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:20200129T163556Z
LOCATION:506
DTSTART;TZID=America/Denver:20191118T143500
DTEND;TZID=America/Denver:20191118T150000
UID:submissions.supercomputing.org_SC19_sess137_ws_xloop104@linklings.com
SUMMARY:Scientific Image Restoration Anywhere
DESCRIPTION:Workshop\n\nScientific Image Restoration Anywhere\n\nAbeykoon,
  Liu, Kettimuthu, Fox, Foster\n\nThe use of deep learning models within sc
 ientific experimental facilities frequently requires low-latency inference
 , so that, for example, quality control operations can be performed while 
 data are being collected. Edge computing devices can be useful in this con
 text, as their low cost and compact form factor permit them to be co-locat
 ed with the experimental apparatus.\n\nCan such devices, with their limite
 d resources, can perform neural network feed-forward computations efficien
 tly and effectively? We explore this question by evaluating the performanc
 e and accuracy of a scientific image restoration model, for which both mod
 el input and output are images, on edge computing devices. Specifically, w
 e evaluate deployments of TomoGAN, an image-denoising model based on gener
 ative adversarial networks developed for low-dose x-ray imaging, on the Go
 ogle Edge TPU and NVIDIA Jetson. We adapt TomoGAN for edge execution, eval
 uate model inference performance, and propose methods to address the accur
 acy drop caused by model quantization. We show that these edge computing d
 evices can deliver accuracy comparable to that of a full-fledged CPU or GP
 U model, at speeds that are more than adequate for use in the intended dep
 loyments, denoising a 1024x1024 image in less than a second.  Our experime
 nts also show that the Edge TPU models can provide 3x faster inference res
 ponse than a CPU-based model and 1.5x faster than an edge GPU-based model.
   This combination of high speed and low cost permits image restoration an
 ywhere.\n\nTag: Workshop Reg Pass, Experimental Datasets, High-fidelity Si
 mulations, Scalable Computing\n\nRegistration Category: Workshop Reg Pass,
  Experimental Datasets, High-fidelity Simulations, Scalable Computing
URL:https://sc19.supercomputing.org/presentation/?id=ws_xloop104&sess=sess
 137
END:VEVENT
END:VCALENDAR