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:20220812T074334Z
LOCATION:Samarkand Room
DTSTART;TZID=Europe/Stockholm:20220627T111500
DTEND;TZID=Europe/Stockholm:20220627T114500
UID:submissions.pasc-conference.org_PASC22_sess175_pap101@linklings.com
SUMMARY:Communication Bounds for Convolutional Neural Networks
DESCRIPTION:Paper\n\nCommunication Bounds for Convolutional Neural Network
 s\n\nChen, Demmel, Dinh, Haberle, Holtz\n\nConvolutional neural networks (
 CNNs) are important in a wide variety of machine learning tasks and applic
 ations, so optimizing their performance is essential. Moving words of data
  between levels of a memory hierarchy or between processors on a network i
 s much more expensive than the cost of arithmetic, so minimizing communica
 tion is critical to optimizing performance. In this paper, we present new 
 precise lower bounds on data movement for convolutions in both single-proc
 essor and parallel distributed memory models, as well as algorithms that o
 utperform current implementations such as Im2Col. We obtain performance fi
 gures using GEMMINI, a machine learning accelerator, where our tiling prov
 ides improvements between 13% and 150% over a vendor supplied algorithm.\n
 \nDomain: Computer Science and Applied Mathematics, Humanities and Social 
 Sciences
END:VEVENT
END:VCALENDAR