Version History: 

August 2016: Manuscript v1 (see files attached)

March 2017: Manuscript v2 (see files attached); Errata

April 2017: Published Version (in Tutorials on the Foundations of Cryptography; see Publisher's Version link and also SPRINGER 2017.PDF, below) 

Differential privacy is a theoretical framework for ensuring the privacy of individual-level data when performing statistical analysis of privacy-sensitive datasets. This tutorial provides an introduction to and overview of differential privacy, with the goal of conveying its deep connections to a variety of other topics in computational complexity, cryptography, and theoretical computer science at large. This tutorial is written in celebration of Oded Goldreich’s 60th birthday, starting from notes taken during a minicourse given by the author and Kunal Talwar at the 26th McGill Invitational Workshop on Computational Complexity [1].

Version History: a conference version of this paper appeared in the Proceedings of the 18th International Workshop on Randomization and Computation (RANDOM'14). Full version posted as ECCC TR14-076 and arXiv:1405.7028 [cs.CC].

We present an explicit pseudorandom generator for oblivious, read-once, width-3 branching programs, which can read their input bits in any order. The generator has seed length \(Õ(\log^3 n)\).The previously best known seed length for this model is \(n^{1/2+o(1)}\) due to Impagliazzo, Meka, and Zuckerman (FOCS ’12). Our work generalizes a recent result of Reingold, Steinke, and Vadhan (RANDOM ’13) for permutation branching programs. The main technical novelty underlying our generator is a new bound on the Fourier growth of width-3, oblivious, read-once branching programs. Specifically, we show that for any \(f : \{0, 1\}^n → \{0, 1\}\) computed by such a branching program, and \(k ∈ [n]\),

 \(\displaystyle\sum_{s⊆[n]:|s|=k} \big| \hat{f}[s] \big | ≤n^2 ·(O(\log n))^k\),

where \(\hat{f}[s] = \mathbb{E}_U [f[U] \cdot (-1)^{s \cdot U}]\) is the standard Fourier transform over \(\mathbb{Z}^n_2\). The base \(O(\log n)\) of the Fourier growth is tight up to a factor of \(\log \log n\).

Version History: Full version posted as arXiv:1604.05590 [cs.DS].

We present a new algorithm for locating a small cluster of points with differential privacy [Dwork, McSherry, Nissim, and Smith, 2006]. Our algorithm has implications to private data exploration, clustering, and removal of outliers. Furthermore, we use it to significantly relax the requirements of the sample and aggregate technique [Nissim, Raskhodnikova, and Smith, 2007], which allows compiling of “off the shelf” (non-private) analyses into analyses that preserve differential privacy.
 

Version History: Preliminary version posted as arXiv:1602.03090.

Hypothesis testing is a useful statistical tool in determining whether a given model should be rejected based on a sample from the population. Sample data may contain sensitive information about individuals, such as medical information. Thus it is important to design statistical tests that guarantee the privacy of subjects in the data. In this work, we study hypothesis testing subject to differential privacy, specifically chi-squared tests for goodness of fit for multinomial data and independence between two categorical variables.

We propose new tests for goodness of fit and independence testing that like the classical versions can be used to determine whether a given model should be rejected or not, and that additionally can ensure differential privacy. We give both Monte Carlo based hypothesis tests as well as hypothesis tests that more closely follow the classical chi-squared goodness of fit test and the Pearson chi-squared test for independence. Crucially, our tests account for the distribution of the noise that is injected to ensure privacy in determining significance.

We show that these tests can be used to achieve desired significance levels, in sharp contrast to direct applications of classical tests to differentially private contingency tables which can result in wildly varying significance levels. Moreover, we study the statistical power of these tests. We empirically show that to achieve the same level of power as the classical non-private tests our new tests need only a relatively modest increase in sample size.

Version History: Paper posted as arXiv:1609.04340 [cs.CR].

We provide an overview of the design of PSI (“a Private data Sharing Interface”), a system we are developing to enable researchers in the social sciences and other fields to share and explore privacy-sensitive datasets with the strong privacy protections of differential privacy.

Version History: Full version posted as https://arxiv.org/abs/1605.08294.

In this paper we initiate the study of adaptive composition in differential privacy when the length of the composition, and the privacy parameters themselves can be chosen adaptively, as a function of the outcome of previously run analyses. This case is much more delicate than the setting covered by existing composition theorems, in which the algorithms themselves can be chosen adaptively, but the privacy parameters must be fixed up front. Indeed, it isn’t even clear how to define differential privacy in the adaptive parameter setting. We proceed by defining two objects which cover the two main use cases of composition theorems. A privacy filter is a stopping time rule that allows an analyst to halt a computation before his pre-specified privacy budget is exceeded. A privacy odometer allows the analyst to track realized privacy loss as he goes, without needing to pre-specify a privacy budget. We show that unlike the case in which privacy parameters are fixed, in the adaptive parameter setting, these two use cases are distinct. We show that there exist privacy filters with bounds comparable (up to constants) with existing pri- vacy composition theorems. We also give a privacy odometer that nearly matches non-adaptive private composition theorems, but is sometimes worse by a small asymptotic factor. Moreover, we show that this is inherent, and that any valid privacy odometer in the adaptive parameter setting must lose this factor, which shows a formal separation between the filter and odometer use-cases.

