BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Insights from quantum field theory and AdS/CFT for machine learnin
 g - Johanna Erdmenger (Julius-Maximilians-Universität Würzburg)
DTSTART:20230914T080000Z
DTEND:20230914T093000Z
UID:TALK205036@talks.cam.ac.uk
DESCRIPTION:In recent years\, important new relations between information 
 theory and quantum gravity have been found in the context of generalising 
 the AdS/CFT correspondence. Information measures have been used to describ
 e the quantum nature of black holes. Conversely\, concepts such as the ren
 ormalisation group and relative entropy are used towards quantifying the l
 earning ability of neural networks\, and hyperbolic geometry gives rise to
  neural networks with improved properties. &nbsp\;I will review these deve
 lopments based on three examples: 1) Geometric phases and symplectic forms
  characterise Hilbert spaces and hidden information on both sides of the A
 dS/CFT duality\; &nbsp\;2) The relative entropy or Kullback-Leibler diverg
 ence shows similar behaviour for the Ising model under RG transformations 
 and for feedforward neural networks as function of depth\; 3) Recent progr
 ess towards establishing a holographic duality for regular tilings of hype
 rbolic space may also have implications for hyperbolic graph neural networ
 ks.\nBased on arXiv 2107.06898\,&nbsp\; 2205.05693 and 2306.00055&nbsp\;
LOCATION:Seminar Room 2\, Newton Institute
END:VEVENT
END:VCALENDAR