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:20200129T163603Z
LOCATION:502-503-504
DTSTART;TZID=America/Denver:20191117T153000
DTEND;TZID=America/Denver:20191117T160000
UID:submissions.supercomputing.org_SC19_sess101_ws_dls124@linklings.com
SUMMARY:Scaling Distributed Training of Flood-Filling Networks on HPC Infr
 astructure for Brain Mapping
DESCRIPTION:Workshop\n\nScaling Distributed Training of Flood-Filling Netw
 orks on HPC Infrastructure for Brain Mapping\n\nDong, Keceli, Vescovi, Li,
  Adams...\n\nMapping all the neurons in the brain requires automatic recon
 struction of entire cells from volume electron microscopy data. The flood-
 filling network (FFN) architecture has demonstrated leading performance fo
 r segmenting structures from this data. However, the training of the netwo
 rk is computationally expensive. In order to reduce the training time, we 
 implemented synchronous and data-parallel distributed training using the H
 orovod library, which is different from the asynchronous training scheme u
 sed in the published FFN code. We demonstrated that our distributed traini
 ng scaled well up to 2048 Intel Knights Landing (KNL) nodes on the Theta s
 upercomputer. Our trained models achieved similar level of inference perfo
 rmance, but took less training time compared to previous methods. Our stud
 y on the effects of different batch sizes on FFN training suggests ways to
  further improve training efficiency. Our findings on optimal learning rat
 e and batch sizes agree with previous works.\n\nTag: Workshop Reg Pass, De
 ep Learning, Scientific Computing\n\nRegistration Category: Workshop Reg P
 ass, Deep Learning, Scientific Computing
URL:https://sc19.supercomputing.org/presentation/?id=ws_dls124&sess=sess10
 1
END:VEVENT
END:VCALENDAR