A Review of the Security and Usability Of Differential Privacy

MATADEACU√ĎA, J. (2017). A Review of the Security and Usability Of Differential Privacy (MSc ASDF Dissertation). Edinburgh Napier University (Macfarlane, R., Aaby, P.).



The increasing trend for data collection in modern companies creates a challenge
for privacy protection. Cases where the advances in data processing have
been used to recover identifiable information from published, curated datasets
have started to disprove traditional methods of anonymisation. This, together
with the progressively tightening legal framework being expanded with regulations
like the General Data Protection Regulation (GDPR), which ultimate aim
is to protect the privacy of users, creates an urgent necessity for developing
more thorough and verifiable privacy preservation mechanisms.
In response to this problem, different lines of research have risen, with varied
propositions that seek to tackle this issue. Techniques like randomisation,
anonymisation, query auditing or generalisation have been proposed by academic
researchers to protect the privacy of the users. Inside the randomisation
category, a promising concept called differential privacy can be highlighted.
This approach proposes a formal definition of privacy, based on solid mathematical
foundations and supported by companies like Google orMicrosoft in the
late years.
This work aims to contribute to the further study and development of techniques
based on differential privacy. Using an open sourced project called Randomized
Aggregatable Privacy-Preserving Ordinal Response (RAPPOR), created
by Google researchers and incorporated into the Chrome web browser, a platform
for generating RAPPOR simulations from a given dataset is produced to
elaborate scenarios with different parameters that can be compared to evaluate
the effects of altering the input values on the recovered data.
The experimental results show how the increase of the privacy requirement of
the mechanism leads to producing noisier results, generating distributions increasingly
distant to the original data as the privacy guarantee becomes more
constraining. Also, it is observed how smaller populations suffer the effect of
high privacy requirements more than populations with a higher number of reports.
Finally, an additional experiment performed at the end of the project
allowed to appreciate how this loss of fidelity in the recovered data grows linearly
when the privacy guarantee is increased.
[Read More]


Areas of Expertise

Electronic information now plays a vital role in almost every aspect of our daily lives. So the need for a secure and trustworthy online infrastructure is more important than ever. without it, not only the growth of the internet but our personal interactions and the economy itself could be at risk.

Associated Projects

    Keywords: cyber security