BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Talks.cam//talks.cam.ac.uk// X-WR-CALNAME:Talks.cam BEGIN:VEVENT SUMMARY:BSU Seminar: "\;Harnessing Extra Randomness: Replicability\, F lexibility and Causality"\; - Richard Guo\, Stas Lab\, University of C ambridge DTSTART:20231114T140000Z DTEND:20231114T150000Z UID:TALK207550@talks.cam.ac.uk CONTACT:Alison Quenault DESCRIPTION:Many modern statistical procedures are randomized in the sense that the output is a random function of data. For example\, many procedur es employ data splitting\, which randomly divides the dataset into disjoin t parts for separate purposes. Despite their flexibility and popularity\, data splitting and other constructions of randomized procedures have obvio us drawbacks. First\, two analyses of the same dataset may lead to differe nt results due to the extra randomness introduced. Second\, randomized pro cedures typically lose statistical power because the entire sample is not fully utilized.\n\nTo address these drawbacks\, in this talk\, I will stud y how to properly combine the results from multiple realizations (such as through multiple data splits) of a randomized procedure. I will introduce rank-transformed subsampling as a general method for delivering large samp le inference of the combined result under minimal assumptions. I will illu strate the method with three applications: (1) a “hunt-and-test” proce dure for detecting cancer subtypes using high-dimensional gene expression data\, (2) testing the hypothesis of no direct effect in a sequentially ra ndomized trial and (3) calibrating cross-fit “double machine learning” confidence intervals. For these problems\, our method is able to derandom ize and improve power. Moreover\, in contrast to existing approaches for c ombining p-values\, our method enjoys type-I error control that asymptotic ally approaches the nominal level. This new development opens up the possi bility of designing procedures that explicitly randomize and derandomize: extra randomness is introduced to make the problem easier before being mar ginalized out.\n\nThis talk is based on joint work with Rajen Shah. \n\nB io: Richard Guo is a research associate in the Statistical Laboratory at t he University of Cambridge\, mentored by Prof. Rajen Shah. Previously\, he was the Richard M. Karp Research Fellow in the 2022 causality program at the Simons Institute for the Theory of Computing. He received his PhD in S tatistics from University of Washington in 2021\, advised by Thomas Richar dson. His research interests include graphical models\, causal inference\, semiparametric methods and replicability of data analysis. Dr. Guo will s tart as an assistant professor in Biostatistics at University of Washingto n in January 2024.\n LOCATION:MRC Biostatistics Unit\, East Forvie Building\, Forvie Site Robin son Way Cambridge CB2 0SR. END:VEVENT END:VCALENDAR