BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/New_York
X-LIC-LOCATION:America/New_York
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260422T143140Z
LOCATION:B304
DTSTART;TZID=America/New_York:20241118T111000
DTEND;TZID=America/New_York:20241118T112500
UID:submissions.supercomputing.org_SC24_sess749_ws_drbsd110@linklings.com
SUMMARY:Filling the Void: Data-Driven Machine Learning-based Reconstructio
 n of Sampled Spatiotemporal Scientific Simulation Data
DESCRIPTION:Ayan Biswas (Los Alamos National Laboratory (LANL)); Aditi Mis
 hra (Arizona State University); Meghanto Majumder (Los Alamos National Lab
 oratory (LANL)); Subhashis Hazarika (Fujitsu Research of America Inc.); Al
 exander Most and Juan Castorena (Los Alamos National Laboratory (LANL)); C
 hristopher Bryan (Arizona State University); and Patrick McCormick, James 
 Ahrens, Earl Lawrence, and Aric Hagberg (Los Alamos National Laboratory (L
 ANL))\n\nAs high-performance computing systems continue to advance, the ga
 p between computing performance and I/O capabilities is widening. This bot
 tleneck limits the storage capabilities of increasingly large-scale simula
 tions, which generate data at never-before-seen granularities while only b
 eing able to store a small subset of the raw data. Recently, strategies fo
 r data-driven sampling have been proposed. However, a thorough analysis of
  how such intelligent samples can be used for data reconstruction is lacki
 ng. We propose a data-driven machine learning approach based on training n
 eural networks to reconstruct full-scale datasets based on a simulation’s 
 sampled output. Compared to current state-of-the-art reconstruction approa
 ches, we demonstrate that our machine learning-based reconstruction has se
 veral advantages, including reconstruction quality, time-to-reconstruct, a
 nd knowledge transfer to unseen timesteps and grid resolutions. We propose
  and evaluate strategies that balance the sampling rates with model traini
 ng (pretraining and fine-tuning) and data reconstruction time to demonstra
 te its efficacy.\n\nTag: Data Compression, Data Movement and Memory, Middl
 eware and System Software\n\nRegistration Category: Workshop Reg Pass\n\nS
 ession Chairs: Sheng Di (Argonne National Laboratory (ANL), University of 
 Chicago); Ana Gainaru (Oak Ridge National Laboratory (ORNL)); Sian Jin (Te
 mple University); Xin Liang (Oregon State University); Kento Sato (RIKEN C
 enter for Computational Science (R-CCS)); and Dingwen Tao (Institute of Co
 mputing Technology, Chinese Academy of Sciences; University of Chinese Aca
 demy of Sciences)\n\n
END:VEVENT
END:VCALENDAR