Version History: Full version posted on Cryptology ePrint Archive, Report 2016/820.

Differential privacy is a mathematical definition of privacy for statistical data analysis. It guarantees that any (possibly adversarial) data analyst is unable to learn too much information that is specific to an individual. Mironov et al. (CRYPTO 2009) proposed several computa- tional relaxations of differential privacy (CDP), which relax this guarantee to hold only against computationally bounded adversaries. Their work and subsequent work showed that CDP can yield substantial accuracy improvements in various multiparty privacy problems. However, these works left open whether such improvements are possible in the traditional client-server model of data analysis. In fact, Groce, Katz and Yerukhimovich (TCC 2011) showed that, in this setting, it is impossible to take advantage of CDP for many natural statistical tasks.

Our main result shows that, assuming the existence of sub-exponentially secure one-way functions and 2-message witness indistinguishable proofs (zaps) for NP, that there is in fact a computational task in the client-server model that can be efficiently performed with CDP, but is infeasible to perform with information-theoretic differential privacy.

Version History: Full version posted as arXiv:1504.07553.

We prove new upper and lower bounds on the sample complexity of \((\varepsilon, \delta)\) differentially private algorithms for releasing approximate answers to threshold functions. A threshold function \(c_x\) over a totally ordered domain \(X\) evaluates to \(c_x(y)=1 \) if \(y \leq {x}\), and evaluates to \(0\) otherwise. We give the first nontrivial lower bound for releasing thresholds with \((\varepsilon, \delta)\) differential privacy, showing that the task is impossible over an infinite domain \(X\), and moreover requires sample complexity \(n \geq \Omega(\log^* |X|)\), which grows with the size of the domain. Inspired by the techniques used to prove this lower bound, we give an algorithm for releasing thresholds with \(n ≤ 2^{(1+o(1)) \log^∗|X|}\) samples. This improves the previous best upper bound of \(8^{(1+o(1)) \log^∗ |X|}\)(Beimel et al., RANDOM ’13).

Our sample complexity upper and lower bounds also apply to the tasks of learning distri- butions with respect to Kolmogorov distance and of properly PAC learning thresholds with differential privacy. The lower bound gives the first separation between the sample complexity of properly learning a concept class with \((\varepsilon, \delta)\) differential privacy and learning without privacy. For properly learning thresholds in \(\ell\) dimensions, this lower bound extends to \(n ≥ Ω(\ell·\log^∗ |X|)\).

To obtain our results, we give reductions in both directions from releasing and properly learning thresholds and the simpler interior point problem. Given a database \(D\) of elements from \(X\), the interior point problem asks for an element between the smallest and largest elements in \(D\). We introduce new recursive constructions for bounding the sample complexity of the interior point problem, as well as further reductions and techniques for proving impossibility results for other basic problems in differential privacy.

For Boolean functions computed by read-once, depth-D circuits with unbounded fan-in over the de Morgan basis, we present an explicit pseudorandom generator with seed length \(\tilde{O}(\log^{D+1} n)\). The previous best seed length known for this model was \(\tilde{O}(\log^{D+4} n)\), obtained by Trevisan and Xue (CCC ‘13) for all of AC0 (not just read-once). Our work makes use of Fourier analytic techniques for pseudorandomness introduced by Reingold, Steinke, and Vadhan (RANDOM ‘13) to show that the generator of Gopalan et al. (FOCS ‘12) fools read-once AC0. To this end, we prove a new Fourier growth bound for read-once circuits, namely that for every \(F : \{0,1\}^n\rightarrow \{0,1\}\) computed by a read-once, depth-\(D\) circuit,

\(\left|\hat{F}[s]\right| \leq O\left(\log^{D-1} n\right)^k,\)

where \(\hat{F}\) denotes the Fourier transform of \(F\) over \(\mathbb{Z}_2^n\).

The privacy risks inherent in the release of a large number of summary statistics were illustrated by Homer et al. (PLoS Genetics, 2008), who considered the case of 1-way marginals of SNP allele frequencies obtained in a genome-wide association study: Given a large number of minor allele frequencies from a case group of individuals diagnosed with a particular disease, together with the genomic data of a single target individual and statistics from a sizable reference dataset independently drawn from the same population, an attacker can determine with high confidence whether or not the target is in the case group.

In this work we describe and analyze a simple attack that succeeds even if the summary statistics are significantly distorted, whether due to measurement error or noise intentionally introduced to protect privacy. Our attack only requires that the vector of distorted summary statistics is close to the vector of true marginals in \(\ell_1 \)norm. Moreover, the reference pool required by previous attacks can be replaced by a single sample drawn from the underlying population.

The new attack, which is not specific to genomics and which handles Gaussian as well as Bernouilli data, significantly generalizes recent lower bounds on the noise needed to ensure differential privacy (Bun, Ullman, and Vadhan, STOC 2014; Steinke and Ullman, 2015), obviating the need for the attacker to control the exact distribution of the data.