Taking a Lesson from Quantum Particles for Statistical Data Privacy

Farhad Farokhi

REVIEW 4 major objections 4 minor 32 references

Taking a Lesson from Quantum Particles for Statistical Data Privacy

T0 review · 4 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims that the privacy-optimal additive noise, found by minimizing Fisher information under a utility constraint, must satisfy the time-independent Schrödinger equation, and that a Heisenberg-like bound $JQ \ge 1$ governs the…

desk verdict The central variational claim is wrong as stated—bounded-support noise with zero Fisher information beats the paper's Schrödinger solution—so the main example collapses, though the unbounded Gaussian case is salvageable. read the letter →

arxiv 1908.04954 v1 pith:MEP6PKVN submitted 2019-08-14 cs.CR cs.ITmath.ITquant-ph

Farhad Farokhi This is my paper · ORCID

classification cs.CRcs.ITmath.ITquant-ph MSC 94A1762B1081Q05

keywords differentialprivacyFisherinformationSchrödingerequationadditivenoiseutilitytrade-offCramér-Raoboundinformation-theoreticquantummechanics

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Privacy is treated as control over what an adversary can infer from a noisy query answer. The paper proposes measuring leakage not by mutual information, which requires knowing the distribution of the private data, but by the Fisher information of the added noise. It then shows that the noise minimizing Fisher information subject to a bound on its expected cost is described by the time-independent Schrödinger equation, with the noise density written as the squared wave function $\psi^2$. A second result is a universal trade-off: for mean-square noise cost, the Fisher information $J$ and expected cost $Q$ always satisfy $JQ \ge 1$. If the framework is right, designing privacy-preserving noise becomes solving an eigenvalue problem instead of tuning a differential-privacy budget.

What carries the argument

The load-bearing object is the change of variables $\psi=\sqrt{p_w}$, which rewrites the probability density as the squared amplitude of a wave function and converts the constrained minimization of Fisher information into an eigenvalue problem for the Schrödinger operator $-\nabla^2 + V(w)$, with $V(w)=(\mu+\lambda g(w))/4$ inside the support and $+\infty$ outside. This substitution is what lets known quantum-mechanical solutions, such as the infinite square well, supply the optimal noise densities. The companion identity is the "Privacy Principle" $JQ\ge1$, derived from the lower bound $J\ge 1/\mathbb{E}[w^\top w]$ for any density, which encodes the privacy–utility trade-off.

What would settle it

Take a scalar query with a Gaussian prior $x\sim\mathcal{N}(0,\sigma^2)$ and a fixed noise variance. Compare the paper's optimal Gaussian noise, whose Fisher information is $1/\sigma_n^2$, with Laplace noise of the same variance, whose Fisher information is larger. Compute the adversary's Bayesian mean-squared error under both noises. If the Laplace noise ever yields a smaller posterior mean-squared error, then minimizing Fisher information does not maximize privacy against a prior-aware adversary, and the paper's central premise collapses.

Watch

Extended reading notes

Core claim

The core claim is that the optimal privacy-preserving additive noise for a query $y=f(x)+w$ is the one that minimizes the Fisher information $J$ of the density $p_w$ subject to a utility constraint $\mathbb{E}[g(w)]\le\rho$. The variational solution of this problem is that $\psi(w)=\sqrt{p_w(w)}$ must obey the time-independent Schrödinger equation $\nabla^2\psi(w)-\frac{1}{4}(\mu+\lambda g(w))\psi(w)=0$ on the support of the noise, with infinite potential outside the support, where $\lambda\le0$ and $\mu$ are Lagrange multipliers. In the scalar bounded-support case with $g=0$ the optimal density is the ground-state sine-squared shape $\sin^2(\pi(w-a)/(2a))/a$; in the unbounded quadratic-cost case the optimal density is Gaussian with variance $\rho$. The paper also proves the "Privacy Principle" $JQ\ge1$ for $g(w)=w^\top w$, where $J$ is the Fisher information and $Q=\mathbb{E}[w^\top w]$, showing that improving privacy forces a larger expected noise cost, analogous to Heisenberg uncertainty.

Load-bearing premise

The whole argument rests on equating privacy with a small Fisher information for the added noise, and the mathematical link between Fisher information and estimation error only directly limits estimators that are not systematically biased; if the adversary uses prior knowledge or a biased rule, low Fisher information may not mean privacy.

