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:20211207T055404Z
LOCATION:223
DTSTART;TZID=America/Chicago:20211115T114500
DTEND;TZID=America/Chicago:20211115T121000
UID:submissions.supercomputing.org_SC21_sess332_ws_pdsw110@linklings.com
SUMMARY:Data-Aware Storage Tiering for Deep Learning
DESCRIPTION:Workshop\n\nData-Aware Storage Tiering for Deep Learning\n\nXu
 , Bhattacharya, Foltin, Byna, Faraboschi\n\nDNN models trained with large 
 datasets can perform rich deep learning tasks with high accuracy. However,
  feeding huge volumes of training data exerts significant pressure on IO s
 ubsystems as the entire data is re-loaded in random order on every iterati
 on to enable convergence, with very little scope for reuse. To address thi
 s challenge, we co-optimize data tiering and iteration in DNN training for
  any given dataset and model with bandwidth and convergence conscious mini
 -epoch training (MET). This approach can substantially reduce the IO bandw
 idth required to provide sustained read throughput. Further, we introduce 
 two different feedback mechanisms to adjust the repeating factor over each
  mini-epoch during the training. We have evaluated three different applica
 tions with MET. Most of them work out-of-box with modest MET parameters. T
 he adaptive repeating factor design was able to gain back most of the accu
 racy drop due lo large MET parameters.\n\nTag: Data Analytics, Data Manage
 ment, File Systems and I/O, Storage\n\nRegistration Category: Workshop Reg
  Pass
END:VEVENT
END:VCALENDAR