BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Los_Angeles
X-LIC-LOCATION:America/Los_Angeles
BEGIN:DAYLIGHT
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
TZNAME:PDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
TZNAME:PST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20240626T180033Z
LOCATION:3001\, 3rd Floor
DTSTART;TZID=America/Los_Angeles:20240626T113000
DTEND;TZID=America/Los_Angeles:20240626T114500
UID:dac_DAC 2024_sess102_RESEARCH1361@linklings.com
SUMMARY:GNNavigator: Towards Adaptive Training of Graph Neural Networks vi
 a Automatic Guideline Exploration
DESCRIPTION:Research Manuscript\n\nTong Qiao, Jianlei Yang, Yingjie Qi, an
 d Ao Zhou (Beihang University); Chen Bai and Bei Yu (The Chinese Universit
 y of Hong Kong); and Weisheng Zhao and Chunming Hu (Beihang University)\n\
 nGraph Neural Networks (GNNs) succeed significantly in many applications r
 ecently. However, balancing GNNs training runtime cost, memory consumption
 , and attainable accuracy for various applications is non-trivial. Previou
 s training methodologies suffer from inferior adaptability and lack a unif
 ied training optimization solution. To address the problem, this work prop
 oses GNNavigator, an adaptive GNN training configuration optimization fram
 ework. GNNavigator meets diverse GNN application requirements due to our u
 nified software-hardware co-abstraction, proposed GNNs training performanc
 e model, and practical design space exploration solution. Experimental res
 ults show that GNNavigator can achieve up to 3.1X speedup and 44.9% peak m
 emory reduction with comparable accuracy to state-of-the-art approaches.\n
 \nTopic: AI\n\nKeyword: AI/ML Algorithms\n\nSession Chairs: Parivesh Choud
 hary (Synopsys) and Anca Molnos (CEA-List)
END:VEVENT
END:VCALENDAR