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:20260522T150116Z
LOCATION:C2/3/4 Ballroom
DTSTART;TZID=America/Chicago:20181113T083000
DTEND;TZID=America/Chicago:20181113T170000
UID:submissions.supercomputing.org_SC18_sess325_spost127@linklings.com
SUMMARY:Precomputing Outputs of Hidden Layers to Speed Up Deep Neural Netw
 ork Training
DESCRIPTION:Sohil Lal Shrestha (University of Texas, Arlington)\n\nDeep le
 arning has recently emerged as a powerful technique for many tasks includi
 ng image classification. A key bottleneck of deep learning is that the tra
 ining phase takes a lot of time, since state-of-the-art deep neural networ
 ks have millions of parameters and hundreds of hidden layers. The early la
 yers of these deep neural networks have the fewest parameters but take up 
 the most computation.\n\nIn this work, we reduce training time by progress
 ively freezing hidden layers, pre-computing their output and excluding the
 m from training in both forward and backward paths in subsequent iteration
 s. We compare this technique to the most closely related approach for spee
 ding up the training process of neural network.\n\nThrough experiments on 
 two widely used datasets for image classification, we empirically demonstr
 ate that our approach can yield savings of up to 25% wall-clock time durin
 g training with no loss in accuracy.\n\nRegistration Category: Tech Progra
 m Reg Pass, Exhibits Reg Pass\n\n
END:VEVENT
END:VCALENDAR