BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20190719T085744Z
LOCATION:HG F 3
DTSTART;TZID=Europe/Stockholm:20190613T181500
DTEND;TZID=Europe/Stockholm:20190613T184500
UID:submissions.pasc-conference.org_PASC19_sess161_msa322@linklings.com
SUMMARY:Learning to Optimize Machine Learning Workloads
DESCRIPTION:Minisymposium\nComputer Science and Applied Mathematics, Emerg
 ing Application Domains\n\nLearning to Optimize Machine Learning Workloads
 \n\nTomioka\n\nMore and more specialized hardware accelerators targeting m
 achine learning and computer vision are becoming available with different 
 architectural advantages and constraints. How can we maximize developer pr
 oductivity while achieving high performance on a variety of hardware accel
 erators? We are developing a machine-learning-based compiler towards this 
 goal. Our key observations are: first, representing programs as graphs (ra
 ther than a sequence of tokens) allows us to handle locality properly whic
 h is important because most optimization opportunities are local; second, 
 our reinforcement-learning-based agent operating on rewrite rules achieves
  some non-trivial optimizations including common subexpression elimination
  and inlining decisions.
END:VEVENT
END:VCALENDAR