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:20260202T201804Z
LOCATION:242
DTSTART;TZID=America/Chicago:20251116T114500
DTEND;TZID=America/Chicago:20251116T115500
UID:submissions.supercomputing.org_SC25_sess216_ws_isav106@linklings.com
SUMMARY:Lightning Talk: Differentiable In-Situ Diagnostics for Particle Ac
 celerator Digital Twins
DESCRIPTION:Axel Huebl, Samuel Barber, and Christopher Doss (Lawrence Berk
 eley National Laboratory (LBNL)); Auralee Edelen (SLAC National Laboratory
 ); Arianna Formenti and Chad E Mitchell (Lawrence Berkeley National Labora
 tory (LBNL)); Ryan Roussel (SLAC National Laboratory); and Jean-Luc Vay, D
 an Wang, and Remi Lehe (Lawrence Berkeley National Laboratory (LBNL))\n\nS
 imulation codes are extensively used to design and operate particle accele
 rators. \nSignificant effort is invested towards digital twins that closel
 y mirror a physical system and inform tuning of particle accelerators in r
 eal time, during operation. Some aspects of these digital twins benefit fr
 om using differentiable simulation codes that can automatically compute th
 e gradient of their output with respect to certain input parameters. Gradi
 ents can also be used to find the optimal operating point of an accelerato
 r (e.g. maximizing beam energy) and thereby inform tuning of the physical 
 accelerator.\n\nThis talk will point out the crucial need for in-situ diag
 nostic in this type of gradient-based workflow. It is crucial that modelin
 g reproduces the accelerator diagnostics in-situ rather than in post-proce
 ssing, so that the gradients can propagate throughout the simulation pipel
 ine. We demonstrate a real-world accelerator beamline model from PyTorch a
 nd will describe how to incorporate differentiable diagnostics in a C++ co
 de.\n\nTag: AI, Machine Learning, & Deep Learning, Clouds & Distributed Co
 mputing, Performance Evaluation, Scalability, & Portability, Scientific & 
 Information Visualization\n\nRecording: Partially Livestreamed, Partially 
 Recorded\n\nRegistration Category: Technical Program Reg Pass, Workshop Re
 g Pass\n\nSession Chairs: Axel Huebl (Lawrence Berkeley National Laborator
 y (LBNL)); Will Usher (Luminary Cloud Inc); Nicola Ferrier (Argonne Nation
 al Laboratory (ANL), University of Chicago); Estelle Dirand (TotalEnergies
 , France); Victor Mateevitsi (Argonne National Laboratory (ANL), Universit
 y of Illinois Chicago); and Jens Henrik Göbbert (Forschungszentrum Jülich,
  Jülich Supercomputing Centre (JSC))\n\n
END:VEVENT
END:VCALENDAR