable rehabilitation system with local intelligence providing patient feedback and ensuring clinical relevance of the physical activity will be developed. Fusion of clinical data acquired in controlled (hospital) conditions with data coming from the personalised system will be used to verify the appropriateness of the rehabilitation schemes to enhance the quality and frequency of the follow-up of the patient's progress. [The Advanced ICT Platform for Rehabilitation (ICT4Rehab) Project is a 3 years strategic platform project funded by IWOIB.The ICT4Rehab project addresses a new rehabilitation paradigm in the domain of muscle spasticity rehabilitation. It will require the integration of various data sources, the development of novel algorithms and putting into practice several ICT tools supporting 2D/3D user interaction. Two important aspects of clinical spasticity management will be combined: maintaining optimal patient motivation during her/his physical therapy data handling and processing of the spasticity-related clinical data.A portable rehabilitation system with local intelligence providing patient feedback and ensuring clinical relevance of the physical activity will be developed. Fusion of clinical data acquired in controlled (hospital) conditions with data coming from the personalised system will be used to verify the appropriateness of the rehabilitation schemes to enhance the quality and frequency of the follow-up of the patient's progress.]

"Unravelling the information processing patterns of SH2 domains participating in the JAK/STAT signaling pathway" [''Unravelling the information processing patterns of SH2 domains participating in the JAK/STAT signaling pathway'']
This project aims to provide the structural basis for the different allosteric properties of several SH2 domains, which participate in the common JAK-STAT signaling pathway. On a broad level, we aim to grasp the functional relevance, evolutionary importance and the effect of artificial or disease-related mutations in this class of SH2-containing proteins. To achieve these general goals, we focus here on the three SH2 domains belonging to the proteins SOCS3 and SHP2, which are both implicated in the attenuation of the JAK/STAT pathway. After the in silico identification of those residues implicated in the long-range communication through the SH2 structures, these dynamics will be validated using NMR relaxation experiments and functional assays in living cells. The residues implicated in signal transmission are expected to correspond to a particular sequence pattern, whose conservation within the family of SH2 domains will be evaluated. These new insights will provide a novel perspective on the existing classifications of this family and, in addition, a validation of alternative methods to multiple sequence alignments to predict the allosteric patterns within that family. Finally, the knowledge derived from the previous steps will be used to design an algorithm to create artificial members of the SH2 family with communication patterns similar to the one(s) previously analyzed. These artificial versions will be examined for folding and binding and their dynamics will be compared with their natural counterparts using NMR. Once they behave correctly, chimeras of the SOCS3 and SHP2 proteins will be tested within living cells. Overall, we aim not only to provide insight into the allosteric nature of SH2 domains, but also to decipher the intimate mechanism of SOCS3 and SHP2 function in modulation of JAK/STAT signaling. [This project aims to provide the structural basis for the different allosteric properties of several SH2 domains, which participate in the common JAK-STAT signaling pathway. On a broad level, we aim to grasp the functional relevance, evolutionary importance and the effect of artificial or disease-related mutations in this class of SH2-containing proteins. To achieve these general goals, we focus here on the three SH2 domains belonging to the proteins SOCS3 and SHP2, which are both implicated in the attenuation of the JAK/STAT pathway. After the in silico identification of those residues implicated in the long-range communication through the SH2 structures, these dynamics will be validated using NMR relaxation experiments and functional assays in living cells. The residues implicated in signal transmission are expected to correspond to a particular sequence pattern, whose conservation within the family of SH2 domains will be evaluated. These new insights will provide a novel perspective on the existing classifications of this family and, in addition, a validation of alternative methods to multiple sequence alignments to predict the allosteric patterns within that family. Finally, the knowledge derived from the previous steps will be used to design an algorithm to create artificial members of the SH2 family with communication patterns similar to the one(s) previously analyzed. These artificial versions will be examined for folding and binding and their dynamics will be compared with their natural counterparts using NMR. Once they behave correctly, chimeras of the SOCS3 and SHP2 proteins will be tested within living cells. Overall, we aim not only to provide insight into the allosteric nature of SH2 domains, but also to decipher the intimate mechanism of SOCS3 and SHP2 function in modulation of JAK/STAT signaling. ]

