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:20210916T132454Z
LOCATION:Michel Mayor
DTSTART;TZID=Europe/Stockholm:20210708T183000
DTEND;TZID=Europe/Stockholm:20210708T190000
UID:submissions.pasc-conference.org_PASC21_sess167_msa365@linklings.com
SUMMARY:Active Learning for Machine Learning Potentials
DESCRIPTION:Minisymposium\n\nActive Learning for Machine Learning Potentia
 ls\n\nSmith\n\nMachine learning (ML), trained on quantum mechanics calcula
 tions, is a powerful tool for modeling potential energy surfaces of molecu
 les and materials. The development of new neural network architectures and
  intelligent data selection schemes has led to broadly applicable ML-based
  potentials in recent years. Message passing neural network model potentia
 ls, e.g. HIP-NN, which inherently accounts for long-range interactions, ha
 ve been applied for end-to-end learning of potential surfaces. Our recent 
 efforts have further improved such models by generalizing the distance bas
 ed descriptors to a tensorial representation. Further, a critical factor i
 n ML potential development is the quality and diversity of the training da
 taset. For this presentation I will discuss our recent developments for hi
 ghly automated active learning-based dataset construction, as well as our 
 recent improvements and applications of ML-based potentials.\n\nDomain: CS
  and Math, Chemistry and Materials, Physics
END:VEVENT
END:VCALENDAR