Editorial extensions

If this is right

For scalar queries with bounded support and no utility cost, the optimal privacy-preserving noise is the ground-state density $\sin^2(\pi(w-a)/(2a))/a$, with Fisher information $\pi^2/a$, and every higher mode $n\ge2$ is strictly worse for privacy.
For scalar queries with unbounded support and a quadratic utility cost, the optimal noise is Gaussian with variance $\rho$, which links the framework to the relaxed differential-privacy guarantees that Gaussian noise provides.
The design of the optimal noise does not require a prior distribution over the private data; only the utility function $g$ and the chosen bound $\rho$ enter the calculation.
The inequality $JQ\ge1$ implies that any privacy gain, measured as a decrease in Fisher information, must be paid for by an increase in the expected noise cost, because the query quality degrades.
The need to set a differential-privacy budget is replaced by choosing the utility level $\rho$ and solving a Schrödinger eigenvalue problem, which also lets the support of admissible noises be specified in advance.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Editorial inference: The spectral form of the solution suggests that families of privacy-optimal noise shapes could be indexed by quantum-like eigenvalues, so mechanism design becomes a matter of selecting states of the associated Schrödinger operator rather than tuning parametric families.
Beyond the paper: For multivariate queries, replacing scalar Fisher information with a Fisher-information matrix should yield a system of coupled Schrödinger-like equations, and the scalar uncertainty bound $JQ\ge1$ may have a matrix counterpart involving the trace of the inverse information matrix.
Testable extension: Because the trade-off is proved only for $g(w)=w^\top w$, a natural extension is to derive analogous inequalities for other utility costs using weighted Poincaré or uncertainty inequalities; the paper does not attempt this.
Extension with an open flank: Since the Cramér–Rao reasoning covers unbiased estimators, a robust privacy claim against Bayesian or otherwise biased adversaries would need an additional minimax argument, testing whether the same noise also minimizes worst-case error over biased estimators.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Rough take: the central claim doesn't hold as stated. The variational problem in (5) has no boundary condition, and for W=[-a,a], g=0, the uniform density is feasible and gives I=0, so the sine-wave "Schrödinger" solution is not a minimizer. The displayed Fisher information for that solution is off by a factor of a (correct value n²π²/a²). These two issues undermine the main example.

What's genuinely there: the idea of choosing additive noise by minimizing Fisher information is a reasonable prior-free way to frame privacy, and the unbounded-support Gaussian case (for g(w)=w²) is correct and known. The prose is clear and honest about the prior literature, and the paper is short and readable. The connection to quantum mechanics is a nice expository hook, though it is not new: Frieden and Reginatto already use the Schrödinger equation derived from minimum Fisher information. The contribution is the privacy application and the Heisenberg analogy, not the mathematics.

The soft spots are in the load-bearing parts. Besides the boundary-condition problem and the dimensional error, the "Privacy Principle" JQ≥1 is true but not tight in multiple dimensions; the correct scalar bound is IQ≥m², so calling it "reminiscent of Heisenberg" is a stretch. The trade-off is essentially the Cramér-Rao bound. And the privacy interpretation rests on assuming the adversary is unbiased and that scalar Fisher information is a worst-case measure; the paper doesn't defend that assumption.

If you fix the boundary condition (e.g., restrict to W=R), correct the square-well computation, and state the sharp inequality, you'd have a reasonable short note. As written, it doesn't deliver on its main claims.

I'd send it to peer review because the flaw is subtle and the paper is honest about the literature, but the verdict would be reject without major revision. For a reading group, it's a useful cautionary example about boundary conditions in variational problems.

Referee Report

4 major / 4 minor

Summary. The paper proposes to use the Fisher information of additive noise as a privacy metric, arguing that minimizing it is independent of the adversary's actions. It formulates a constrained optimization problem, problem (5), and claims that the minimizer solves a time-independent Schrödinger equation, Eq. (6). It then gives a square-well example, a Gaussian example, and derives a 'Privacy Principle' JQ ≥ 1, Eq. (9), as a fundamental privacy-utility trade-off. The abstract frames the contribution as an information-theoretic privacy framework that avoids prior assumptions on the private data and exhibits a Heisenberg-like uncertainty relation.

