Privacy against statistical inference

By Flávio du Pin Calmon, Nadia Fawaz. Allerton Conference on Communication, Control, and Computing, 2012

370 citations recorded by the source corpus.

We propose a general statistical inference framework to capture the privacy threat incurred by a user that releases data to a passive but curious adversary, given utility constraints. We show that applying this general framework to the setting where the adversary uses the self-information cost function naturally leads to a non-asymptotic information-theoretic approach for characterizing the best achievable privacy subject to utility constraints. Based on these results we introduce two privacy metrics, namely average information leakage and maximum information leakage. We prove that under both metrics the resulting design problem of finding the optimal mapping from the user's data to a privacy-preserving output can be cast as a modified rate-distortion problem which, in turn, can be formulated as a convex program. Finally, we compare our framework with differential privacy.

Subfields: Artificial Intelligence, Electrical and Electronic Engineering

View Paper: https://www.semanticscholar.org/paper/592b28828c0a528396e41110dcafb974194beb83

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Electrical and Electronic Engineering

Citations in the corpus

Learning Generative Adversarial RePresentations (GAP) under Fairness and Censoring Constraints (2019, methodology): “Thus, under the log-loss in (15), GAP is equivalent to using MI as the censoring metric Calmon and Fawaz [2012]. The 0-1 loss captures a strong guessing adversary; in contrast, log-loss or information-loss models a belief refining adversary.”
A systems and control perspective of CPS security (2019, background): “3 The latter is called inference attacks which are to infer the private information of a system using the access to some potentially legitimate parts of the system [80]. may create.”
Generative Adversarial Privacy: A Data-Driven Approach to Information-Theoretic Privacy (2018, methodology): “Context-aware privacy notions fall under the domain of information theoretic (IT) privacy; see, for example, [4, 5, 25, 26, 27].”
Toward Information Privacy for the Internet of Things: A Nonparametric Learning Approach (2018, methodology): “Our main contributions are as follows: 1) We adopt the concept of -information privacy from [28], and show that a sufficient condition to achieve information privacy when the private hypothesis is binary, is to ensure that the average of the Type I and II detection error probabilities are large.”
Privacy and Utility Preserving Sensor-Data Transformations (2019, methodology): “In this paper, we present mechanisms for transforming time-windows of sensor data to preserve privacy and utility during information disclosure [14, 15, 16] to a honest-but-curious app running on users’ devices.”
Censored and Fair Universal Representations using Generative Adversarial Models (2019, methodology): “Thus, under the log-loss in (22), the CFUR formulation is equivalent to using MI as the privacy metric [Calmon and Fawaz, 2012].”
A rate-disortion perspective on local differential privacy (2014, background): “From the distortion constraint, we can see that the set of channels which guarantees distortion D universally over the class P is a piecewise linear region.”
Technical Privacy Metrics (2015, background): “It quantifies the maximum amount of information about private events or data X ∗ that can be gained by an adversary observing a single output y, where the maximum is taken over all possible outputs [37].”
AttriGuard: A Practical Defense Against Attribute Inference Attacks via Adversarial Machine Learning (2018, methodology): “One category of defenses (e.g., [13– 16]) against general inference attacks leverage game theory.”
On the Relation Between Identifiability, Differential Privacy, and Mutual-Information Privacy (2016, background): “When mutual information is used as the privacy notion, the problem of finding the optimal tradeoff between privacy and distortion can usually be formulated as a rate–distortion problem in the field of information theory (see [32] for an introduction) [25], [27]–[31].”
On the Robustness of Information-Theoretic Privacy Measures and Mechanisms (2018, background): “Of particular relevance for the present papers are the works of Rebollo-Monedero et al. [49], Sankar et al. [9], Calmon and Fawaz [8], Asoodeh et al. [10], and references therein.”
Utility-Privacy Tradeoffs in Databases: An Information-Theoretic Approach (2011, background): “In addition, there has been recent work comparing differential privacy guarantee with Renyi entropy [15] and Shannon entropy [16].”
Context-Aware Generative Adversarial Privacy (2017, methodology): “Thus, under the log-loss in (6), GAP is equivalent to using MI as the privacy metric [38].”
On the relation between identifiability, differential privacy, and mutual-information privacy (2014, neutral): “In this paper, we consider a form of identifiability that prevents the input database from being identified given the observation of the output database, prior knowledge and the mechanism.”