BEGIN:VCALENDAR
VERSION:2.0
CALSCALE:GREGORIAN
PRODID:UW-Madison-Physics-Events
BEGIN:VEVENT
SEQUENCE:3
UID:UW-Physics-Event-6873
DTSTART:20220413T160000Z
DTEND:20220413T171500Z
DTSTAMP:20260414T153413Z
LAST-MODIFIED:20220406T211500Z
LOCATION:Chamberlin 5280 (Zoom link also available for online particip
 ants who signed up on our mailing list)
SUMMARY:A duality connecting neural network and cosmological dynamics\
 , Physics ∩ ML Seminar\, Sven Krippendorf\, Ludwig Maximilian Univer
 sity
DESCRIPTION:We demonstrate that the dynamics of neural networks traine
 d with gradient descent and the dynamics of scalar fields in a flat\, 
 vacuum energy dominated Universe are structurally profoundly related. 
 This duality provides the framework for synergies between these system
 s\, to understand and explain neural network dynamics and new ways of 
 simulating and describing early Universe models. Working in the contin
 uous-time limit of neural networks\, we analytically match the dynamic
 s of the mean background and the dynamics of small perturbations aroun
 d the mean field\, highlighting potential differences in separate limi
 ts. We perform empirical tests of this analytic description and quanti
 tatively show the dependence of the effective field theory parameters 
 on hyperparameters of the neural network. As a result of this duality\
 , the cosmological constant is matched inversely to the learning rate 
 in the gradient descent update.
URL:https://www.physics.wisc.edu/events/?id=6873
END:VEVENT
END:VCALENDAR