BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Certified Randomized Methods for High-Dimensional Approximation: N
 on-asymptotic bounds for the PCA/POD error and near-optimal sampling strat
 egies
DTSTART:20261125T161500
DTSTAMP:20261122T092244Z
UID:bcab048f0de5aa6e8660dd788837d23deb24ef68f9b0209efaa2ce91
CATEGORIES:Conferences - Seminars
DESCRIPTION:Kathrin Smetana (EPFL)\nIn this talk\, I will present randomiz
 ed methods that provide a certified approximation of the solutions of para
 metric partial differential equations (PDEs) with high-dimensional paramet
 er sets. To that end\, we will exploit data-dependent sampling strategies 
 and concentration phenomena to discover and leverage low-dimensional struc
 ture.\nFirst\, I will present non-asymptotic error bounds for Principal Co
 mponent Analysis (PCA) or Proper Orthogonal Decomposition (POD) under the 
 assumption that the norm of the solution is bounded for almost all paramet
 er values. The resulting bounds are independent of the dimension of the pa
 rameter space and remain valid in infinite-dimensional settings. Under sui
 table spectral-gap conditions\, we show that our bounds are asymptotically
  optimal up to constants.\nWhile the PCA/POD result holds for general samp
 ling distributions\, the quality of the randomized approximation can be si
 gnificantly improved by exploiting problem-dependent sampling. We therefor
 e turn to provably near-optimal sampling strategies. Sampling from Christo
 ffel-type measures provides a near-optimal strategy for identifying large 
 function values from finitely many samples. Building on this idea\, we des
 ign a randomized greedy algorithm that guarantees\, with high probability\
 , the desired approximation accuracy for almost every parameter value in t
 he parameter set. We show that\, with high probability\, the resulting ran
 domized greedy algorithm achieves quasi-optimal convergence rates. If we h
 ad already found a space that approximates the set of all solutions well\,
  volume sampling based on that space would yield a quasi-best approximatio
 n in expectation with the number of samples on the order of the dimension 
 of the reduced space. However\, in this case\, we would no longer need to 
 sample\, as we would already have a good approximation space. We therefore
  propose constructing a surrogate sampling distribution\, for instance\, f
 rom the data functions of the PDE\, and analyze how this affects the appro
 ximation error.
LOCATION:CM 1 517
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR
