BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Topics in differential privacy: optimal noise and record perturbat
 ion baseddata sets - Jordi Soria-Comas (Universitat Rovira i Virgili )
DTSTART:20161202T140000Z
DTEND:20161202T150000Z
UID:TALK69403@talks.cam.ac.uk
CONTACT:INI IT
DESCRIPTION:<span>We explore two different aspects of differential privacy
 . First we explore the optimality&nbsp\;of noise distributions in noise ad
 dition. In particular\, we show that the Laplace distribution&nbsp\;is nea
 rly optimal in the univariate case\, but not in the multivariate case. Opt
 imal&nbsp\;distributions are described. Then we explore the generation of 
 differentially private data sets&nbsp\;via perturbative masking of the ori
 ginal records. This approach is<br>remarkably more efficient than&nbsp\;hi
 stogram-based approaches but a naive application of it may completely dama
 ge the data utility.&nbsp\;In particular\, we analyze the use of microaggr
 egation to reduce the<br>&nbsp\;sensitivity and\, thus\, the&nbsp\;amount 
 of noise required to attain differential privacy.</span>
LOCATION:Seminar Room 2\, Newton Institute
END:VEVENT
END:VCALENDAR