BEGIN:VCALENDAR
VERSION:2.0
CALSCALE:GREGORIAN
PRODID:UW-Madison-Physics-Events
BEGIN:VEVENT
SEQUENCE:2
UID:UW-Physics-Event-7982
DTSTART:20230111T170000Z
DTEND:20230111T181500Z
DTSTAMP:20260417T012332Z
LAST-MODIFIED:20230110T151232Z
LOCATION:Online Seminar: Please sign up for our mailing list at www.ph
 ysicsmeetsml.org for zoom link. We will also livestream the talk in Ch
 amberlin 5280.
SUMMARY:Implicit Regularization in Quantum Tensor Networks\, Physics �
 �� ML Seminar\, Nadav Cohen\, Tel Aviv University
DESCRIPTION:The mysterious ability of neural networks to generalize is
  believed to stem from an implicit regularization\, a tendency of grad
 ient-based optimization to fit training data with predictors of low �
 �complexity.” Despite vast efforts\, a satisfying formalization of t
 his intuition is lacking. In this talk I will present a series of work
 s theoretically analyzing the implicit regularization in quantum tenso
 r networks\, known to be equivalent to certain (non-linear) neural net
 works. Through dynamical characterizations\, I will establish an impli
 cit regularization towards low tensor ranks\, different from any type 
 of norm minimization\, in contrast to prior beliefs. I will then discu
 ss implications of this finding to both theory (potential explanation 
 for generalization over natural data) and practice (compression of neu
 ral network layers\, novel regularization schemes). An underlying them
 e of the talk will be the potential of quantum tensor networks to unra
 vel mysteries behind deep learning.
\n
\nWorks covered in the talk wer
 e in collaboration with Sanjeev Arora\, Wei Hu\, Yuping Luo\, Asaf Mam
 an and Noam Razin.
URL:https://www.physics.wisc.edu/events/?id=7982
END:VEVENT
END:VCALENDAR