Significance. If the central variational claim were correct, the paper would offer an elegant and potentially useful reduction: privacy-optimal additive noise would be obtained by solving a quantum eigenvalue problem without a prior on the private data. The use of Fisher information is a reasonable and concrete modeling choice, and the connection to the Schrödinger equation is mathematically suggestive. However, the paper's main derivation is not valid as stated because the variational problem has no boundary conditions that force the claimed square-well solutions, and the paper contains concrete errors in the example and in the multivariate privacy principle. These issues directly affect the central claims, so the significance of the contribution as written is not established.

major comments (4)

[§3, Eq. (6)] The variational derivation implicitly imposes a Dirichlet boundary condition that is not part of problem (5). Problem (5) minimizes I over densities on W with only the moment constraint Q ≤ ρ; no condition at ∂W is imposed. For W = [-a,a] and g(w) = 0, the uniform density p(w) = 1/(2a) is feasible for every ρ ≥ 0 and gives ∇ log p = 0 a.e., hence I = 0, while the paper's square-well candidate has I = n²π²/a² > 0. Thus Eq. (6) characterizes the solution of a restricted problem with an additional ψ = 0 boundary condition, not the minimizer of the stated privacy problem. This is a load-bearing flaw because the central claim that the optimal privacy-preserving noise must satisfy the Schrödinger equation rests on Eq. (6).
[§3, square-well example] The reported value I = n²π²/a is dimensionally incorrect. For the density p(w) = sin²(nπ(w - a)/(2a))/a on W = [-a,a], direct differentiation gives I = ∫ (p')²/p dw = n²π²/a², not n²π²/a. The dimension of the Fisher information in (3) is 1/length², so the paper's expression has the wrong dimension. This is not a typo of presentation only: the example is used to demonstrate that the Schrödinger solution is the privacy-optimal policy, and the boundary-condition counterexample above shows the claimed optimality is false.
[§3, Eq. (9)] The 'Privacy Principle' JQ ≥ 1 is stated as a general inequality for any density, but it is false under the paper's own definitions for multivariate noise. For a zero-mean Gaussian on R^m with covariance σ²I, I = m/σ² and E{wᵀw} = mσ², so the product equals m², which is larger than 1 but still consistent; however, for a uniform density on a bounded set, I = 0 and E{wᵀw} > 0, giving JQ = 0. Thus Eq. (9) does not hold without additional regularity and support conditions that are not stated. The cited lower bound [29] cannot apply to densities with nonvanishing boundary values or to multivariate settings in the form used.
[§2, Eq. (2)] The privacy interpretation relies on the Cramér-Rao bound, which bounds the error of unbiased estimators. The paper asserts that minimizing I is a great measure of privacy because it is independent of the adversary's actions, but it does not prove that minimizing scalar Fisher information maximizes estimation error for biased, Bayesian, or prior-informed adversaries. Since the Cramér-Rao bound only lower-bounds a certain class of estimators, a density with smaller I need not provide larger worst-case estimation error against an arbitrary adversary. This modeling premise is not validated in the paper.

minor comments (4)

[§2, Eq. (3) and §3, Eq. (9)] The quantity defined in (3) is called I, but the 'Privacy Principle' in (9) uses J. The notation should be consistent.
[§3, normalization of ψ] The sentence 'c = 1/√a' is correct only after the fact that ∫ ψ² over W is 1; it would be clearer to state that the normalization constant is determined by this condition.
[References] Some references are informal or lack page numbers, such as [29]; given that [29] is used to justify the universal lower bound in Eq. (9), the precise statement and hypotheses of that theorem should be provided.
[§1 and §2] The phrase 'minimizing I can capture the most privacy-preserving additive noise' should be hedged to reflect that this is a modeling choice, not a theorem, given that the paper does not analyze adversarial estimators beyond the Cramér-Rao bound.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the Schrödinger equation is derived from a standard variational argument, and the one self-citation is not load-bearing.

full rationale

The paper's central derivation is self-contained. It starts from an explicit optimization problem (5), forms the Lagrangian, and applies standard Euler–Lagrange conditions from an external textbook (Theorem 5.3 in [12]) to obtain the optimality condition (6), the time-independent Schrödinger equation. No parameter is fitted to data, and the Lagrange multipliers μ and λ are determined by the constraints rather than by any target result. The free-particle example explicitly solves the resulting ODE, and the Gaussian example is presented as a separate algebraic solution. The Privacy Principle (9) is obtained by substituting the known inequality J ≥ 1/E{w^T w} from [29] into the definition Q = E{w^T w}; this is a direct application, not a renamed input. The only self-citation, [13], appears in the final paragraph as an acknowledgment that Fisher information was previously used for privacy, and it plays no role in deriving Eq. (6) or (9). The skeptic's boundary-condition objection concerns whether the variational problem's optimum satisfies an unstated Dirichlet condition; that is a mathematical correctness issue, not a circularity, and is outside the scope of this pass.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The central claim rests on modeling assumptions: the Cramér-Rao bound for unbiased estimators, the validity of the Lagrange multiplier rule, and the identification of Fisher information with privacy. No new entities or fitted parameters are introduced; the Lagrangian multipliers and utility budget are problem inputs, not fitted constants.

