A0858
Title: An efficient framework for model-free variable selection
Authors: Xin He - City University of Hong Kong (Hong Kong) [presenting]
Junhui Wang - Chinese University of Hong Kong (Hong Kong)
Abstract: Variable selection plays a crucial role in the analysis of high dimensional datasets. While dealing with an ultra-high dimensional dataset, most existing methods require different model assumptions. We propose a model-free variable selection method which can exactly identify informative variables that are related to the conditional mean function by measuring the corresponding gradients. More importantly, this method allows the number of predictors grows exponentially with the sample size, which is opposed to most existing model-free methods which can only handle finite number of predictors. Some applications of the proposed method under various model assumptions are also considered in this paper. The proposed method is implemented via an efficient and fast computing algorithm. The asymptotic estimation and variable selection consistencies are established in the model-free framework, which assures that the truly informative variables are correctly identified with high probability. The effectiveness of the proposed method is also supported by a variety of simulated and real-life examples.
