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:20230831T095746Z
LOCATION:Sertig
DTSTART;TZID=Europe/Stockholm:20230627T160000
DTEND;TZID=Europe/Stockholm:20230627T163000
UID:submissions.pasc-conference.org_PASC23_sess170_msa242@linklings.com
SUMMARY:Accelerating Earth System Emulation with Spherical Fourier Neural 
 Operators
DESCRIPTION:Minisymposium\n\nAnima Anandkumar (California Institute of Tec
 hnology, NVIDIA Inc.); Thorsten Kurth, Boris Bonev, Christian Hundt, and K
 arthik Kashinath (NVIDIA Inc.); and Mike Pritchard (NVIDIA Inc.; Universit
 y of California, Irvine)\n\nNVIDIA's Earth-2 initiative aims to build digi
 tal twins of Earth and has two central goals: (i) Computational: enable hi
 gh-resolution weather and climate predictions with principled physics-insp
 ired ML; and (ii) Societal: nimbly serve high-fidelity, high-resolution pr
 edictions via an intuitive interactive system. The above-mentioned goals d
 epend on achieving orders-of-magnitude speedup and data compression via a 
 combination of advances in AI and computing technologies, in particular, A
 I at supercomputing scale. In this presentation we discuss the latest deve
 lopments in our Fourier Neural Operator-based Earth System emulator. We sh
 ow how a novel geometric deep learning algorithm called the Spherical Four
 ier Neural Operator (SFNO) that respects the spherical geometry of Earth a
 chieves stable year-long autoregressive rollouts whilst maintaining remark
 able skill for high-resolution medium-range global weather predictions. Th
 is advancement has important implications for data-driven subseasonal-to-s
 easonal prediction, and eventually for climate prediction, although many c
 hallenges need to be addressed. We discuss the optimizations, parallelism,
  I/O and memory challenges that were addressed to enable efficient predict
 ion on multi-GPU, multi-node systems. Finally, we discuss opportunities an
 d challenges towards scaling these developments to km-scale weather and cl
 imate simulations. We conclude with machine learning algorithmic and engin
 eering innovations required for breakthroughs that will truly help realize
  digital twins of Earth.\n\nDomain: Climate, Weather and Earth Sciences\n\
 nSession Chair: Karthik Kashinath (NVIDIA Inc., Lawrence Berkeley National
  Laboratory)
END:VEVENT
END:VCALENDAR