"Discovery of the molecular pathways regulating pancreatic beta cell dysfunction and apoptosis in diabetes using functional genomics and bioinformatics" [''Discovery of the molecular pathways regulating pancreatic beta cell dysfunction and apoptosis in diabetes using functional genomics and bioinformatics'']
The two main forms of diabetes mellitus are type 1 and type 2 diabetes (T1D and T2D). They affect 30 million individuals in Europe, decreasing their life quality and expectancy. Of particular concern is that the prevalence of both forms of diabetes is increasing; it is expected to double in the next two decades. A reduction in functional pancreatic beta cell mass, caused by progressive loss of beta cell function and increased apoptosis, is a key component of both T1D and T2D. The molecular mechanisms underlying this decreased functional beta cell mass remain to be clarified. Molecular signaling in the beta cells is decisive for their survival or death in diabetes. We hypothesize that crosstalk between key gene networks and insufficient protective responses, due to inherent features of beta cells, trigger dysfunction and the apoptosis program. This crosstalk is modulated by the genetic background of the individuals at risk, but it is presently unknown how candidate genes for diabetes affect beta cell function and survival, and how they interact with environmental agents that may trigger disease, e.g. glucolipotoxicity in T2D and viral infections in T1D. Against this background, the aim of the present proposal is to utilize functional genomics and advanced molecular biology and bioinformatics tools to identify molecular signatures and pathways responsible for beta cell dysfunction and apoptosis in diabetes, and to use this knowledge to define novel targets for intervention to preserve beta cell mass. [The two main forms of diabetes mellitus are type 1 and type 2 diabetes (T1D and T2D). They affect 30 million individuals in Europe, decreasing their life quality and expectancy. Of particular concern is that the prevalence of both forms of diabetes is increasing; it is expected to double in the next two decades. A reduction in functional pancreatic beta cell mass, caused by progressive loss of beta cell function and increased apoptosis, is a key component of both T1D and T2D. The molecular mechanisms underlying this decreased functional beta cell mass remain to be clarified. Molecular signaling in the beta cells is decisive for their survival or death in diabetes. We hypothesize that crosstalk between key gene networks and insufficient protective responses, due to inherent features of beta cells, trigger dysfunction and the apoptosis program. This crosstalk is modulated by the genetic background of the individuals at risk, but it is presently unknown how candidate genes for diabetes affect beta cell function and survival, and how they interact with environmental agents that may trigger disease, e.g. glucolipotoxicity in T2D and viral infections in T1D. Against this background, the aim of the present proposal is to utilize functional genomics and advanced molecular biology and bioinformatics tools to identify molecular signatures and pathways responsible for beta cell dysfunction and apoptosis in diabetes, and to use this knowledge to define novel targets for intervention to preserve beta cell mass.]

Preference dynamics in adaptive networks [Preference dynamics in adaptive networks]
Many strategic and economic situations are characterized by participant preferences that take into account the well-being of others. These preferences also guide humans in their choices with whom to participate in economical or social activities. Little attention has been given to how this group formation shapes these preferences and, vice versa, how these preferences shape the network of strategic interactions. Using methods of experimental economics and computational modeling we here will examine the interplay between these two dynamics in multiplayer games in order to provide supported answers to the origin and evolution of other regarding preferences. [Many strategic and economic situations are characterized by participant preferences that take into account the well-being of others. These preferences also guide humans in their choices with whom to participate in economical or social activities. Little attention has been given to how this group formation shapes these preferences and, vice versa, how these preferences shape the network of strategic interactions. Using methods of experimental economics and computational modeling we here will examine the interplay between these two dynamics in multiplayer games in order to provide supported answers to the origin and evolution of other regarding preferences. ]

Network dynamics of social capital [Network dynamics of social capital]
Social capital is always embedded in social networks. Such networks are inherently dynamic. Their dynamics is affected in various ways by endogenous and exogenous effects. The study of analogous dynamic networks is being carried out in various other fields, ranging from epidemics to telecommunications. With this cross-disciplinary project, we intend to study the interplay of social capital and network dynamics, using advanced modeling tools from research on complex networks to analyze and interpret available social network data. [Social capital is always embedded in social networks. Such networks are inherently dynamic. Their dynamics is affected in various ways by endogenous and exogenous effects. The study of analogous dynamic networks is being carried out in various other fields, ranging from epidemics to telecommunications. With this cross-disciplinary project, we intend to study the interplay of social capital and network dynamics, using advanced modeling tools from research on complex networks to analyze and interpret available social network data. ]

