BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Chicago
X-LIC-LOCATION:America/Chicago
BEGIN:DAYLIGHT
TZOFFSETFROM:-0600
TZOFFSETTO:-0500
TZNAME:CDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0500
TZOFFSETTO:-0600
TZNAME:CST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20230124T170804Z
LOCATION:C1-2-3
DTSTART;TZID=America/Chicago:20221116T083000
DTEND;TZID=America/Chicago:20221116T170000
UID:submissions.supercomputing.org_SC22_sess225_spostg101@linklings.com
SUMMARY:Efficiently Learning Locality Optimizations by Decomposing Transfo
 rmation Domains
DESCRIPTION:ACM Student Research Competition: Graduate Poster, ACM Student
  Research Competition: Undergraduate Poster, Posters\n\nEfficiently Learni
 ng Locality Optimizations by Decomposing Transformation Domains\n\nPataban
 di\n\nAchieving full automation of program optimization is still an open p
 roblem for compiler writers. This work explores machine learning as a pote
 ntial solution to learn data locality optimizations for tensor application
 s. Training models with supervised-learning for loop-nest optimization oft
 en requires prohibitively expensive training data generation for learning 
 the combined effects of a transformation sequence. As a solution, this wor
 k proposes a novel learning strategy called Composed Singular Prediction (
 CSP) that significantly reduces the training data generation cost in the c
 ontext of learned loop transformation models. The learned models are then 
 deployed to predict data locality optimization schedules for Conv2d kernel
 s to achieve performance improvements up to 4x against Intel oneDNN while 
 saving over 100x in training data collection time over exhaustive search.\
 n\nRegistration Category: Tech Program Reg Pass, Exhibits Reg Pass
END:VEVENT
END:VCALENDAR