BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Machine Learning Reading Group @ CUED
SUMMARY:Are we making progress in unlearning?  - Eleni Tri
 antafillou (Google DeepMind)
DTSTART;TZID=Europe/London:20241113T110000
DTEND;TZID=Europe/London:20241113T123000
UID:TALK224401AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/224401
DESCRIPTION:Machine unlearning is the problem of removing the 
 influence of a subset of training data from machin
 e learning models. This problem is enjoying increa
 sed attention recently due to excitement around us
 ing this technology to remove outdated\, harmful\,
  private or no-longer-permissible data from traine
 d models\, in order to increase their accuracy\, s
 afety\, or protect privacy. A straightforward solu
 tion to the problem is to remove the unwanted data
  from the training set and retrain a new model fro
 m scratch. However\, that solution is inefficient 
 and impractical\, especially in the era of increas
 ingly large models that are increasingly expensive
  to train. Can we instead cause models to "forget"
  a subset of their training data after the fact? W
 hile this problem has close ties to many research 
 areas\, including continual learning\, transfer le
 arning and privacy\, machine unlearning is still a
 t its infancy\, with many open questions remaining
 \, both in how to evaluate success as well as how 
 to improve upon existing methods. In this talk\, I
  will discuss recent progress and challenges remai
 ning\, highlighting open questions and important d
 irections for the community.
LOCATION:Cambridge University Engineering Department\, CBL 
 Seminar room BE4-38.
CONTACT:Xianda Sun
END:VEVENT
END:VCALENDAR