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:20220812T074334Z
LOCATION:Nairobi Room
DTSTART;TZID=Europe/Stockholm:20220628T110000
DTEND;TZID=Europe/Stockholm:20220628T113000
UID:submissions.pasc-conference.org_PASC22_sess143_msa122@linklings.com
SUMMARY:Efficient Top-Down Parameterization of Machine Learning-Based Mode
 ls
DESCRIPTION:Minisymposium\n\nEfficient Top-Down Parameterization of Machin
 e Learning-Based Models\n\nZavadlav\n\nMolecular modeling has become a cor
 nerstone of many disciplines, including material science. However, the qua
 lity of predictions critically depends on the employed potential energy mo
 del. A class of models with tremendous success in recent years are neural 
 network (NN) potentials due to their flexibility and capacity of learning 
 many-body interactions. Traditionally, these models are trained bottom-up 
 on quantum mechanical data. Top-down approaches that learn NN potentials d
 irectly from experimental data have received less attention, typically fac
 ing numerical and computational challenges when backpropagating through mo
 lecular dynamics (MD) simulations. We recently developed the Differentiabl
 e Trajectory Reweighting (DiffTRe) method, which bypasses differentiation 
 through the MD simulation for time-independent observables. Leveraging the
 rmodynamic perturbation theory, we avoid exploding gradients and achieve a
 round 2 orders of magnitude speed-up in gradient computation for top-down 
 learning. The effectiveness of DiffTRe is showcased on atomistic and coars
 e-grained models using diverse experimental observables including thermody
 namic, structural, and mechanical properties. Our approach opens the way t
 o high fidelity molecular models, particularly when bottom-up data is unav
 ailable or insufficiently accurate.\n\nDomain: Chemistry and Materials, Ph
 ysics
END:VEVENT
END:VCALENDAR