报告地点：腾讯会议：711 873 080
报告题目：Controlling the False Discovery Rate in Structural Sparsity: Split Knockoffs
报告摘要：Controlling the False Discovery Rate in variable selection is critical for reproducible discoveries, which receives extensive studies in sparse linear models. However, in many scenarios such as medical image analysis, partial order ranking, and trend filtering etc., sparsity constraint is not directly imposed on parameters, but on alinear transformation of the parameters to be estimated. In this talk, we present a new data adaptive variable selection framework with FDR control in this structural sparsity setting, the Split Knockoff. This proposed method adopts the variable splitting scheme from optimisation, which relaxes the linear subspace constraint on the parameters to its neighbourhood. Such a scheme will yield orthogonal design matrices and enjoy new statistical benefits in terms of FDR control and power. In particular, two variants of Split Knockoffs will be introduced to overcome the challenge of broken symmetry in exchangeability and achieve provable FDR control.
报告人简介：Prof. Yuan Yao is currentlyAssociate Professor of Mathematics and Chemical & Biological Engineering atHong Kong University of Science and Technology. He received Ph.D. in Mathematics from UC Berkeley and has been working on mathematics of machine learning and data science.