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:20250626T233527Z
LOCATION:B302-B305
DTSTART;TZID=America/New_York:20241120T100000
DTEND;TZID=America/New_York:20241120T170000
UID:submissions.supercomputing.org_SC24_sess533_post224@linklings.com
SUMMARY:SanQus: Staleness and Quantization-Aware Full-Graph Decentralized 
 Training in GNNs
DESCRIPTION:Hongbo Yin (Hong Kong University of Science and Technology, Gu
 angzhou); Jingshu Peng (Hong Kong University of Science and Technology); Q
 iyu Liu (Southwest University); Zhao Chen (Hong Kong University of Science
  and Technology); Yingxia Shao (Beijing University of Posts and Telecommun
 ications); Yanyan Shen (Shanghai Jiao Tong University); Lei Chen (Hong Kon
 g University of Science and Technology); and Jiannong Cao (The Hong Kong P
 olytechnic University)\n\nGraph neural networks (GNNs) have demonstrated s
 ignificant success in modeling graphs; however, they encounter challenges 
 in efficiently scaling to large graphs. To address this, we propose the Sa
 nQus system, advancing our previous work, Sancus. SanQus reduces the need 
 for expensive communication among distributed workers by utilizing Stalene
 ss and Quantization-Aware broadcasting. SanQus manages embedding staleness
 , skips unnecessary broadcasts, and treats decentralized GNN processing as
  sequential matrix operations. To further reduce communication, SanQus cac
 hes historical embeddings and performs quantization-aware broadcast. Theor
 etically, SanQus demonstrates bounded approximation errors and optimal con
 vergence rates. Extensive experiments on big graphs with common GNN models
  show that SanQus reduces communication by up to 86% and triples throughpu
 t without sacrificing accuracy, outperforming state-of-the-art systems.\n\
 nRegistration Category: Tech Program Reg Pass, Exhibits Reg Pass\n\nSessio
 n Chairs: Ayesha Afzal (Friedrich-Alexander University, Erlangen-Nuremberg
 ; Erlangen National High Performance Computing Center); Sally Ellingson (U
 niversity of Kentucky); and Alan Sussman (University of Maryland)\n\n
END:VEVENT
END:VCALENDAR