BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Denver
X-LIC-LOCATION:America/Denver
BEGIN:DAYLIGHT
TZOFFSETFROM:-0700
TZOFFSETTO:-0600
TZNAME:MDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0600
TZOFFSETTO:-0700
TZNAME:MST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260422T000712Z
LOCATION:403-404
DTSTART;TZID=America/Denver:20231116T153000
DTEND;TZID=America/Denver:20231116T160000
UID:submissions.supercomputing.org_SC23_sess163_pap202@linklings.com
SUMMARY:Phases, Modalities, Spatial and Temporal Locality: Domain Specific
  ML Prefetcher for Accelerating Graph Analytics
DESCRIPTION:Pengmiao Zhang (University of Southern California (USC)), Rajg
 opal Kannan (DEVCOM US Army Research Lab), and Viktor K. Prasanna (Univers
 ity of Southern California (USC))\n\nMemory performance is a bottleneck in
  graph analytics acceleration. Existing Machine Learning (ML) prefetchers 
 struggle with phase transitions and irregular memory accesses in graph pro
 cessing. We propose MPGraph, an ML-based Prefetcher for Graph analytics us
 ing domain specific models. MPGraph introduces three novel optimizations: 
 soft detection for phase transitions, phase-specific multi-modality models
  for access delta and page predictions, and chain spatio-temporal prefetch
 ing (CSTP) for prefetch control.\n\nOur transition detector achieves 34.17
 –82.15% higher precision compared with Kolmogorov–Smirnov Windowing and de
 cision tree. Our predictors achieve 6.80–16.02% higher F1-score for delta 
 and 11.68–15.41% higher accuracy-at-10 for page prediction compared with L
 STM and vanilla attention models. Using CSTP, MPGraph achieves 12.52–21.23
 % IPC improvement, outperforming state-of-the-art non-ML prefetcher BO by 
 7.58–12.03% and ML-based prefetchers Voyager and TransFetch by 3.27–4.58%.
  For practical implementation, we demonstrate MPGraph using compressed mod
 els with reduced latency shows significantly superior accuracy and coverag
 e compared with BO, leading to 3.58% higher IPC improvement.\n\nTag: Archi
 tecture and Networks, Data Movement and Memory, Graph Algorithms and Frame
 works, Performance Measurement, Modeling, and Tools, Programming Framework
 s and System Software\n\nRegistration Category: Tech Program Reg Pass\n\nR
 eproducibility Badges: Artifact Available, Artifact Functional, Results Re
 produced\n\nSession Chair: Mahantesh Halappanavar (Pacific Northwest Natio
 nal Laboratory (PNNL))\n\n
END:VEVENT
END:VCALENDAR