Contribution à l'apprentissage statistique à base de modèles génératifs pour données complexes.

Abstract : This manuscript presents my research activities, which mainly focus on designing parametric, parsimonious and meaningful generative models for complex data. Several kinds of complex data have been studied. Data sampled from different populations (transfer learning) has been addressed by designing parametric models for the link between the different populations. Thus, statistical models can be adapted from one population to another one by sparing a large collect of new data. Ranking data, which results from ranking of objects by a judge according to a preference order, ordinal data, which are categorical data with ordered categories, and functional data, in which the statistical unit consists of one or several curves, have also been studied. For this three kinds of complex data, generative models have been developed and used for the clustering of multidimensional data. The last kind of complex data, high dimensional data, has been studied in a regression context. In this domain, two approaches are proposed by two Ph.D. students I co-supervise\string: the first one uses combinatorial optimization algorithms in order to efficiently explore the feature space and the second one defines a regression model in which the variables having a similar effect on the output are grouped together.
Document type :
Habilitation à diriger des recherches
Statistics [math.ST]. Université des Sciences et Technologie de Lille - Lille I, 2012


https://tel.archives-ouvertes.fr/tel-00761184
Contributor : Julien Jacques <>
Submitted on : Wednesday, December 5, 2012 - 9:52:48 AM
Last modification on : Wednesday, December 5, 2012 - 10:23:17 AM

Identifiers

  • HAL Id : tel-00761184, version 1

Collections

Citation

Julien Jacques. Contribution à l'apprentissage statistique à base de modèles génératifs pour données complexes.. Statistics [math.ST]. Université des Sciences et Technologie de Lille - Lille I, 2012. <tel-00761184>

Export

Share

Metrics

Consultation de
la notice

338

Téléchargement du document

279