BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20190719T085743Z
LOCATION:HG E 3
DTSTART;TZID=Europe/Stockholm:20190612T140000
DTEND;TZID=Europe/Stockholm:20190612T143000
UID:submissions.pasc-conference.org_PASC19_sess126_msa148@linklings.com
SUMMARY:Cosmological Parameter Estimation with Deep Learning
DESCRIPTION:Minisymposium\nComputer Science and Applied Mathematics, Emerg
 ing Application Domains, Physics, Life Sciences\n\nCosmological Parameter 
 Estimation with Deep Learning\n\nPanyasantisuk, Schöngens, Herbel, Kacprza
 k, Refregier\n\nThe point spread function (PSF) describes the smearing of 
 astronomical images due to Earth's atmosphere and imperfections in the opt
 ics. When constraining cosmological models using imaging data, it is very 
 important to account for the PSF, since it can bias the results when left 
 uncorrected. Recently, it was shown that convolutional neural networks (CN
 N) are a promising means of speeding up PSF modeling. However, training a 
 CNN on a single GPU is time consuming for large datasets. This problem cou
 ld be alleviated by exploiting suitable parallelization strategies amenabl
 e to modern supercomputers. In this work, we used MPI-based training frame
 works such as Cray's Machine Learning Plugin and Horovod, to parallelize t
 he CNN training of PSF parameter estimation in TensorFlow, on the CSCS Piz
  Daint supercomputer and ETH's Leonhard Open GPU Cluster. Our experiments 
 have shown good performance both for strong and weak scalability. Due to i
 ts general applicability, the approach can be interesting for a broad rang
 e of use cases.
END:VEVENT
END:VCALENDAR