BEGIN:VCALENDAR
VERSION:2.0
CALSCALE:GREGORIAN
PRODID:UW-Madison-Physics-Events
BEGIN:VEVENT
SEQUENCE:2
UID:UW-Physics-Event-8933
DTSTART:20241007T213000Z
DTEND:20241007T223000Z
DTSTAMP:20260408T073823Z
LAST-MODIFIED:20240929T201752Z
LOCATION:5310 Chamberlin Hall
SUMMARY:Machine Learning Symmetries in Physics from First Principles\,
  Theory Seminar (High Energy/Cosmology)\, Konstantin Matchev\, Univers
 ity of Florida
DESCRIPTION:Symmetries are the cornerstones of modern theoretical phys
 ics\, as they imply fundamental conservation laws. The recent boom in 
 AI algorithms and their successful application to high-dimensional lar
 ge datasets from all aspects of life motivates us to approach the prob
 lem of discovery and identification of symmetries in physics as a mach
 ine-learning task. In a series of papers\, we have developed and teste
 d a deep-learning algorithm for the discovery and identification of th
 e continuous group of symmetries present in a labeled dataset. We use 
 fully connected neural network architectures to model the symmetry tra
 nsformations and the corresponding generators. Our proposed loss funct
 ions ensure that the applied transformations are symmetries and that t
 he corresponding set of generators is orthonormal and forms a closed a
 lgebra. One variant of our method is designed to discover symmetries i
 n a reduced-dimensionality latent space\, while another variant is cap
 able of obtaining the generators in the canonical sparse representatio
 n. Our procedure is completely agnostic and has been validated with se
 veral examples illustrating the discovery of the symmetries behind the
  orthogonal\, unitary\, Lorentz\, and exceptional Lie groups.
URL:https://www.physics.wisc.edu/events/?id=8933
END:VEVENT
END:VCALENDAR