free parameters (2)

Lagrange multipliers λ, μ
Dual variables for the utility and normalization constraints in the variational problem. They parameterize the family of candidate densities and are fixed by the constraints, not fitted to external data.
Utility budget ρ
The curator's chosen bound on expected cost of noise. It is an input to the optimization problem, not an estimated quantity.

assumptions (3)

domain assumption Cramér-Rao lower bound applies to the adversarial estimation problem
Section 2 uses E||x - xhat||² >= α(x)/I, which assumes unbiased estimators and a known Jacobian of f. This is the bridge from Fisher information to privacy.
standard math Minimizer of the Lagrangian exists and Lagrange multiplier rule holds
The paper invokes Theorem 5.3 of Edwards [12] without checking regularity conditions on the density class and constraints. This is standard but not proven for this specific setting.
domain assumption Fisher information of the additive noise is a valid worst-case privacy measure
Section 2 states that minimizing I is a great measure of privacy because it is independent of the adversary. This is a modeling assumption, not a theorem.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Taking a Lesson from Quantum Particles for Statistical Data Privacy." pith.science (2026). https://pith.science/paper/MEP6PKVN

@misc{pith2026190804954,
  author       = {Pith},
  title        = {Pith review of: Taking a Lesson from Quantum Particles for Statistical Data Privacy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MEP6PKVN}},
  note         = {Machine review of arXiv:1908.04954}
}

read the original abstract

Privacy is under threat from artificial intelligence revolution fueled by unprecedented abundance of data. Differential privacy, an established candidate for privacy protection, is susceptible to adversarial attacks, acts conservatively, and leads to miss-implementations because of lacking systematic methods for setting its parameters (known as the privacy budget). An alternative is information-theoretic privacy using entropy with the drawback of requiring prior distribution of the private data. Here, by using the Fisher information, information-theoretic privacy framework is extended to avoid unnecessary assumptions on the private data. The optimal privacy-preserving additive noise, extracted by minimizing the Fisher information, must follow the time-independent Schrodinger's equation. A fundamental trade-off between privacy and utility is also proved, reminiscent of the Heisenberg uncertainty principle.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 31 canonical work pages

[29]

A lower bou nd for the ﬁsher information measure

Manuel S Stein, Amine Mezghani, and Josef A Nossek. A lower bou nd for the ﬁsher information measure. IEEE Signal Process. Lett. , 21(7):796–799, 2014

work page 2014
[1]

The US census bureau adopts diﬀerential privacy

John M Abowd. The US census bureau adopts diﬀerential privacy . In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & D ata Mining , pages 2867–2867. ACM, 2018

work page 2018
[2]

Apple diﬀerential privacy technical overview, 2016

Apple. Apple diﬀerential privacy technical overview, 2016. https://www.apple.com/privacy/docs/ Differential_Privacy_Overview.pdf

work page 2016
[3]

Fool’s gold: an illustrated critique of diﬀerential privacy

Jane Bambauer, Krishnamurty Muralidhar, and Rathindra Sarat hy. Fool’s gold: an illustrated critique of diﬀerential privacy. Vand. J. Ent. & Tech. L. , 16:701, 2013

work page 2013
[4]

On generalized Cram´ er–Rao inequalit ies, generalized Fisher information and characterizations of generalized q-Gaussian distributions

Jean-Fran¸ cois Bercher. On generalized Cram´ er–Rao inequalit ies, generalized Fisher information and characterizations of generalized q-Gaussian distributions. Journal of Physics A: Mathematical and Theoretical, 45(25):255303, 2012

work page 2012
[5]

Learning with privacy in consensus + obfuscation

Paolo Braca, Riccardo Lazzeretti, Stefano Marano, and Vincen zo Matta. Learning with privacy in consensus + obfuscation. IEEE Signal Processing Letters , 23(9):1174–1178, 2016

work page 2016
[6]

