Variable Selection via Smoothing Spline Adaptive Response Transformation

• Activity Number: 505
• Type: Topic Contributed
• Date/Time: Wednesday, August 4, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Wenxuan Zhong*+
• Companies: University of Illinois
• Keywords: variable selection ; nonlinear regression ; smoothing spline ; regularization ; penalized least square ; lasso

Abstract

Nonlinear regression in high dimensional space suffers from the curse of dimensionality. Variable selection for nonlinear regression is very important but face many challenges:(1) The prior specification of the nonlinear link function is infeasible in the high dimensional regression. (2) Dimension reduction have been a major theme to overcome such shortcomings by treating the nonlinear link function as a nuisance meta parameter. However, dimension reduction methods typically assume some restrictive condition which is difficult if not impossible to verify in real practice. Moreover, they fail when the mean regression function is symmetric. In this talk, we present a smoothing spline adaptive response transformation method to surmount the challenges.