BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Chicago
X-LIC-LOCATION:America/Chicago
BEGIN:DAYLIGHT
TZOFFSETFROM:-0600
TZOFFSETTO:-0500
TZNAME:CDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0500
TZOFFSETTO:-0600
TZNAME:CST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20211207T055347Z
LOCATION:Online
DTSTART;TZID=America/Chicago:20211115T163000
DTEND;TZID=America/Chicago:20211115T170000
UID:submissions.supercomputing.org_SC21_sess422_ws_pmbsf103@linklings.com
SUMMARY:Bayesian Optimization for Auto-Tuning GPU kernels
DESCRIPTION:Workshop\n\nBayesian Optimization for Auto-Tuning GPU kernels\
n\nWillemsen, van Nieuwpoort, van Werkhoven\n\nFinding optimal parameter c
onfigurations for tunable GPU kernels is a non-trivial exercise for large
search spaces, even when automated. This poses an optimization task on a n
on-convex search space, using an expensive-to-evaluate function with unkno
wn derivative. These characteristics make a good candidate for Bayesian o
ptimization, which previously has not been applied to this problem. \n\nTh
e application of Bayesian optimization to this problem, however, is challe
nging. We demonstrate how to deal with the rough, discrete, constrained se
arch spaces, containing invalid configurations. We introduce a novel cont
extual variance exploration factor, as well as new acquisition functions
with improved scalability, combined with an informed acquisition function
selection mechanism. By comparing the performance of our Bayesian optimiza
tion implementation on various test cases to the existing search strategie
s in Kernel Tuner, as well as other Bayesian optimization implementations,
we demonstrate that our search strategies generalize well and consistentl
y outperform other search strategies by a wide margin.\n\nTag: Online Only
, Accelerator-based Architectures, Applications, Computational Science, Em
erging Technologies, Extreme Scale Computing, File Systems and I/O, Hetero
geneous Systems, Parallel Programming Languages and Models, Performance, S
cientific Computing, Software Engineering\n\nRegistration Category: Worksh
op Reg Pass
END:VEVENT
END:VCALENDAR