Forum de Statistiques
Vous souhaitez réagir à ce message ? Créez un compte en quelques clics ou connectez-vous pour continuer.
Les posteurs les plus actifs de la semaine
Aucun utilisateur

Le Deal du moment : -34%
-34% LG OLED55B3 – TV OLED 4K 55″ 2023 ...
Voir le deal
919 €

CDD - Paris

Aller en bas

CDD - Paris Empty CDD - Paris

Message par joyeux_lapin13 Jeu 6 Fév 2014 - 6:35


Type de poste:
Post-doc / IR
CDD
Durée du poste:
2 years
Ville:
Paris
Laboratoire:
UMR518 AgroParisTech/INRA
Nom du contact:
Tristan Mart-Huard
Julien Chiquet
Email du contact:
tristan.mary-huard@agroparistech.fr
julien.chiquet@agroparistech.fr
Url:
http://www.agroparistech.fr/mmip/maths/essaimia/accueil:stages
Date de validité:
Sa, 20/12/2014
Description du poste:

Regularization and sparse methods has been intensively and successfully used in bioinformatics to prediction problems such as cancer diagnosis or physiological response prediction. In most of these applications, the response variable is univariate, or different response variables are investigated one at a time, neglecting the possible strong correlations between the responses. The application of regularization methods to multivariate regression turns out to be non trivial, and raises important methodological questions. In this context, the post-doc fellow's research will focus on the following questions:

- How to incorporate prior knowledge in the optimization process, first in order to narrow the domain where optimization is performed and reduce the computational burden, and then to obtain interpretable prediction functions?

- How to handle correlations between the different traits/responses? How to impose constraints both on the covariance matrix of the response variables and on the norm of the regression coefficients?

- How do the state-of-the-art optimization algorithms compare in the context of multivariate regression? How to obtain efficient calibration of the trade-off between adjustment and smoothness when adjustment is measured on several response variables?

- How to generalize the regularized multivariate regression approach in the context of multi-task learning, where the learning task has to be jointly performed on several inhomogeneous training populations ?

The developed methodology will be applied to genomic selection, where the goal is to predict the performance of individuals according to their available genotypic information (molecular markers). Animal and plant genetic datasets will be made available from our biological partners.

IMPORTANT: The candidate should be eligible according to the Agreenskills excellence program criteria.
joyeux_lapin13
joyeux_lapin13

Nombre de messages : 1927
Age : 40
Localisation : Mayotte
Date d'inscription : 21/04/2010

https://lemakistatheux.wordpress.com/

Revenir en haut Aller en bas

Revenir en haut

- Sujets similaires

 
Permission de ce forum:
Vous ne pouvez pas répondre aux sujets dans ce forum