Title:   Survey on privacy in data publishing
Speaker: Yannis Katsis (UCSD), Avinash Vyas (UCSD)

Abstract:

In data publishing scenarios, data owners are willing to share
part of their data, while maintaining some other part of them private.
For example health insurance companies would like to publish the birth date,
zip code and condition of patients to aid statistical analyses but
they do not want to reveal the identity of the individuals.

Conventional access control methods used to secure computer systems
are not suited to protect data in these scenarios, since even if a
query against the database conceals the private data, the attacker
may combine results of different queries or also use his own
background knowledge to infer the hidden data. For instance, a
researcher retrieved the zip code and age of the governor of MA from
the voter registration list and used it to reveal his condition
from his health insurance database.

In this talk we present the solutions proposed in the literature to
guarantee privacy. These can be distinguished in two large categories:
The first approach, stemming from practical needs,
focuses on generalizing the data before publishing them to reduce the
possibility of privacy breaches. The second on the other hand,
does not obfuscate the data but instead uses probability theory to check
whether the published views of a database cause a data leak.