The tracker: A threat to statistical database security

Dorothy E Denning, Peter J Denning, and Mayer D Schwartz. The tracker: A threat to statistical database security. ACM Transactions on Database Systems , 4(1):76–96, 1979

work page 1979
[7]

Revealing information while preserving p rivacy

Irit Dinur and Kobbi Nissim. Revealing information while preserving p rivacy. In Proceedings of the twenty-second ACM SIGMOD-SIGACT-SIGART symposium on Prin ciples of database systems , pages 202–210. ACM, 2003

work page 2003

Show all 32 references

[8]

Our data, ourselves: Privacy via distributed noise generation

Cynthia Dwork, Krishnaram Kenthapadi, Frank McSherry, Ilya M ironov, and Moni Naor. Our data, ourselves: Privacy via distributed noise generation. In Serge Vaud enay, editor, Advances in Cryptology - EUROCRYPT 2006 , pages 486–503, Berlin, Heidelberg, 2006. Springer Berlin Heidelber g

2006
[9]

C alibrating noise to sensitivity in private data analysis

Cynthia Dwork, Frank McSherry, Kobbi Nissim, and Adam Smith. C alibrating noise to sensitivity in private data analysis. In Shai Halevi and Tal Rabin, editors, Theory of Cryptography , pages 265–284, Berlin, Heidelberg, 2006. Springer Berlin Heidelberg

2006
[10]

Diﬀerential data privacy , April 13 2010

Cynthia Dwork and Frank D Mcsherry. Diﬀerential data privacy , April 13 2010. US Patent 7,698,250

2010
[11]

The algorithmic foundations of d iﬀerential privacy

Cynthia Dwork and Aaron Roth. The algorithmic foundations of d iﬀerential privacy. Foundations and Trends in Theoretical Computer Science , 9(3–4):211–407, 2014

2014
[12]

C. H. Edwards. Advanced Calculus of Several Variables . Academic Press, 1973

1973
[13]

Fisher information as a m easure of privacy: Preserving privacy of households with smart meters using batteries

Farhad Farokhi and Henrik Sandberg. Fisher information as a m easure of privacy: Preserving privacy of households with smart meters using batteries. IEEE Transactions on Smart Grid , 9(5):4726–4734, 2018

2018
[14]

Charles Fried. Privacy. The Yale Law Journal , 77:475–483, 1968

1968
[15]

Fisher information, disorder, and the equilibrium d istributions of physics

B Roy Frieden. Fisher information, disorder, and the equilibrium d istributions of physics. Physical Review A , 41(8):4265, 1990

1990
[16]

Issues e ncountered deploying diﬀerential privacy

Simson L Garﬁnkel, John M Abowd, and Sarah Powazek. Issues e ncountered deploying diﬀerential privacy. In Proceedings of the 2018 Workshop on Privacy in the Electroni c Society , pages 133–137, 2018

2018
[17]

Privacy and the limits of law

Ruth Gavison. Privacy and the limits of law. The Yale Law Journal , 89(3):421–471, 1980

1980
[18]

How one of Apple’s key privacy safeguards fa lls short, 2017

Andy Greenberg. How one of Apple’s key privacy safeguards fa lls short, 2017. https://www.wired. com/story/apple-differential-privacy-shortcomings/

2017
[19]

Diﬀ erential privacy under ﬁre

Andreas Haeberlen, Benjamin C Pierce, and Arjun Narayan. Diﬀ erential privacy under ﬁre. In USENIX Security Symposium, 2011. 4

2011
[20]

Explorin g estimator bias-variance tradeoﬀs using the uniform CR bound

Alfred O Hero, Jeﬀrey A Fessler, and Mohammad Usman. Explorin g estimator bias-variance tradeoﬀs using the uniform CR bound. IEEE Transactions on Signal Processing , 44(8):2026–2041, 1996

2026
[21]

Zero duality gaps in inﬁnite-dim ensional programming

V Jeyakumar and Henry Wolkowicz. Zero duality gaps in inﬁnite-dim ensional programming. Journal of Optimization Theory and Applications , 67(1):87–108, 1990

1990
[22]

Information th eoretic security

Yingbin Liang, H Vincent Poor, and Shlomo Shamai. Information th eoretic security. Foundations and Trends in Communications and Information Theory , 5(4–5):355–580, 2009

2009
[23]

Information- theoretic privacy in smart metering systems using cascaded rechargeable batteries

