BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Adversarial Explanations - You Shouldn't Trust Me: Learning Models
  Which Conceal Unfairness From Multiple Explanation Methods - Botty Dimano
 v (University of Cambridge)
DTSTART:20200609T121500Z
DTEND:20200609T131500Z
UID:TALK142618@talks.cam.ac.uk
CONTACT:Mateja Jamnik
DESCRIPTION:"ONLINE link.":https://teams.microsoft.com/l/meetup-join/19%3a
 meeting_ODE2MmI1ZGUtMzJhMi00ZGMyLWFlZGMtNWFkM2VhZjE1Mzc5%40thread.v2/0?con
 text=%7b%22Tid%22%3a%2249a50445-bdfa-4b79-ade3-547b4f3986e9%22%2c%22Oid%22
 %3a%22760af26a-1349-4870-a967-af40fbad85e9%22%7d\n\nTransparency of algori
 thmic systems has been discussed as a way for end-users and regulators to 
 develop appropriate trust in machine learning models. One popular approach
 \, LIME (Ribeiro et al 2016) even suggests that model explanations can ans
 wer the question ``Why should I trust you?'' Here we show a straightforwar
 d method for modifying a pre-trained model to manipulate the output of man
 y popular feature importance explanation methods with little change in acc
 uracy\, thus demonstrating the danger of trusting such explanation methods
 . We show how this explanation attack can mask a model's discriminatory us
 e of a sensitive feature\, raising strong concerns about using such explan
 ation methods to check model fairness.
LOCATION:Online on Teams
END:VEVENT
END:VCALENDAR