Pith. sign in

REVIEW 4 major objections 5 minor 35 references

Private Semantic Communications with Separate Blind Encoders

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read The paper's central result is that the privacy-utility trade-off for a task-blind privacy encoder is bracketed by explicit entropy expressions $\max\{L^1_h(\epsilon),L^2_h(\epsilon)\} \le h_\epsilon(P_{S,f(X)}) \le H(f(X)|S)+\epsilon$.

desk verdict The blind two-encoder setup is genuinely new, but Theorem 1 is false as stated because it omits the eps ≤ H(S) condition. read the letter →

arxiv 2412.17658 v1 pith:UMN2SVDV submitted 2024-12-23 cs.IT math.IT

Amirreza Zamani , Mikael Skoglund This is my paper · ORCID
classification cs.ITmath.IT MSC 94A1594A1768P27
keywords privatesemanticcommunicationblindencodersprivacy-utilitytrade-offFunctionalRepresentationLemmaStrongcommoninformationmutualrandomizedresponse
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

This paper studies a statistical privacy layer in semantic communication that is 'blind' in two ways: the first encoder sees the source but not the private data, and the second encoder sees the private data and a semantic summary but not the user's task. The central claim is that the best privacy-utility trade-off $h_\epsilon(P_{S,f(X)})=\sup\{I(f(X);U):I(U;S)\le\epsilon\}$ is bracketed by explicit entropy expressions, $\max\{L^1_h(\epsilon),L^2_h(\epsilon)\}\le h_\epsilon(P_{S,f(X)})\le H(f(X)|S)+\epsilon$, and that the user's utility $I(h(X);U)$ lies in an interval of the same width plus $H(h(X)|f(X))$. The lower bounds come from constructive extended Functional Representation Lemma mechanisms, so they are achievable by a finite alphabet code rather than an abstract limit. A designer who trusts these bounds can certify privacy and utility from entropies alone, without solving the optimization over channels. The paper's numerical MNIST example shows the lower and upper utility bounds can be within about $1.4$ nats when the semantic is tightly coupled to the task.

What carries the argument

The argument rests on two constructive information-theoretic tools: the Extended Functional Representation Lemma (EFRL) and the Extended Strong Functional Representation Lemma (ESFRL), taken from the paper's earlier work [17]. These lemmas produce a channel output $\bar U$ (or $\tilde U$) that keeps $I(\bar U;f(X))$ large for a prescribed small $I(\bar U;S)=\epsilon$, with the randomized-response technique of Warner [34] used to hit the leakage target exactly. The derivation also uses the identity $I(f(X);U)=I(S;U)+H(f(X)|S)-I(U;S|f(X))-H(f(X)|S,U)$, plus the Markov-chain expansions of $I(U;h(X),f(X))$, to convert a bound on semantic privacy into a bound on task utility. Common information, in the sense of Wyner [35], is the sharper condition that makes the upper bound tight without requiring $S$ to be a function of $f(X)$.

What would settle it

Run the linear-programming computation of $h_\epsilon(P_{S,f(X)})$ on a small finite joint distribution where $S$ is not a function of $f(X)$ and compare the exact value with $\max\{L^1_h(\epsilon),L^2_h(\epsilon)\}$; one instance with exact $h_\epsilon$ below the max disproves Theorem 1's lower bound. Alternatively, implement the EFRL/ESFRL mechanism for that distribution and measure $I(U;S)$ from samples; if it is not $\epsilon$ to within simulation error, the inherited construction fails for the present setting.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that separating the encoder into a privacy-unaware semantic designer and a task-unaware privacy filter does not destroy the information-theoretic control of the trade-off. Theorem 1 states that for any discrete semantic $f(X)$ satisfying constraints (1)–(2), the value of the privacy-utility trade-off obeys the sandwich (6), with $L^1_h(\epsilon)=H(f(X)|S)-H(S|f(X))+\epsilon$ and $L^2_h(\epsilon)=H(f(X)|S)-\alpha H(S|f(X))+\epsilon-(1-\alpha)(\log(I(S;f(X))+1)+4)$, where $\alpha=\epsilon/H(S)$. For the user's utility the theorem gives $\max\{L^1(\epsilon),L^2(\epsilon),L^3(\epsilon)\}\le I(h(X);U)\le H(f(X)|S)+\epsilon+H(h(X)|f(X))$, with $L^1,L^2,L^3$ defined in (10)–(12). The upper bound is attained when the private $S$ is a deterministic function of $f(X)$; Corollary 1 widens this to equality of common information and mutual information between $S$ and $f(X)$, and Corollary 2 removes the dependence on the chosen semantic by bounding $I(U;h(X))\le\epsilon+\gamma_2-\gamma_1+H(h(X))$. The proof route is the identity chain (4)–(5), the EFRL/ESFRL constructions from [17], and the randomized-response step of [34].

