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:20260422T143138Z
LOCATION:B311
DTSTART;TZID=America/New_York:20241121T140000
DTEND;TZID=America/New_York:20241121T143000
UID:submissions.supercomputing.org_SC24_sess381_pap575@linklings.com
SUMMARY:Large Language Models for Anomaly Detection in Computational Workf
 lows: from Supervised Fine-Tuning to In-Context Learning
DESCRIPTION:Hongwei Jin (Argonne National Laboratory (ANL)), George Papadi
 mitriou (University of Southern California (USC)), Krishnan Raghavan (Argo
 nne National Laboratory (ANL)), Pawel Zuk (University of Southern Californ
 ia (USC)), Prasanna Balaprakash (Oak Ridge National Laboratory (ORNL)), Co
 ng Wang and Anirban Mandal (Renaissance Computing Institute (RENCI)), and 
 Ewa Deelman (University of Southern California (USC))\n\nAnomaly detection
  in computational workflows is critical for ensuring system reliability an
 d security. However, traditional rule-based methods struggle to detect nov
 el anomalies. This paper explores leveraging large language models (LLMs) 
 for workflow anomaly detection by exploiting their ability to learn comple
 x data patterns. Two approaches are investigated: 1) supervised fine-tunin
 g (SFT), where pre-trained LLMs are fine-tuned on labeled data for sentenc
 e classification to identify anomalies, and 2) in-context learning (ICL) w
 here prompts containing task descriptions and examples guide LLMs in few-s
 hot anomaly detection without fine-tuning. The paper evaluates the perform
 ance, efficiency, generalization of SFT models, and explores zero-shot and
  few-shot ICL prompts and interpretability enhancement via chain-of-though
 t prompting. Experiments across multiple workflow datasets demonstrate the
  promising potential of LLMs for effective anomaly detection in complex ex
 ecutions.\n\nTag: Accelerators, Applications and Application Frameworks, A
 rtificial Intelligence/Machine Learning, Modeling and Simulation, Numerica
 l Methods\n\nRegistration Category: Tech Program Reg Pass\n\nSession Chair
 : Murali Emani (Argonne National Laboratory (ANL))\n\n
END:VEVENT
END:VCALENDAR