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:20210916T132452Z
LOCATION:Henry Dunant
DTSTART;TZID=Europe/Stockholm:20210707T153000
DTEND;TZID=Europe/Stockholm:20210707T160000
UID:submissions.pasc-conference.org_PASC21_sess152_msa304@linklings.com
SUMMARY:Reinforcement Learning at Exascale
DESCRIPTION:Minisymposium\n\nReinforcement Learning at Exascale\n\nMohd Yu
 sof, Ramakrishnaiah\n\nMost contemporary researches in Reinforcement Learn
 ing (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 resources, such as the U.S. Department of Energy (DOE) leadership-clas
 s machines. Many scientific environments are complex and have actions that
  can take multiple hours to get a response - even on large clusters (e.g.,
  NWChem; LAMMPS). Therefore, these complex scientific environments not onl
 y require scalable learning but also the ability to run environments in pa
 rallel on multiple nodes. OpenAI Gym is ubiquitously used for creating RL 
 environments, but it does not inherently support scalable environments. No
 t only that, OpenAI Gym does not provide scalable RL algorithms for scienc
 e problems. To overcome these limitations, the ExaLearn Control team devel
 oped the Easily eXtendable Architecture for Reinforcement Learning (EXARL)
  framework for scientific environments. EXARL extends OpenAI Gym’s e
 nvironment registry to agents and workflows making it easy to incorporate 
 a suite of different environments, agents, and workflows. Uniquely, EXARL 
 supports parallel execution of scientific environments in addition to para
 llel training of RL policies. CANDLE tools are utilized and extended for e
 ase of hyper-parameter access and optimization.\n\nDomain: CS and Math, Li
 fe Sciences
END:VEVENT
END:VCALENDAR