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:20211207T055402Z
LOCATION:Online
DTSTART;TZID=America/Chicago:20211115T163500
DTEND;TZID=America/Chicago:20211115T170000
UID:submissions.supercomputing.org_SC21_sess342_ws_worksp112@linklings.com
SUMMARY:A Performance Characterization of Scientific Machine Learning Work
 flows
DESCRIPTION:Workshop\n\nA Performance Characterization of Scientific Machi
 ne Learning Workflows\n\nKrawczuk, Papadimitriou, Tanaka, Do, Subramanya..
 .\n\nScientific workflows are one of the well-established pillars of moder
 n large-scale computational science. More recently, scientists have starte
 d to leverage machine learning (ML) capabilities in their workflows, leadi
 ng to a new category of scientific workflows, denoted as scientific ML wor
 kflows. ML is not only about training and inference, modern ML workflows a
 lso involve complex data processing steps before the training can start, w
 hich are not often accounted for in most performance studies. In this work
 , we consider scientific ML workflows, from data pre-processing to trainin
 g, inference, and model evaluation. We aim to explore (i) how scientific M
 L workflows differ from more traditional scientific workflows and; (ii) ho
 w we can characterize ML workflows both in terms of execution time and dat
 a movements when executing on an exemplary cloud platform. We select three
  representative workflows, ranging from image classification to natural la
 nguage processing and image segmentation, which have been executed using t
 he academic cloud platform, Chameleon. We build four realistic deployment 
 scenarios for each workflow, which stress data movements during workflow e
 xecutions. Then, we compare the performance observed when utilizing these 
 different configurations and study how different settings impact overall w
 orkflows performance and efficiency when running on cloud infrastructures.
  Finally, we summarize our findings and discuss performance impacts when a
 ugmenting scientific workflows with ML techniques and how traditional work
 flow management systems can improve their support for such workflows.\n\nT
 ag: Online Only, Cloud and Distributed Computing, Scientific Computing, Wo
 rkflows\n\nRegistration Category: Workshop Reg Pass
END:VEVENT
END:VCALENDAR