Export Reviews, Discussions, Author Feedback and Meta-Reviews

The authors propose a two-stage feature selection approach for linear regression. Candidate features are selected in the first stage and the selection is further refined in the second stage. They define "Strong screening consistency" as a criterion for performing the initial stage, and then investigate the implications of that definition when the screening values are computed by a linear operator. From the content and the references it appears that the authors address the statistics community. My expertise is more in machine learning/computer science, and my review is written from that perspective. My main point is the following: The vector beta to be estimated has 0 value for its ith coordinate iff the ith feature should not be part of the optimal selection.