Load-bearing premise

The load-bearing premise is that the lower-bound construction inherited from [17] achieves leakage exactly $\epsilon$ for the pair $(S,f(X))$ through the randomized-response step of [34], including the constant term $\log(I(S;f(X))+1)+4$; if that exact-leakage equality (15) fails, the lower bounds $L^1_h$ and $L^2_h$ do not follow as stated.

Editorial extensions

If this is right

  • A designer can certify that no privacy mechanism can disclose more than $H(f(X)|S)+\epsilon$ nats of the semantic to a user, and that the EFRL/ESFRL constructions disclose at least the computed $L^1_h$ or $L^2_h$ nats.
  • When $H(S|f(X))=0$ or common information equals mutual information, the trade-off is exactly $H(f(X)|S)+\epsilon$, so the semantic can be sized precisely from the residual uncertainty about it given the private data.
  • The user-utility lower bounds $L^1(\epsilon)$ and $L^2(\epsilon)$ are computable from entropies alone and are achieved by simple mechanisms, making the result a practical design recipe rather than an existence proof.
  • The semantic-free bound $I(U;h(X))\le\epsilon+\gamma_2-\gamma_1+H(h(X))$ lets a designer evaluate the privacy cost of a task before choosing the semantic $f(X)$.
  • In the MNIST example, the width of the utility interval is $H(h(X)|f(X))+H(f(X)|h(X))\simeq 1.4$ nats, so for highly correlated semantic/task pairs the bounds are a tight design target.

Reading between the lines

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

  • A natural next step the paper leaves open is to combine these EFRL/ESFRL bounds with the separation-technique lower bounds of [33]; the resulting lower bracket would likely be tighter but more complex.
  • The finite-alphabet assumption is likely load-bearing: the exact-leakage equality and the log term are proven for discrete alphabets, so extending to Gaussian or continuous semantic sources would require rechecking (15).
  • One could stress-test the task-blind assumption by giving encoder 2 a coarse description of the task; the utility bounds should interpolate between the present bounds and the full-access design of [4].
  • An empirical protocol follows directly: on any labeled dataset, compute the entropies of $(S,f(X),h(X))$, predict the achievable utility interval, train a real privacy mechanism, and see whether the measured $I(h(X);U)$ lands inside the interval; any miss identifies either a suboptimal mechanism or a violated assumption.
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.

Referee Report

4 major / 5 minor

Summary. The paper studies a semantic communication system with two separate encoders: encoder 1 observes the source X and designs a semantic f(X) for a task h(X), without access to private data S; encoder 2 observes S and f(X) and outputs disclosed data U, without access to the task. The privacy-utility trade-off is h_epsilon(P_{S,f(X)}) = sup_{P_{U|S,f(X)}: I(U;S)<=epsilon} I(U;f(X)), and the user utility is I(U;h(X)). The main result, Theorem 1, gives upper and lower bounds on h_epsilon and on I(U;h(X)), with lower bounds obtained from the Extended Functional Representation Lemma and the Extended Strong Functional Representation Lemma imported from the authors' earlier work [17]. The paper also states a corollary involving common information, a corollary giving a task-independent upper bound, and a numerical MNIST experiment comparing the bounds.

