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:20240626T180034Z
LOCATION:3003\, 3rd Floor
DTSTART;TZID=America/Los_Angeles:20240626T140000
DTEND;TZID=America/Los_Angeles:20240626T141500
UID:dac_DAC 2024_sess125_RESEARCH1179@linklings.com
SUMMARY:GSPO: A Graph Substitution and Parallelization Joint Optimization 
 Framework for DNN Inference
DESCRIPTION:Research Manuscript\n\nZheng Xu (Tsinghua University); Xu Dai 
 (Shanghai Artificial Intelligence Laboratory); and Shaojun Wei, Yang Hu, a
 nd Shouyi Yin (Tsinghua University)\n\nThis work proposes GSPO, an automat
 ic unified framework that jointly applies graph substitution and paralleli
 zation for DNN inference. GSPO uses joint optimization computation graph (
 JOCG) to represent both graph substitution and parallelization at the oper
 ator level. Then, a novel cost model customized for joint optimization is 
 used to quickly evaluate the computation graph execution time. Combined wi
 th backtracking search algorithm, GSPO is able to find the optimal joint o
 ptimization solution within acceptable search time. Compared to existing f
 rameworks applying equivalent graph substitution or parallelization, GSPO 
 can achieve up to 27.1% end-to-end performance improvement and reduce sear
 ch time by up to 94.3%.\n\nTopic: AI, Design\n\nKeyword: AI/ML System and 
 Platform Design\n\nSession Chairs: Amin Firoozshahian (Rain AI) and Thierr
 y Tambe (Stanford University)
END:VEVENT
END:VCALENDAR