BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Stability - Bin Yu\, University of California\, Berkeley
DTSTART:20121126T160000Z
DTEND:20121126T170000Z
UID:TALK40802@talks.cam.ac.uk
CONTACT:Richard Samworth
DESCRIPTION:Reproducibility is imperative for any scientific discovery. Of
 ten than not\,\nmodern scientific findings rely on statistical analysis of
  high-dimensional\ndata. At a minimum\, reproducibility manifests itself i
 n stability of\nstatistical results relative to “reasonable” perturbat
 ions to data and to the\nmodel used. Jacknife\, bootstrap\, and cross-vali
 dation are based on\nperturbations to data\, while robust statistics metho
 ds deal with perturbations\nto models.\n\nIn this talk\, a case is made fo
 r the importance of stability in\nstatistics. Firstly\, we motivate the ne
 cessity of stability of interpretable\nencoding\nmodels for movie reconstr
 uction from brain fMRI signals. Secondly\, we find\nstrong evidence in the
  literature to demonstrate the central role of stability\nin statistical i
 nference. Thirdly\, a smoothing parameter selector based on\nestimation st
 ability (ES)\, ES-CV\, is proposed for Lasso\, in order to bring\nstabilit
 y to bear on cross-validation (CV). ES-CV is then utilized in the\nencodin
 g models to reduce the number of predictors by 60% with almost no loss\n(1
 .3%) of prediction performane across over 2\,000 voxels. Last\, a novel\n
 “stability” argument is seen to drive new results that shed light on t
 he\nintriquing interactions between sample to sample varibility and heavie
 r tail\nerror distribution (e.g. double-exponential) in high dimensional r
 egression\nmodels with p predictors and n independent samples. In particul
 ar\, when p/n →\nκ ∈ (0.3\, 1) and error is double-exponential\, OLS 
 is a better estimator than\nLAD.
LOCATION:MR12\, CMS\, Wilberforce Road\, Cambridge\, CB3 0WB
END:VEVENT
END:VCALENDAR