Challenging Differential Privacy:The Case of Non-interactive Mechanisms

Abstract

To cite this version:Raghavendran Balu, Teddy Furon, Sébastien Gambs. Abstract. In this paper, we consider personalized recommendation systems in which before publication, the profile of a user is sanitized by a non-interactive mechanism compliant with the concept of differential privacy. We consider two existing schemes offering a differentially private representation of profiles: BLIP (BLoom-and-flIP) and JLT (JohnsonLindenstrauss Transform). For assessing their security levels, we play the role of an adversary aiming at reconstructing a user profile. We compare two inference attacks, namely single and joint decoding. The first one decides of the presence of a single item in the profile, and sequentially explores all the item set. The latter strategy decides whether a subset of items is likely to be the user profile, and considers all the possible subsets. Our contributions are a theoretical analysis as well as a practical implementation of both attacks, which were evaluated on datasets of real user profiles. The results obtained clearly demonstrates that joint decoding is the most powerful attack, while also giving useful insights on how to set the differential privacy parameter ǫ.

Sanitization of Call Detail Records via Differentially-Private Bloom Filters

“…In that case, the total privacy budget will be 2 instead of . Alternatively, these cardinalities can be estimated directly from the flipped Bloom filters with no cost to the privacy budget using the method introduced recently by Balu, Furon and Gambs in [2]. This method defines a probabilistic relation between the number of items in a set and the number of bits in its Bloom filter, before extending this relation to the flipped Bloom filter.…”

Privacy-preserving Wi-Fi Analytics

“…They have also shown that even after a Bloom filter is flipped in such a way, it is still possible to extract some utility from it such as approximating the similarity between two profiles represented as sets. In a similar line of work Balu, Furon, and Gambs [8], and later Alaggan, Gambs, Matwin, and Tuhin [4] created techniques to estimate the set size and the size of intersection between two sets, given only their corresponding BLIPs. The application considered in [4] was the analysis of mobility patterns using mobile phone usage data (such as Call Detail Records collected by telecom operators).…”

“…The quantification of utility depends what is done with the output of the algorithm. This paper (in particular Section 4), along with the unifying framework of [5], encompass the prior works of [3,4,8] and many others.…”

“…The previous bounds in [3,4,8] apply automatically for their respective applications once we substitute the correct value of the flipping probability µ 0 = 1/2 − η d /2, in which d is the number of intrusions including the output itself. In particular, for [5] all upper and lower bounds after d intrusions follow by replacing η by η d .…”

Secure Two-Party Protocol for Privacy-Preserving Classification via Differential Privacy

“…According to them, Information gain is the most sensitive to noise and max operator function is least sensitive to noise. Balu et al[5] introduces a recommender system that uses differential privacy in a non-interactive manner to sanitize the user profile before publishing. To achieve privacy they use two differentially-private non-interactive mechanisms for profile representation, Bloom-and-Flip (BLIP)[2] and Johnson-Lindenstrauss Transform (JLT)[33].…”