BEGIN:VCALENDAR
VERSION:2.0
CALSCALE:GREGORIAN
PRODID:UW-Madison-Physics-Events
BEGIN:VEVENT
SEQUENCE:3
UID:UW-Physics-Event-5931
DTSTART:20200729T160000Z
DTEND:20200729T170000Z
DTSTAMP:20260415T042130Z
LAST-MODIFIED:20200729T215157Z
LOCATION:Please register for this online event: http://physicsmeetsml.
 org
SUMMARY:Discovering Symbolic Models in Physical Systems using Deep Lea
 rning\, Physics ∩ ML Seminar\, Shirley Ho\,  Flatiron Institute
DESCRIPTION:We develop a general approach to distill symbolic represen
 tations of a learned deep model by introducing strong inductive biases
 . We focus on Graph Neural Networks (GNNs). The technique works as fol
 lows: we first encourage sparse latent representations when we train a
  GNN in a supervised setting\, then we apply symbolic regression to co
 mponents of the learned model to extract explicit physical relations. 
 We find the correct known equations\, including force laws and Hamilto
 nians\, can be extracted from the neural network. We then apply our me
 thod to a non-trivial cosmology example—a detailed dark matter simul
 ation—and discover a new analytic formula that can predict the conce
 ntration of dark matter from the mass distribution of nearby cosmic st
 ructures. The symbolic expressions extracted from the GNN using our te
 chnique also generalized to out-of-distribution-data better than the G
 NN itself. Our approach offers alternative directions for interpreting
  neural networks and discovering novel physical principles from the re
 presentations they learn.
URL:https://www.physics.wisc.edu/events/?id=5931
END:VEVENT
END:VCALENDAR