清华-鲁汶合作项目报告会: Differential Privacy and Differential Identifiability
时间：6月24日上午10:00-11:00
地点：东主楼10-302

1. 题目：Differential Privacy and its Deficiencies
    报告人：万志国
    时间：6月24日上午10:00-10:30
    地点：东主楼10-302
摘要：
Differential privacy, since it was proposed by Dwork from Microsoft
Research, has been recognized as the most up-to-date
privacy-preserving technique for databases. Recent years has witnessed
numerous research results in the database literature. However, it was
also identified that differential privacy has its own deficiencies in
protecting user privacy. In this talk, I will provide an overview
about differential privacy and discuss its deficiencies, and possible
future research directions are also presented in the talk.

2. 题目："Differential Identifiability for Quasi-Identifiers"
    报告人： 鲁汶大学研究员 Ero Balsa
    时间：6月24日上午10:30-11:00
    地点：东主楼10-302
摘要：
The goal of a privacy-preserving statistical database is to provide
analysts with accurate information about the population as a whole
while protecting the privacy of each of the individuals participating
in the database. In her paper "Differential Privacy" [1], Dwork
proposed a measure of the increased privacy risk of an individual
participating on a database. Moreover, she provided a mechanism based
on the addition of random noise to provide epsilon-differential
privacy. However, choosing the appropriate epsilon, i.e., according to
a certain privacy policy, is a non-trivial task. Lee and Clifton
proposed "Differential Identifiability" (DI) [2], a guideline to
choose epsilon such that the probability of identification of an
individual on a database is below a certain threshold. In this talk, I
will explain Lee and Clifton's concept of DI and its constraints and
limitations. Moreover, I will argue about the dependence of DI on the
existence of identifiers and introduce a generalisation of this
concept to deal with quasi-identifiers. This is joint work with Wan
Zhiguo.

[1] Dwork, Cynthia. "Differential privacy." Automata, languages and
programming. Springer Berlin Heidelberg, 2006. 1-12.
[2] Lee, Jaewoo, and Chris Clifton. "Differential identifiability."
Proceedings of the 18th ACM SIGKDD international conference on
Knowledge discovery and data mining. ACM, 2012.