BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Resampling-Based Inference in High Dimensional Linear Models via F
 lipscore Tests
DTSTART:20260421T091500Z
DTEND:20260421T101500Z
DTSTAMP:20260607T052300Z
UID:indico-event-14478@indico.math.cnrs.fr
DESCRIPTION:Speakers: Daniela Corbetta (IMT)\n\nWe consider inference in h
 igh-dimensional linear models where p>>n\, focusing on testing individual 
 regression coefficients and constructing confidence statements on the set 
 of active predictors. Existing methods such as debiased lasso and ridge pr
 ojection rely on approximate inversion of X’X and yield asymptotically v
 alid inference\, but may suffer from instability and loss of accuracy in f
 inite samples\, particularly under strong dependence.We study an alternati
 ve approach based on the flipscore test\, a resampling-based procedure tha
 t approximates the null distribution of score statistics via random sign-f
 lipping. This method avoids explicit matrix inversion\, is robust to varia
 nce misspecification\, and accommodates dependence among predictors. Howev
 er\, its standard formulation requires a full-rank design and is not direc
 tly applicable when p>n.To address this limitation\, we propose a framewor
 k combining conditional resampling with preliminary variable selection. We
  investigate several strategies\, including lasso-based screening\, SVD-ba
 sed dimension reduction\, and stepwise selection\, and analyze their impac
 t on validity. Under suitable conditions on the selected model\, we establ
 ish asymptotic validity of the resulting tests.We further extend the appro
 ach to simultaneous inference on the full coefficient vector via multiple 
 testing procedures\, and exploit the resampling structure to obtain post-h
 oc bounds on the false discovery proportion that adapt to dependence. Simu
 lation results highlight competitive error control and power compared to s
 tate-of-the-art methods. \n\nhttps://indico.math.cnrs.fr/event/14478/
LOCATION:Salle E. Picard (1R2-129)
URL:https://indico.math.cnrs.fr/event/14478/
END:VEVENT
END:VCALENDAR