BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:KNG - A New Mechanism for Data Privacy
DTSTART:20201016T151500
DTEND:20201016T163000
DTSTAMP:20260510T020832Z
UID:d402857a3a4454eef86a8ea30c85a7c9e74c7360a3e30a99fe98e65f
CATEGORIES:Conferences - Seminars
DESCRIPTION:Matthew Reimherr\, Penn State University\n\nIn this presentati
 on we consider a new mechanism for achieving Differential Privacy called K
 NG.\nTwo of the most popular methods\, the exponential mechanism and objec
 tive perturbation\, have had great\nsuccess\, but still have several drawb
 acks that can range from minor to severe depending on the setting.\n\nRece
 ntly it was shown that the exponential mechanism is not asymptotically eff
 icient\, introducing too much\nnoise\, and thus reducing statistical utili
 ty quite broadly. Conversely\, objective perturbation enjoys excellent\nut
 ility\, but can be difficult to generalize and requires strong structural 
 assumptions.\n\nWe show how our new approach\, KNG\, assuages nearly all o
 f these issues\; it is nearly as easy to implement as the exponential\nmec
 hanism\, but has much better asymptotic properties. We highlight how KNG a
 grees with well known mechanisms in simpler settings\, while using its fra
 mework to develop new privacy tools in more complicated\nsettings such as 
 linear and quantile regression.\n 
LOCATION:zoom https://epfl.zoom.us/j/82137336037
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR