Séminaire (organisé par l’équipe de recherche DI)

Maxime Sangnier

Chercheur post-doctoral, Laboratoire Traitement et Communication de l’Information (LTCI), Telecom ParisTech


Joint quantile regression in vector-valued RKHSs


Mardi 29 mars 2016 à 14 h en GI016 (Bâtiment Blaise Pascal)

Résumé :

Building upon kernel-based multi-task learning, a novel methodology for estimating and predicting simultaneously several conditional quantiles is proposed. We particularly focus on curbing the embarrassing phenomenon of quantile crossing. Moreover, this framework comes along with a uniform convergence bound and an efficient coordinate descent learning algorithm. Numerical experiments on benchmark datasets highlight the enhancements of our approach regarding the prediction error, the crossing occurrences and the training time.



Actualités
Vidéothèque
Téléchargements
Annuaire



FR SHIC 3272

Collegium UTC/CNRS