BEGIN:VCALENDAR
PRODID:-//eluceo/ical//2.0/EN
VERSION:2.0
CALSCALE:GREGORIAN
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1656
DTSTAMP:20251215T055054Z
SUMMARY:Sparsifying suprema of Gaussian processes
DESCRIPTION:Speaker: Anindya De (University of Pennsylvania)\n\nAbstract: \
 nWe show that the supremum of any centered Gaussian processcan be approxim
 ated to any arbitrary accuracy by a finite dimensional Gaussian process\,
  where the dimension of the approximator is justdependent on the target er
 ror. As a corollary\, we show that for any norm \\Phi defined over R^n an
 d target error \\eps\, there is a norm \\Psi such that (i) \\Psi is only 
 dependent on t(\\eps) = \\exp (\\exp(poly(1/\\eps))) dimensions and (ii) \
 \Psi(x)/\\Phi(x) \\in [1-\\eps\, 1+\\eps] with probability 1-\\eps (when x
  is sampled from the Gaussianspace). We prove a similar-in-spirit result f
 or sparsifying high-dimensional polytopes in Gaussian space\, and present
  applications to computational learning and property testing. Our proof r
 elies on Talagrand's majorizing measures theorem.\nJoint work with Shivam
  Nadimpalli\, Ryan O'Donnell and Rocco Servedio.\n
URL:https://www.tcs.tifr.res.in/web/events/1656
DTSTART;TZID=Asia/Kolkata:20251216T160000
DTEND;TZID=Asia/Kolkata:20251216T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
END:VCALENDAR