BEGIN:VCALENDAR
METHOD:PUBLISH
PRODID:www-physics-nat-fau-eu//Events//EN
VERSION:2.0
BEGIN:VEVENT
SUMMARY:Promotionsvortrag Physik: Insights into physical states of loo
 p quantum gravity using techniques from machine learning
UID:040000008200E00074C5B7101A82E00800000000006A3F91265BDD010000000000
 00000010000000162F2A31C90BD443BF85C506E05FD0F2
DESCRIPTION:Ankündigung des Promotionsvortrags von: Herrn Waleed Sher
 if Loop quantum gravity seeks a quantum description of spacetime geome
 try\, but implementing its dynamics remains a major challenge. In the 
 canonical formulation\, physical states must satisfy quantum constrain
 ts. In the covariant formulation\, transition amplitudes are assembled
  from local spinfoam vertex amplitudes whose evaluation is computation
 ally demanding. In this talk\, I present a research program that uses 
 neural networks to approach both problems. For the graph-preserving ca
 nonical dynamics\, I reformulate the many weak constraint equations on
  a fixed graph as a single ground state search problem\, enabling one 
 to find approximate physical states using variational Monte Carlo meth
 ods and neural quantum states. Focusing on a 4-dimensional model with 
 Euclidean dynamics in the weak coupling limit\, I show how\, using thi
 s framework\, one can find approximate solutions and probe them using 
 geometric operators and correlation functions. In particular\, I consi
 der how ordering affects the approximate solutions found\, and establi
 sh connections with known kinematical vacua.For the covariant form of 
 the dynamics\, I investigate supervised neural surrogates for the Eucl
 idean Barrett-Crane vertex amplitude. The learning problem separates i
 nto identifying configurations with non-zero amplitudes and approximat
 ing their magnitudes. The results show that\, with the network archite
 ctures used\, support classification transfers beyond the training dat
 a used whereas magnitude regression reproduces the reference data but 
 does not perform well beyond it. I conclude by discussing the scope of
  these results and the extensions required to study more general canon
 ical dynamics and Lorentzian spinfoam models. (Vortrag auf Englisch) D
 em Vortrag schließt sich eine Diskussion von 15 Minuten an. Vortrag u
 nd Diskussion sind öffentlich. Diesen Verfahrensteilen folgt ein nich
 t öffentliches Rigorosum von 45 Minuten.
DTSTART:20261013T130000Z
DTEND:20261013T143000Z
LOCATION:SR 00.061\, ECAP\, Nikolaus-Fiebiger-Str. 2\, Erlangen
DTSTAMP:20261013T192555Z
END:VEVENT
END:VCALENDAR