BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Denver
X-LIC-LOCATION:America/Denver
BEGIN:DAYLIGHT
TZOFFSETFROM:-0700
TZOFFSETTO:-0600
TZNAME:MDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0600
TZOFFSETTO:-0700
TZNAME:MST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20200129T163559Z
LOCATION:502-503-504
DTSTART;TZID=America/Denver:20191118T110000
DTEND;TZID=America/Denver:20191118T113000
UID:submissions.supercomputing.org_SC19_sess115_ws_mlhpce115@linklings.com
SUMMARY:Fine-Grained Exploitation of Mixed Precision for Faster CNN Traini
 ng
DESCRIPTION:Workshop\n\nFine-Grained Exploitation of Mixed Precision for F
 aster CNN Training\n\nJohnston, Young, Schuman, Chae, March...\n\nAs deep 
 convolutional neural networks (CNNs) have become increasingly popular and 
 successful at an ever-widening number of machine learning tasks specialize
 d hardware has become increasingly available for training and deploying th
 em.  NVIDIA's recent Volta architecture includes tensor cores which perfor
 m a fused operation reduced and mixed precision (16-bit multiply, 32-bit a
 ccumulate).  Recent research indicates that, typically, very little is los
 t (in terms of training accuracy) when half precision is used in place of 
 single precision, and performance gains can be made by doing arithmetic in
  reduced precision.  In this work we demonstrate that making layer-by-laye
 r choices as to the arithmetic/data precision can lead to further performa
 nce improvement.  In our study of 25,200 CNNs we demonstrate an average sp
 eedup (over purely half precision) of 1.27x and speedups as high as 3.64x 
 by appropriately combining single and half precision arithmetic and data t
 ypes on a layer-by-layer basis.\n\nTag: Workshop Reg Pass, Machine Learnin
 g\n\nRegistration Category: Workshop Reg Pass, Machine Learning
URL:https://sc19.supercomputing.org/presentation/?id=ws_mlhpce115&sess=ses
 s115
END:VEVENT
END:VCALENDAR

