BEGIN:VCALENDAR
VERSION:2.0
CALSCALE:GREGORIAN
PRODID:UW-Madison-Physics-Events
BEGIN:VEVENT
SEQUENCE:2
UID:UW-Physics-Event-6487
DTSTART:20210811T160000Z
DTEND:20210811T171500Z
DTSTAMP:20260414T115043Z
LAST-MODIFIED:20210908T050523Z
LOCATION:Online Seminar: Please sign up for our mailing list at www.ph
 ysicsmeetsml.org for zoom link
SUMMARY:Interpretable Deep Learning for Physics\, Physics ∩ ML Semin
 ar\, Miles Cranmer\, Princeton University
DESCRIPTION:If we train a neural network on some dynamical system in s
 ome region of phase space\, and it learns a way to execute the dynamic
 s more efficiently than a handwritten code\, how do we distill physica
 l insight from the learned model? In this talk\, I will argue that sym
 bolic learning should play a major role in the process of interpreting
  a machine learning model for physical systems. I will discuss our gen
 eric method for converting a neural network that has been trained on a
  physical system into a symbolic model\, via genetic algorithm-based s
 ymbolic regression. One of the problems with this process is working w
 ith the fact that neural networks have high-dimensional latent spaces\
 , and genetic algorithms scale poorly with the number of features. To 
 work around this issue\, I’ll then introduce our “Disentangled Spa
 rsity Network\,” which encourages a neural network to learn an easy-
 to-interpret representation. I will then share several recent applicat
 ions of our techniques to real physical systems\, and the various insi
 ghts we have discovered and rediscovered.
URL:https://www.physics.wisc.edu/events/?id=6487
END:VEVENT
END:VCALENDAR