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:20220812T074335Z
LOCATION:Singapore Room
DTSTART;TZID=Europe/Stockholm:20220629T123000
DTEND;TZID=Europe/Stockholm:20220629T130000
UID:submissions.pasc-conference.org_PASC22_sess156_msa110@linklings.com
SUMMARY:Parallelized Integrated Nested Laplace Approximations for Fast Bay
 esian Inference
DESCRIPTION:Minisymposium\n\nParallelized Integrated Nested Laplace Approx
 imations for Fast Bayesian Inference\n\nGaedke-Merzhäuser, Rue, Schenk\n\n
 Bayesian computing has been on the rise for years due to its ability to ma
 ke reliable predictions using flexible yet interpretable modelling framewo
 rks, while also providing uncertainty estimates for all parameters in the 
 form of distributions. For more complex models, performing the Bayesian in
 ference task often remains computationally challenging and therefore time-
 consuming. The methodology of integrated nested Laplace approximations (IN
 LA) offers a way to perform inference in much smaller runtimes than Markov
  chain Monte Carlo methods, usually at similar accuracy. INLA is applicabl
 e to a wide subclass of hierarchical Bayesian additive models. Its key com
 ponents include a nested approximation strategy to mitigate the bottleneck
  operation of high-dimensional integration and the usage of sparse Gaussia
 n Markov random fields for the latent parameter space. A versatile and use
 r-friendly implementation exists in the form of an R-package. To continuou
 sly support the increasing amounts of available data and higher-dimensiona
 l parameter spaces, we introduced various levels of parallelism to the R-I
 NLA implementation, drastically improving its performance and scalability.
  We present the ideas behind these parallelization strategies, which inclu
 de the simultaneous computation of all computationally involved operations
  using OpenMP, the inclusion of sparse linear direct solver PARDISO and a 
 parallel line search in INLA’s optimization scheme.\n\nDomain: Computer Sc
 ience and Applied Mathematics
END:VEVENT
END:VCALENDAR