Keywords: Changepoint detection, Subsampling, Knockoffs
While changepoint detection algorithms are well studied in the literature, finite-sample inferential guarantees for changepoint detection carries a sparser literature. In this work, we combine two modern themes in statistical inference and variable selection -- subsampling and knockoff variable selection. Subsampling methods are a popular and flexible class of computational techniques for enabling nonparametric statistical inference under minimal assumptions. Knockoff variable selection is a new class of methods that allow for variable selection in the regression setting, which allows for finite sample false discovery rate control. A careful fusion of these two techniques may produce a tool for changepoint detection with useful statistical guarantees. One goal would be to control in false discoveries among detected changepoints in 1D or time series data after having applied greedy changepoint detection algorithms. We demonstrate the usage of our proposed tools in simulation and real examples, as well as other useful extensions.