A0637
Title: Spatially adaptive false discovery rate thresholding for sparse estimation
Authors: Gourab Mukherjee - University of Southern California (United States) [presenting]
Wenguang Sun - University of Southern California (United States)
Jiajun Luo - Linkedin (United States)
Abstract: A new false discovery rate (FDR) is developed based threshold estimator by extending the elegant FDR based estimator to spatial settings. The idea is to first construct robust and structure-adaptive weights by estimating local sparsity levels, and thereafter to set spatially adaptive thresholds using the weighted Benjamini-Hochberg procedure. We present asymptotic results demonstrating the superior performance of the proposed method. Through numerical experiments we illustrate the importance of spatial adaptation by studying the finite sample performance of the proposed estimator.
