BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Stochastic gradient with least-squares control variates - Fabio No
 bile (EPFL - Ecole Polytechnique Fédérale de Lausanne)
DTSTART:20250703T093000Z
DTEND:20250703T103000Z
UID:TALK233665@talks.cam.ac.uk
DESCRIPTION:The stochastic gradient (SG) method is a widely used approach 
 for solving stochastic optimization problems\, but its convergence is typi
 cally slow. Existing variance reduction techniques\, such as SAGA\, improv
 e convergence by leveraging stored gradient information\; however\, they a
 re restricted to settings where the objective functional is a finite sum\,
  and their performance degrades when the number of terms in the sum is lar
 ge. In this work\, we propose a novel approach that is best suited when th
 e objective is given by an expectation over random variables with a contin
 uous probability distribution. Our method constructs a control variate by 
 fitting a linear model to past gradient evaluations using weighted discret
 e least-squares\, effectively reducing variance while preserving computati
 onal efficiency. We establish theoretical sublinear convergence guarantees
  and demonstrate the method&rsquo\;s effectiveness through numerical exper
 iments on random PDE-constrained optimization problems.
LOCATION:Seminar Room 2\, Newton Institute
END:VEVENT
END:VCALENDAR