Executive Summary : | With the rapid development of modern data collection methods, a vast amount of data with high-dimensional features become increasingly popular in scientific areas such as biology, finance and computer science. This necessitates an advancement of classical statistical methods to handle ultrahigh-dimensionality present in such data. In recent times, considerable attention has been devoted to variable selection and feature screening. The broad goal of this project is to explore two research areas: (i) fast tests for high-dimensional means under sparsity applicable to multiple class problems, and (ii) model-free sure screening methods for classification based on joint variables. |