BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20250822T115805Z
LOCATION:Campussaal - Plenary Room
DTSTART;TZID=Europe/Stockholm:20250617T103000
DTEND;TZID=Europe/Stockholm:20250617T110000
UID:submissions.pasc-conference.org_PASC25_sess150_posC110@linklings.com
SUMMARY:ACMP03 - Distributed Computing for Spatio-Temporal Bayesian Modeli
 ng Using the INLA Method
DESCRIPTION:Vincent Maillou and Alexandros Nikolaos Ziogas (ETH Zurich); O
 laf Schenk (Università della Svizzera italiana); Mathieu Luisier (ETH Zuri
 ch); Håvard Rue (King Abdullah University of Science and Technology); and 
 Lisa Gaedke-merzhaeuser (King Abdullah University of Science and Technolog
 y, Università della Svizzera italiana)\n\nBayesian inference on large-scal
 e spatio-temporal models is limited by its computational feasibility, a tr
 end that is further exacerbated by the continuous increase in data availab
 ility and model refinements. To address this issue, we present a double-la
 yer distributedmemory parallelization strategy for the popular integrated 
 nested Laplace approximations (INLA) method. First, we perform in parallel
  the different objective function evaluations that happen during the hyper
 parameter minimization process. Each of these function evaluations require
 s the assembly and decomposition of large, often structured, sparse precis
 ion matrices. A second layer of parallelism is thus found in the decomposi
 tion of these sparse matrices. They are handled by a GPU-accelerated, dist
 ributed memory, direct solver called Serinv, which we have integrated into
  our framework. The latter, named DALIA, is a novel Python package that ai
 ms at providing modern coding practices and features for spatio-temporal m
 odeling within the INLA method. DALIA outperforms the scalability\nof stat
 e-of-the-art packages, allowing the entire framework to run on distributed
  memory, GPU-accelerated, systems.We showcase the computational performanc
 e of our framework on the CSCS\nAlps supercomputer, scaling further a larg
 e spatio-temporal air temperature prediction model made of a spatial mesh 
 composed of 2865 edges over the course of 365 days.\n\nSession Chair: Davi
 d Moxey (King's College London)\n\n
END:VEVENT
END:VCALENDAR