Robust Group Lasso

• Activity Number: 41
• Type: Contributed
• Date/Time: Sunday, August 1, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Ji Young Kim*+
• Companies: Mount Holyoke College
• Keywords: Variable Selection ; Lasso ; Group Lasso ; Robust ; Multivariate ; Motifs

Abstract

Variable selection has received much attention from researchers in many areas. A number of methods have been developed including Lasso (Tibshirani, 1996), penalized least squares (Fan and Li, 2001) and LARS (Efron et al., 2004). The group Lasso in Yuan and Lin (2006) is an extension of Lasso with the goal of selecting important groups of variables rather than individual variables. In the cases where the errors have heavier tails than Gaussian distributions, the least squares based methods may not work so well. We propose a robust Lasso method for the multivariate time-course response, and develop an algorithm to compute it. The proposed method enables us to handle multivariate responses in variable selection by grouping. A basis representation of the regression parameters is employed to reduce dimensionality. We apply the proposed method to Saccharomyces Cerevisiae to identify motifs.