Séminaires


Retour à la liste de tous les séminaires


Learning sparse penalties for change-point detection using max margin interval regression.

Le : 15/10/2012 11h00
Par : Toby Dylan HOCKING (ENS, INRIA)
Lieu : I103
Lien web :
Résumé : Many statistical models have hyperparameters that must be chosen using cross-validation or various heuristics. In segmentation models, the number of segments is usually chosen using regularized cost functions that compromise between data fitting and model complexity. There are many penalties and many heuristics for choosing the constants in those penalties. In this work, we propose to learn the penalty and its constants in databases of signals with weak change-point annotations. We propose a convex relaxation that yields an interval regression problem, and solve it using accelated proximal gradient methods. Finally, we show that this method achieves state-of-the-art performance on a large database of annotated copy number profiles from neuroblastoma tumors.