Métabolisme urbain [Water urbanism]
Les changements climatiques, la croissance urbaine mais aussi la crise économique augmente l'insécurité pour les agglomérations urbaines d'assurer le bon fonctionnement du cycle de l'eau dans l'avenir: l'alimentation en eau potable, le stockage d'eau pluviale, le traitement des eaux usées, etc. En conséquence, ces difficultés ont influencé négativement la perception de l'eau dans l'aménagement du territoire, en étant plutôt considérée comme une cause de désastre. Cette perception a traditionnellement orienté vers des solutions qui visent la réalisation de grandes infrastructures pour résoudre temporellement les problèmes, mais qui amplifient la rupture entre la ville et l'environnement. Des nouveaux concepts comme la résilience, la sensibilité à l'eau ou le système intégré et décentralisé de la gestion de l'eau initient une nouvelle vision pour l'aménagement du territoire et proposent une réconciliation de l'eau avec la ville. Dans cette perspective, les architectes et les urbanistes ont un rôle important pour établir la connexion entre l'usage du territoire, l'urbanité et les aspirations des citoyens face à l'eau. S'inscrivant dans cette tendance globale, le centre de recherche LoUIsE explore ce questionnement de la gestion des risques naturels et de la résilience dans les processus de reconstruction de la ville, notamment pendant le workshop Resilient Ishinomaki au Japon. En 2013, la gestion de l'eau a ensuite pris forme comme nouvel axe de recherche dans le cadre de LoUIsE, envisageant des recherches théoriques, mais également prospectives, sur l'aménagement de la Région de Bruxelles Capitale. Ces initiatives s'inscrivent dans une nouvelle vision, réconciliant la ville avec l'eau, et visent à la sensibilisation des étudiants, des architectes et des urbanistes dans cette direction. Deux nouveaux chercheurs ont ainsi joint le centre. Marco Ranzato, post-doctorant avec une bourse Prospective Research for Brussels, s'intéresse aux méthodes décentralisées de gestion de l'eau dans la ville tandis que Catalina Dobre, doctorante, oriente sa recherche vers la gestion de l'eau de pluie et la manière qui permettrait à Bruxelles de devenir plus « sensible » à l'eau. [Nowadays, enterprises and public institutions have to face a growing presence of frauds and consequently need automatic systems able to support fraud detection and fight. These systems are essential since it is not always possible or easy for a human analyst to detect fraudulent patterns in transaction datasets, often characterized by a large number of samples, many dimensions and online update.]



theses


Liran Lerman. A Machine Learning Approach for Automatic and Generic Side-Channel Attacks, 2015

Abhilash Miranda. Spectral Factor Model for Times Series Learning, 2011

Benjamin Haibe-Kains. Identification and Assessment of Gene Signatures in Human Breast Cancer., 2009

Olivier Caelen. Sélection Séquentielle en Environnement Aléatoire Appliquée à l'Apprentissage Supervisé, 2009

Yann-ael Le Borgne. Learning in Wireless Sensor Networks for Energy-Efficient Environmental Monitoring, 2009

Kevin Kontos. Gaussian Graphical Model Selection for Gene Regulatory Networké Reverse Engineering and Function Prediction, 2009

Patrick E. Meyer Information. Theoretic Variable Selection and Network Inference from Microarray Data., 2008



prix / awards


Prix De Meurs-François Prize pour la thèse de doctorat, 2009. - Yann-Aël LE BORGNE

Award Solvay pour le mémoire - Souhaib BEN TAIEB

Award Solvay pour la thèse de doctorat - Benjamin HAIBE KAINS



savoir-faire/équipements / know-how, equipment


Laboratoire avec capteurs sans-fil



mots clés pour non-spécialistes / keywords for non-specialists


analyse des données apprentissage automatique bioinformatique biologie numérique la théorie des jeux


disciplines et mots clés / disciplines and keywords


Aménagement urbain Informatique générale

fraud detection muscle spasticity rehabilitation network dynamics pancreatic beta cell dysfunction preference dynamics sh2 domains


codes technologiques DGTRE


Bio-informatique, informatique médicale, biométrie Intelligence artificielle Sciences de l'ordinateur, analyse numérique, systèmes, contrôle