Abstract:
With the rapid development of modern science and technology, the era of big data is coming to us. At this time, the reproducibility of statistical methods is pivotal for improving rigor in scientific research. The knockoff procedure proposed in Barber and Candès
48 is a general variable selection algorithm that can leverage any feature importance score to discover true effects while rigorously controlling false discovery rate (FDR). The main idea is to construct synthetic variables called knockoffs to mimic the correlation structure found within the original variables. This method has received much attention in recent years because it completely bypasses the computation of
p-values, and has become the most popular research area in statistics and machine learning. This paper mainly introduces the newly research advance in knockoff procedure and briefly discusses some future directions.