A1432
Title: Online testing with dependent streaming data
Authors: Seohwa Hwang - Seoul National University (Korea, South) [presenting]
Abstract: The problem of converting offline multiple testing procedures into valid online testing rules under sequential data acquisition is studied. While classical offline methods assume access to the full dataset and control error rates such as the false discovery rate (FDR) in a batch setting, directly applying them in an online environment leads to invalid inference. To address this gap, a principled offline-to-online conversion framework based on sliding-window inference is proposed. At each time step, the procedure constructs local statistics using a finite window of recent observations and applies a calibrated decision rule that mimics its offline counterpart while respecting the online filtration. Sufficient conditions are established under which the resulting procedure provably controls the FDR asymptotically, even in the presence of dependence.
