BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Capacity and errors in classification of object manifolds - Uri Co
 hen (Hebrew University of Jerusalem)
DTSTART:20210603T090000Z
DTEND:20210603T101500Z
UID:TALK160870@talks.cam.ac.uk
CONTACT:Yul Kang
DESCRIPTION:What makes a good object representation? How do object represe
 ntations change along biological or artificial hierarchies? I will introdu
 ce a measure called classification capacity and argue it quantifies the go
 odness of a neural representation with respect to manifold classification.
  A theoretical analysis using tools from statistical physics relates this 
 capacity to the geometry of object manifolds\, thus augmenting the computa
 tional definition with an intuitive geometric perspective. For artificial 
 hierarchies\, theory was used to describe changes in representation along 
 deep convolutional neural networks. For noisy biological hierarchies\, rel
 ating capacity to generalization error with respect to neural noise allows
  one to correctly interpret experimental results.\n\n\nJoin via Zoom:\nhtt
 ps://us02web.zoom.us/j/84958321096?pwd=dFpsYnpJYWVNeHlJbEFKbW1OTzFiQT09\n\
 nMeeting ID: 849 5832 1096\n\nPasscode: 506576
LOCATION:Online on Zoom
END:VEVENT
END:VCALENDAR