BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Scalable simulation and inference in non-Gaussian stochastic PDEs 
 - David Duvenaud (University of Toronto)
DTSTART:20221215T110000Z
DTEND:20221215T120000Z
UID:TALK193870@talks.cam.ac.uk
CONTACT:James Allingham
DESCRIPTION:This talk presents early results along a path to scalable appr
 oximate inference schemes in large spatiotemporal models\, such as weather
  or molecular dynamics simulations. Specifically\, we’ll show how existi
 ng heuristics for scaling physical models such as coarse grids or mutli-sc
 ale temporal models can be learned automatically as auxiliary variables in
  variational posteriors. We’ll also demonstrate a new contribution to pa
 rallelizing adaptive SPDE solvers\, allowing stateless sampling of entire 
 Brownian sheets of any dimension. Finally\, we’ll show how to extend sto
 chastic variational inference in SDEs to include arbitrary jump processes.
LOCATION:Cambridge University Engineering Department\, CBL Seminar room BE
 4-38.
END:VEVENT
END:VCALENDAR