A0278
Title: High-dimensional convex nonparametric least squares with Lasso penalty
Authors: Zhiqiang Liao - Aalto University (Finland) [presenting]
Abstract: The problem of variable selection is studied in convex nonparametric least squares. Whereas the Lasso is a popular technique for simultaneous estimation and variable selection, its performance is unknown in convex regression problems. The performance of the Lasso regularized convex nonparametric least squares estimator is investigated in a high-dimensional setting, and an alternative approach is proposed based on the unique structure of the subgradients. The proposed estimators perform favorably, while generally leading to sparser models relative to the other predictive models via the standard Lasso. Further, the estimators can be expressed as solutions to convex optimization problems and are amenable to modern optimization algorithms.
