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:20210916T132528Z
LOCATION:Henry Dunant
DTSTART;TZID=Europe/Stockholm:20210707T140000
DTEND;TZID=Europe/Stockholm:20210707T160000
UID:submissions.pasc-conference.org_PASC21_sess152@linklings.com
SUMMARY:Exploiting and Supporting Exascale Infrastructure for Deep Learnin
 g Applications
DESCRIPTION:Minisymposium\n\nDeep learning (DL) has the capability to tran
 sform many scientific problems, including COVID-19 research, cancer studie
 s, and a wide range of energy sciences.  Ever more complex couplings of da
 ta sources, analysis, and learning methods demand additional resources in 
 computing capability, data movement, and storage.  In practice, the growin
 g scientific usage of DL faces technical and engineering challenges in the
  exascale era.  This minisymposium will illustrate two different learning-
 based applications and two related systems software contributions to DL at
  exascale. Exascale challenges for DL include scaling models, scaling work
 flows, integrating methods, integrating software, and down-scaling results
  for return to users with varying levels of available computing power.  Th
 e presenters of this minisymposium are developing a suite of software to s
 upport scalable DL on exascale supercomputing resources. The software suit
 e includes capabilities for rapid prototyping and application development,
  portability across machine types and scales, support for typical DL workf
 low patterns such as error analysis and hyperparameter optimization, and h
 ierarchical parallelism via distribution of larger models across multiple 
 nodes.  We will also present a range of research results in the usage of d
 eep learning ensembles, including recursive search patterns used to look f
 or outliers in experimental data.\n\nTargeting Billion-Scale Compound Libr
 aries with Deep Learning: A SARS-Cov-2 Example\n\nBrettin, Babuji, Clyde, 
 Jain, Liu...\n\nDeep learning-based surrogate models can be effective appr
 oaches to speed up costly computations in large workflows. In this work, w
 e present the evaluation of compounds from a large chemical database using
  surrogate DL models in place of the slower traditional docking approaches
 . Our work demonstra...\n\n---------------------\nProgrammable Infrastruct
 ure for Diverse, Scalable Learning at Exascale\n\nWozniak\n\nA wide range 
 of problems in deep learning can be addressed with ensembles of training r
 uns, including hyperparameter sweeps and optimization, robustness, sensiti
 vity, and statistical studies, and incremental training approaches to stud
 y the underlying data.  Many of these cases can benefit from...\n\n--
 -------------------\nHealth Data Science at Scale\n\nGounley\n\nInformatio
 n extraction from clinical text documents is an important tool for cancer 
 surveillance and research, with phenotyping used to stratify patients into
  cohorts based on tumor biology, responsiveness to treatment, and many oth
 er factors. To facilitate this work, natural language processing tec...\n\
 n---------------------\nReinforcement Learning at Exascale\n\nMohd Yusof, 
 Ramakrishnaiah\n\nMost contemporary researches in Reinforcement Learning (
 RL) use simple environments where the action results in an immediate state
  change. In addition, the learning is not usually done at scale with HPC r
 esources, such as the U.S. Department of Energy (DOE) leadership-class mac
 hines. Many scientific...\n\n\nDomain: CS and Math, Life Sciences
END:VEVENT
END:VCALENDAR