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:20220812T074335Z
LOCATION:Singapore Room
DTSTART;TZID=Europe/Stockholm:20220629T153000
DTEND;TZID=Europe/Stockholm:20220629T160000
UID:submissions.pasc-conference.org_PASC22_sess155_msa126@linklings.com
SUMMARY:Neural-Network Approaches for High-Dimensional Optimal Control Pro
 blems
DESCRIPTION:Minisymposium\n\nNeural-Network Approaches for High-Dimensiona
 l Optimal Control Problems\n\nRuthotto\n\nWe consider neural network appro
 aches to solving high-dimensional optimal control problems with determinis
 tic and randomly perturbed dynamics. The training process simultaneously a
 pproximates the value function of the control problem and identifies the p
 art of the state space likely to be visited by optimal trajectories. The l
 atter is important to avoid the curse of dimensionality associated with so
 lving these problems globally (that is, for all states). We, therefore, co
 nsider our approach an approximate semi-global method.<br />Our network de
 sign and the training problem leverage insights from optimal control theor
 y. We approximate the value function of the control problem using a neural
  network and use the Pontryagin maximum principle to express the optimal c
 ontrol (and therefore the sampling) in terms of the value function. Our tr
 aining loss consists of a weighted sum of the objective functional of the 
 control problem and penalty terms that enforce the Hamilton Jacobi Bellman
  equations along the sampled trajectories. Importantly, training is unsupe
 rvised in that it does not require solutions to the control problem.<br />
 In our numerical experiments, we compare our method to existing solvers fo
 r a more general class of semilinear PDEs.\n\nDomain: Computer Science and
  Applied Mathematics
END:VEVENT
END:VCALENDAR