Threshold selection for model selection

Jeudi 10 décembre 2015 14:00-15:00 - Sylvain Sardy - Université de Genève

Résumé : Model selection is a recurrent theme in Statistics and is becoming more challenging with big data. Sparse model selection assumes a low dimensional model exists underneath this avalanche of data. The goal is to identify it. To that aim, many thresholding techniques have been proposed. They all require the challenging selection of a scalar $\lambda$ called threshold. We propose a unified framework to select $\lambda$ based on the concept of a zero-thresholding function. The work is motivated by two collaborations with scientists from the University of Geneva in cancer research and cosmology. Based on

Lieu : Salle 117-119

Threshold selection for model selection  Version PDF