BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Learning Symmetries in Neural Networks - James Allingham and Bruno
  Mlodozeniec  
DTSTART:20240221T110000Z
DTEND:20240221T123000Z
UID:TALK212554@talks.cam.ac.uk
CONTACT:Isaac Reid
DESCRIPTION:While the importance of incorporating symmetries into NNs has 
 been well understood for some time\, until recently\, the standard approac
 h has been to incorporate these inductive biases into the architecture of 
 the neural networks (e.g.\, CNNs have translation\, or even rotation\, inv
 ariance). Unfortunately\, this requires prior knowledge of which symmetrie
 s are present in the dataset. To motivate why this might be a problem\, co
 nsider a model trained to classify digits. If this model is fully rotation
 ally invariant\, it cannot distinguish between some 6s and 9s. But\, there
  is certainly some rotation invariance due to natural variations in handwr
 iting. Thus\, we need to learn how invariant our classifier should be to r
 otations. This reading group will explore methods to learn such invariance
 s directly from the data. We will tackle invariance learning in both the s
 upervised and unsupervised settings.\n\nRequired reading: None
LOCATION:Cambridge University Engineering Department\, CBL Seminar room BE
 4-38.
END:VEVENT
END:VCALENDAR