A firm foundation for private data analysis

By Cynthia Dwork. 2011

941 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Sociology and Political Science, Privacy & Security

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 Sociology and Political Science

In Privacy & Security

Citations in the corpus

Semi-supervised Knowledge Transfer for Deep Learning from Private Training Data (2016, background): “Differential privacy (Dwork et al., 2006b; Dwork, 2011) It will be useful to deﬁne the privacy loss and the privacy loss random variable .”
Generative Models for Effective ML on Private, Decentralized Datasets (2019, background): “Differential privacy (DP) (Dwork et al., 2006; Dwork, 2011; Dwork & Roth, 2014) provides a well-tested formalization for the release of information derived from private data.”
Differential Privacy Has Disparate Impact on Model Accuracy (2019, positive): “We use the standard definitions [12–14].”
The Power of the Hybrid Model for Mean Estimation (2018, methodology): “The classic DP algorithm for this problem was designed for the TCM and analyzed in [23].”
Differentially-Private "Draw and Discard" Machine Learning (2018, background): “Differential privacy (DP) [13] has become the de-facto standard for privacy-preserving data analysis [15], [12], [17].”
Privacy Loss in Apple's Implementation of Differential Privacy on MacOS 10.12 (2017, background): “Although the choice of ϵ is typically treated as a social choice by the theoretical computer scientists [6], it is of crucial importance in practical deployments, as the meaning of a privacy risk of exp ( 1 ) vs exp ( 50 ) is radically different.”
BLENDER: Enabling Local Search with a Hybrid Differential Privacy Model (2017, background): “Privacy parameters, and δ: Academic literature on differential privacy views the selection of the parameter as a “social question” [9] and thus”
A Comprehensive Comparison of Multiparty Secure Additions with Differential Privacy (2015, background): “Privacy of data subjects is defined by differential privacy, which is the state-of-the-art privacy notion [21], [22], [36] that assures a strong and provable privacy guarantee for aggregated data.”
RAPPOR: Randomized Aggregatable Privacy-Preserving Ordinal Response (2014, methodology): “For our analysis, we adopt the rigorous notion of privacy, differential privacy, which was introduced by Dwork et al [12] and has been widely adopted [10].”
Boosting and Differential Privacy (2010, background): “See [7], [ 6 ] for more properties and results concerning differential privacy.”
Federated Learning: Challenges, Methods, and Future Directions (2019, background): “Among these various privacy approaches, differential privacy [32, 33, 34] is most widely used due to its strong information theoretic guarantees, algorithmic simplicity, and relatively small systems overhead.”
Learning Differentially Private Language Models Without Losing Accuracy (2017, background): “Differential privacy (DP) [12, 9, 11] However, for problems like language modeling, protecting individual examples is insufﬁcient—each typed word makes its own contribution to the RNN’s training objective, so one user may contribute many thousands of examples to the training data.”
Preserving Statistical Validity in Adaptive Data Analysis (2014, background): “See [14] for a short survey and [19] for a textbook introduction to differential privacy.”
Privacy via the Johnson-Lindenstrauss Transform (2012, methodology): “We adopt a rigorous approach to privacy introduced by Dwork et al. [12], which has gained widespread recognition in recent years (see a survey [10]), and has been used to demonstrate the feasibility of privacy-preserving data releases [3, 20, 23, 24].”
The Johnson-Lindenstrauss Transform Itself Preserves Differential Privacy (2012, background): “In addition, the classical Randomized Response (see [War65]) preserves diﬀerential privacy as discussed in recent surveys [DS10, Dwo11].”
Characterizing the sample complexity of private learners (2013, methodology): “Dwork et al. [12] show how to privately boost accuracy, that is, given a private learning algorithms that have a big classi.cation error, they produce a private learning algorithm with small error.”
Adversarial machine learning (2019, methodology): “The most common approach to achieving differential privacy is to add noise to the target non-private mechanism’s response S(D), typically Laplace noise [22, 19, 6, 20, 56] but more generally an exponential mechanism is used to ran-”