November 3, 2016
All Day
209 W. Eighteenth Ave. (EA), Room 170
Title
CoCoLasso for High-Dimensional Error-in-Variables Regression
Speaker
Hui Zou, University of Minnesota
Abstract
Loh and Wainwright (2012, AOS) proposed a non-convex modification of the Lasso for doing high-dimensional regression with noisy and missing data. Their method requires three tuning parameters and two are difficult to set in practice. Moreover, it is generally agreed that the virtues of convexity contribute fundamentally the success and popularity of the Lasso. In light of this, we propose a new method named CoCoLasso that is convex and can handle a general class of corrupted datasets. CoCoLasso operates pretty much like the standard Lasso. Theoretical and numerical results are provided to support CoCoLasso. This is joint work with Abhirup Datta.