BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Gold Doesn't Always Glitter: Spectral Removal of Linear and Nonlin
 ear Guarded Attribute Information - Yftah Ziser\, University of Edinburgh
DTSTART:20220616T100000Z
DTEND:20220616T110000Z
UID:TALK175718@talks.cam.ac.uk
CONTACT:Marinela Parovic
DESCRIPTION:We describe a simple and effective method (Spectral Attribute 
 removaL\; SAL) to remove guarded information from neural representations. 
 Our method uses singular value decomposition and eigenvalue decomposition 
 to project the input representations into directions with reduced covarian
 ce with the guarded information rather than maximal covariance\, as normal
 ly\, these factorization methods are used. We begin with linear informatio
 n removal and proceed to generalize our algorithm to the case of nonlinear
  information removal using kernels. Our experiments demonstrate that our a
 lgorithm retains better main task performance after removing the guarded i
 nformation compared to previous methods. In addition\, our experiments dem
 onstrate that we need a relatively small amount of guarded attribute data 
 to remove information about these attributes\, which lowers the exposure t
 o such possibly sensitive data and fits better low-resource scenarios.
LOCATION:https://cam-ac-uk.zoom.us/j/97599459216?pwd=QTRsOWZCOXRTREVnbTJBd
 XVpOXFvdz09
END:VEVENT
END:VCALENDAR