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:20260202T201803Z
LOCATION:242
DTSTART;TZID=America/Chicago:20251116T111000
DTEND;TZID=America/Chicago:20251116T112000
UID:submissions.supercomputing.org_SC25_sess216_ws_isav102@linklings.com
SUMMARY:Lightning Talk: Adaptive In-Situ Sampling Framework for Accelerate
 d Online Change Point Detection in Scientific Workflows
DESCRIPTION:Vijayalakshmi Saravanan (University of Texas at Tyler, UTD); S
 ai Karthik Navuluru (UTD); Seetharam Vanam Rao (UTT); Tamil Lakshman (UTD)
 ; and Khaled Z Ibrahim (Lawrence Berkeley National Laboratory (LBNL))\n\nH
 igh-performance computing (HPC) systems at the\nExascale generate monitori
 ng and simulation data at rates that\nexceed the capacity of traditional o
 ffline analysis, especially for\nchange point detection (CPD) in large-sca
 le scientific workflows.\nWe present an adaptive in-situ sampling framewor
 k that combines\ntruncated singular value decomposition (SVD)-based manifo
 ld\nlearning with the Kernel Cumulative Sum (KCUSUM) method\nfor statistic
 al CPD. Implemented with MPI4py for scalable\ninterprocess communication a
 nd ADIOS2 for high-throughput\nstreaming I/O, the framework dynamically ad
 justs sampling rates\nin real time based on anomaly likelihood, enabling s
 elective data\nretention without sacrificing scientific information.\nEval
 uations on synthetic datasets and large-scale molecular\ndynamics (MD) sim
 ulations from NWChem demonstrate up to\n59% memory reduction, sub-second d
 etection latency for critical\nevents, and near-perfect detection accuracy
 , all while eliminating\nthe need for storing full trajectories. These pre
 liminary results\nhighlight the framework’s potential to deliver resource-
 efficient,\nreal-time anomaly detection in data-intensive Exascale scienti
 fic\ncomputing environments,\n\nTag: AI, Machine Learning, & Deep Learning
 , Clouds & Distributed Computing, Performance Evaluation, Scalability, & P
 ortability, Scientific & Information Visualization\n\nRecording: Partially
  Livestreamed, Partially Recorded\n\nRegistration Category: Technical Prog
 ram Reg Pass, Workshop Reg Pass\n\nSession Chairs: Axel Huebl (Lawrence Be
 rkeley National Laboratory (LBNL)); Will Usher (Luminary Cloud Inc); Nicol
 a Ferrier (Argonne National Laboratory (ANL), University of Chicago); Este
 lle Dirand (TotalEnergies, France); Victor Mateevitsi (Argonne National La
 boratory (ANL), University of Illinois Chicago); and Jens Henrik Göbbert (
 Forschungszentrum Jülich, Jülich Supercomputing Centre (JSC))\n\n
END:VEVENT
END:VCALENDAR