Controlling The False Discovery Rate For Latent Factors Via Unit-Rank Deflation

While sparse factor regression is often encountered under high dimensions, it still is unclear how to control the false discovery rate (FDR) of latent factors. In this paper, we propose a variable selection procedure to address the issue and prove that the FDR can be asymptotically controlled at a target level. Moreover, our approach is scalable and memory-efficient in practice owing to the divide-and-conquer strategy. (C) 2021 Elsevier B.V. All rights reserved.