MODAL Research team
MOdel for Data Analysis and Learning
- Leader : Christophe Biernacki
- Type : Project team
- Research center(s) : Lille
- Field : Applied Mathematics, Computation and Simulation
- Theme : Optimization, machine learning and statistical methods
- Partner(s) : CNRS,Université des sciences et technologies de Lille (Lille 1),Université Lille 2
- Collaborator(s) : Laboratoire Paul Painlevé (LPP) (UMR8524)
Team presentationThe main focus of MODAL is to design generative models dealing with complex multivariate and/or heterogeneous data. Typical instances of such data are
- nominal covariables for the multivariate setting,
- and the combination of continuous and nominal variables for the heterogeneous setting.
Obviously, other widespread complex covariables are of interest such as ordinal, ranks, and intervals data.
From these generative models, a convenient and efficient statistical analysis remains to be carried out, leading to data analysis (visualization, clustering) and data learning (supervised and semi-supervised classification, density estimation).
Research themesMODAL is focused on generative models, that is models describing the generation process of data, unlike predictive models.
Generative models are of great interest. On the one hand, they are required in several statistical objectives such as clustering, semi-supervised classification, and density estimation, where predictive models are useless. On the other hand, these models enable data visualization. Indeed, they provide a full description of the data distribution, which gives access to several aspects of the data such as high density areas for instance.
In supervised classification, generative and predictive models directly compete with one another. However, the lack of flexibility of the generative approach, as opposed to the predictive one, is completely balanced by the use of model selection.
In addition, among generative approaches, parametric ones such as mixture models are preferred. Provided parameters are meaningful and parsimonious, mixture models allow valuable data interpretation.
International and industrial relationsCurrent collaborations
- PGXIS UK, PharmacoGenomic Innovative Solutions
- Institut Pasteur de Paris
- IBL, Institut Biologique de Lille
Research teams of the same theme :
- BONUS - Big Optimization aNd Ultra-Scale Computing
- CELESTE - mathematical statistics and learning
- GEOSTAT - Geometry and Statistics in acquisition data
- INOCS - INtegrated Optimization with Complex Structure
- MISTIS - Modelling and Inference of Complex and Structured Stochastic Systems
- RANDOPT - Randomized Optimization
- REALOPT - Reformulations based algorithms for Combinatorial Optimization
- SELECT-POST - Model selection in statistical learning
- SEQUEL - Sequential Learning
- SIERRA - Statistical Machine Learning and Parsimony
- TAU - TAckling the Underspecified