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:20181221T160727Z
LOCATION:D167/174
DTSTART;TZID=America/Chicago:20181111T160000
DTEND;TZID=America/Chicago:20181111T163000
UID:submissions.supercomputing.org_SC18_sess221_ws_mlhpce103@linklings.com
SUMMARY:Auto-Tuning TensorFlow Threading Model for CPU Backend
DESCRIPTION:Workshop\nApplications, Deep Learning, Machine Learning, Works
 hop Reg Pass\n\nAuto-Tuning TensorFlow Threading Model for CPU Backend\n\n
 Hasabnis\n\nTensorFlow is a popular deep learning framework used to solve 
 machine learning and deep learning problems such as image classification a
 nd speech recognition. It also allows users to train neural network models
  or deploy them for inference using GPUs, CPUs, and custom-designed hardwa
 re such as TPUs. Even though TensorFlow supports a variety of optimized ba
 ckends, realizing the best performance using a backend requires additional
  efforts. Getting the best performance from a CPU backend requires tuning 
 of its threading model. Unfortunately, the best tuning approach used today
  is manual, tedious, time-consuming, and, more importantly, may not guaran
 tee the best performance.\n\nIn this paper, we develop an automatic approa
 ch, called TENSORTUNER, to search for optimal parameter settings of Tensor
 Flow’s threading model for CPU backends. We evaluate TENSORTUNER on both E
 igen and Intel’s MKL CPU backends using a set of neural networks from Tens
 orFlow’s benchmarking suite. Our evaluation results demonstrate that the p
 arameter settings found by TENSORTUNER produce 2% to 123% performance impr
 ovement for the Eigen CPU backend and 1.5% to 28% performance improvement 
 for the MKL CPU backend over the performance obtained using their best-kno
 wn parameter settings. This highlights the fact that the default parameter
  settings in Eigen CPU backend are not the ideal settings; and even for a 
 carefully hand-tuned MKL backend, the settings are sub-optimal. Our evalua
 tions also revealed that TENSORTUNER is efficient at finding the optimal s
 ettings — it is able to converge to the optimal settings quickly by prunin
 g more than 90% of the parameter search space.
URL:https://sc18.supercomputing.org/presentation/?id=ws_mlhpce103&sess=ses
 s221
END:VEVENT
END:VCALENDAR