Ensuring data confidentiality through mathematical noise

In 2013, Peter Millican, a computational linguistics researcher, managed to demonstrate through literary style analysis that Robert Galbraith, the alleged author of the hero detective series Cormoran Strike, was in fact J.K. Rowling. This anecdote illustrates how privacy is not so much about anonymity as it is about the impossibility of cross-checking data.

Just because an individual responds to a survey anonymously doesn't mean that he or she won't be identified. How then to guarantee the confidentiality of the data?  Whether it is large datasets or very specific local data, the problem of confidentiality remains.

One solution is to add noise or "fuzziness" to the data. For example "Location guard" is a navigation application that hides the exact location of a user. So a tourist in Paris might not want to reveal that he is visiting the Moulin Rouge cabaret, so the application will locate him in a larger area, including for example the Notre-Dame cathedral or the Louvre. This technique is called "local differential privacy".

Style privacy and meaning preservation

Natasha Fernandes' work has taken local differential privacy a step further with metrics by applying it to text documents in such a way that a computer can understand what the text is about without being able to identify its author.

"To ensure that machine learning cannot detect a given author's writing style, we added a distance metric and mathematically showed that it was possible to retain the document's classification while protecting against identification." 

For example, this blurring property would ensure the anonymity of a whistleblower, a client, or any author of textual communication.

One of Natasha Fernandes' main goals is to make it easier to understand privacy risks.

"If people need a math degree to understand the degree of risk of a privacy breach, it doesn't inspire the confidence to feel comfortable enough to share their data,"

For the full article:

How maths makes online privacy safer
https://www.inria.fr/fr/mathematiques-confidentialite-securite-vie-privee

How JK Rowling was unmasked - https://www.bbc.com/news/entertainment-arts-23313074

Visit inria.fr

Add to my playlists