Yuhan Helena Liu, Si-Hyeon Lee, and Ashish Khisti. Information- theoretic privacy in smart metering systems using cascaded rechargeable batteries. IEEE Signal Processing Letters , 24(3):314–318, 2017

2017
[24]

Privacy and the modern grid

Sonia K McNeil. Privacy and the modern grid. Harvard Journal of Law & Technology , 25:199–224, 2011

2011
[25]

Krish Muralidhar and Rathindra Sarathy. Does diﬀerential priva cy protect terry gross’ privacy? In Josep Domingo-Ferrer and Emmanouil Magkos, editors, Privacy in Statistical Databases , pages 200–209, Berlin, Heidelberg, 2010. Springer Berlin Heidelberg

2010
[26]

Derivation of the equations of nonrelativist ic quantum mechanics using the principle of minimum ﬁsher information

Marcel Reginatto. Derivation of the equations of nonrelativist ic quantum mechanics using the principle of minimum ﬁsher information. Physical Review A , 58(3):1775, 1998

1998
[27]

Big data: The end of privacy or a new beginning? International Data Privacy Law ,

Ira Rubinstein. Big data: The end of privacy or a new beginning? International Data Privacy Law ,
[28]

Mathematical Statistics

Jun Shao. Mathematical Statistics . Springer Texts in Statistics. Springer-Verlag New York, 2003

2003
[30]

Privacy loss in Apple’s implementation of diﬀerential privacy on macOS 10.12

Jun Tang, Aleksandra Korolova, Xiaolong Bai, Xueqiang Wang, an d Xiaofeng Wang. Privacy loss in Apple’s implementation of diﬀerential privacy on macOS 10.12. arXiv preprint arXiv:1709.02753 , 2017

2017 arXiv
[31]

A. D. Wyner. The wire-tap channel. Bell System Technical Journal, The , 54(8):1355–1387, 1975

1975
[32]

A source coding problem for sources with a dditional outputs to keep secret from the receiver or wiretappers

Hirosuke Yamamoto. A source coding problem for sources with a dditional outputs to keep secret from the receiver or wiretappers. IEEE Transactions on Information Theory , 29(6):918–923, 1983. 5

1983

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [29]

A lower bou nd for the ﬁsher information measure

Manuel S Stein, Amine Mezghani, and Josef A Nossek. A lower bou nd for the ﬁsher information measure. IEEE Signal Process. Lett. , 21(7):796–799, 2014

work page 2014

[2] [1]

The US census bureau adopts diﬀerential privacy

John M Abowd. The US census bureau adopts diﬀerential privacy . In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & D ata Mining , pages 2867–2867. ACM, 2018

work page 2018

[3] [2]

Apple diﬀerential privacy technical overview, 2016

Apple. Apple diﬀerential privacy technical overview, 2016. https://www.apple.com/privacy/docs/ Differential_Privacy_Overview.pdf

work page 2016

[4] [3]

Fool’s gold: an illustrated critique of diﬀerential privacy

Jane Bambauer, Krishnamurty Muralidhar, and Rathindra Sarat hy. Fool’s gold: an illustrated critique of diﬀerential privacy. Vand. J. Ent. & Tech. L. , 16:701, 2013

work page 2013

[5] [4]

On generalized Cram´ er–Rao inequalit ies, generalized Fisher information and characterizations of generalized q-Gaussian distributions

Jean-Fran¸ cois Bercher. On generalized Cram´ er–Rao inequalit ies, generalized Fisher information and characterizations of generalized q-Gaussian distributions. Journal of Physics A: Mathematical and Theoretical, 45(25):255303, 2012

work page 2012

[6] [5]

Learning with privacy in consensus + obfuscation

Paolo Braca, Riccardo Lazzeretti, Stefano Marano, and Vincen zo Matta. Learning with privacy in consensus + obfuscation. IEEE Signal Processing Letters , 23(9):1174–1178, 2016

work page 2016

[7] [6]

The tracker: A threat to statistical database security

Dorothy E Denning, Peter J Denning, and Mayer D Schwartz. The tracker: A threat to statistical database security. ACM Transactions on Database Systems , 4(1):76–96, 1979

work page 1979

[8] [7]

Revealing information while preserving p rivacy

Irit Dinur and Kobbi Nissim. Revealing information while preserving p rivacy. In Proceedings of the twenty-second ACM SIGMOD-SIGACT-SIGART symposium on Prin ciples of database systems , pages 202–210. ACM, 2003

work page 2003