Since the time of Gauss, it has been generally accepted that $\ell_2$-methods of combining observations by minimizing sums of squared errors have significant computational advantages over earlier $\ell_1$-methods based on minimization of absolute errors advocated by Boscovich, Laplace and others. However, $\ell_1$-methods are known to have significant robustness advantages over $\ell_2$-methods in many applications, and related quantile regression methods provide a useful, complementary approach to classical least-squares estimation of statistical models. Combining recent advances in interior point methods for solving linear programs with a new statistical preprocessing approach for $\ell_1$-type problems, we obtain a 10- to 100-fold improvement in computational speeds over current (simplex-based) $\ell_1$-algorithms in large problems, demonstrating that $\ell_1$-methods can be made competitive with $\ell_2$-methods in terms of computational speed throughout the entire range of problem sizes. Formal complexity results suggest that $\ell_1$-regression can be made faster than least-squares regression for $n$ sufficiently large and $p$ modest.
The aim of Statistical Science is to present the full range of contemporary statistical thought at a technical level accessible to the broad community of practitioners, teachers, researchers, and students of statistics and probability. The journal publishes discussions of methodological and theoretical topics of current interest and importance, surveys of substantive research areas with promising statistical applications, comprehensive book reviews, discussions of classic articles from statistical literature, and interviews with distinguished statisticians and probabilists.
The purpose of the Institute of Mathematical Statistics (IMS) is to foster the development and dissemination of the theory and applications of statistics and probability. The Institute was formed at a meeting of interested persons on September 12, 1935, in Ann Arbor, Michigan, as a consequence of the feeling that the theory of statistics would be advanced by the formation of an organization of those persons especially interested in the mathematical aspects of the subject. The Annals of Statistics and The Annals of Probability (which supersede The Annals of Mathematical Statistics), Statistical Science, and The Annals of Applied Probability are the scientific journals of the Institute. These and The IMS Bulletin comprise the official journals of the Institute. The Institute has individual membership and organizational membership. Dues are paid annually and include a subscription to the newsletter of the organization, The IMS Bulletin. Members also receive priority pricing on all other IMS publications.
This item is part of JSTOR collection
For terms and use, please refer to our Terms and Conditions
Statistical Science
© 1997 Institute of Mathematical Statistics
Request Permissions
