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:20211207T054802Z
LOCATION:230-231-232
DTSTART;TZID=America/Chicago:20211117T103000
DTEND;TZID=America/Chicago:20211117T110000
UID:submissions.supercomputing.org_SC21_sess168_pap560@linklings.com
SUMMARY:Distributed Multigrid Neural Solver on Megavoxel Domains
DESCRIPTION:Paper\n\nDistributed Multigrid Neural Solver on Megavoxel Doma
 ins\n\nBalu, Botelho, Khara, Rao, Sarkar...\n\nWe consider the distributed
  training of neural networks that serve as PDE solvers producing full fiel
 d outputs. We specifically consider neural solvers for the generalized 3D 
 Poisson equation over megavoxel domains. A scalable framework is presented
  that integrates two distinct advances. First, we accelerate training a la
 rge model via a method analogous to the multigrid technique used in numeri
 cal linear algebra. Here, the network is trained using a hierarchy of incr
 easing resolution inputs in sequence, analogous to the 'V’, 'W’, F', and '
 Half-V' cycles used in multigrid approaches. In conjunction with the multi
 -grid approach, we implement a distributed deep learning framework which s
 ignificantly reduces the time to solve. We show the scalability of this ap
 proach on both GPU and CPU clusters. This approach is deployed to train a 
 generalized 3D Poisson solver that scales well to predict output full-fiel
 d solutions up to the resolution of 512 x 512 x 512.\n\nTag: Machine Learn
 ing and Artificial Intelligence, Numerical Algorithms\n\nRegistration Cate
 gory: Tech Program Reg Pass
END:VEVENT
END:VCALENDAR