EcoSta 2018: Registration
View Submission - EcoSta2018
A0629
Title: Uniform knockoff filter for high-dimensional controlled graph recovery Authors:  Jia Zhou - University of Science and Technology of China (China) [presenting]
Zemin Zheng - University of Science and Technology of China (China)
Abstract: Learning the dependence structures in high-dimensional graphical models is of fundamental importance in many contemporary applications. Despite the fast growing literature, procedures with both guaranteed false discovery rate (FDR) control and high power for recovering the graphical structures remain largely unexplored. We develop a new method called uniform knockoff filter that controls the overall FDR in graph recovery based on control variables. Instead of controlling the FDR in a nodewise way, the proposed procedure utilizes a uniform threshold for the statistics based on a large-scale mixture of regression models associated with the graph, which enjoys not only theoretical guarantees of FDR control but also significantly higher power. Furthermore, a scalable implementation approach is developed for the uniform knockoff filter such that all control variables can be generated through a single estimation of the overall graphical structure. Numerical studies verify that our method outperforms existing approaches in power with FDR control.