Significance. If correct, the results provide computable, parameter-free bounds for a meaningful two-encoder privacy scenario and the underlying mechanisms are constructive. The upper bound H(f(X)|S)+epsilon is simple and the lower bounds are stated in closed form. However, the mathematical novelty beyond the authors' prior work [17] is modest: the main technique is a direct application of previously established lemmas to the pair (S,f(X)). The central theorem as stated is false on part of its claimed domain, and several proof details are not rigorous. The scenario itself is relevant to semantic communication with privacy constraints, and the numerical comparison is useful, so the paper is worth revising.

major comments (4)
  1. [Section III, Theorem 1, Eqs. (6)-(8)] The statement 'For any 0 <= epsilon' is false for epsilon > H(S). Since I(U;f(X)) <= H(f(X)) for every U, we always have h_epsilon(P_{S,f(X)}) <= H(f(X)). But L1_h(epsilon) = H(f(X)|S) - H(S|f(X)) + epsilon = H(f(X)) - H(S) + epsilon, so for epsilon > H(S) the claimed lower bound exceeds H(f(X)). For example, if S and f(X) are independent uniform binary variables and epsilon = 2 bits, then h_epsilon = H(f(X)) = 1 bit while L1_h(epsilon) = 2 bits, contradicting (6). The theorem must be restricted to 0 <= epsilon <= H(S), with H(S) > 0 so that alpha = epsilon/H(S) is defined. This also affects the numerical experiment in Section IV, where H(S) is approximately 0.325 nats but epsilon is swept to 0.35 nats; the sweep should be limited to the valid range or the statement amended.
  2. [Section III, Eq. (4) and proof of Theorem 1] Equation (4) is not consistent: the left-hand side is I(f(X);U), but the right-hand side is the standard expansion for I(h(X);U), with h(X) appearing in all three entropy/conditional terms. The proof then says 'Using (4) we have I(U;f(X)) = ...' and concludes I(U;h(X)) <= epsilon + H(f(X)|S), which does not follow from (4) as written. The upper bound in (6) is correct and can be proved directly from the identity I(U;f(X)) = I(U;S) + H(f(X)|S) - H(f(X)|U,S) - I(U;S|f(X)) <= I(U;S) + H(f(X)|S). Please replace Eq. (4) with the correct identity and rewrite the proof of the upper bounds accordingly.
  3. [Section III, proof of Theorem 1, Eqs. (13)-(15)] The lower bounds are imported from [17, Theorem 2] through the sentence 'Using the same arguments in [17, Theorem 2]', without stating the hypotheses under which the EFRL/ESFRL constructions achieve leakage exactly epsilon for the pair (S,f(X)). Since L1_h and L2_h are increasing in epsilon, if the imported construction only yields I(U;S) < epsilon, the claimed lower bounds at level epsilon do not automatically follow. The proof should either state the precise lemma from [17] being invoked and verify that its conditions (discrete alphabets, finiteness, range of epsilon, etc.) hold for the pair (S,f(X)), or provide the construction directly. As written, the lower-bound half of Theorem 1 is not self-contained.
  4. [Section III, Corollary 2 proof] The final inequality in the proof of Corollary 2, H(f(X)|S) + H(h(X)|f(X)) <= H(h(X)) + gamma_2 - gamma_1, is not a consequence of (1) and (2); in general it is false. The claimed bound is nevertheless true vacuously because I(U;h(X)) <= H(h(X)) and gamma_2 >= gamma_1, so the right-hand side of (20) is at least H(h(X)). Please replace the argument with a correct proof or explicitly note that the bound is trivial.
minor comments (5)
  1. [Section II, Eq. (2)] In Eq. (2), h(f(X)) should presumably be H(f(X)) (the entropy of the semantic), since h is used elsewhere for the task function; the current notation conflates entropy with the task function and is confusing.
  2. [Section III, proof of Theorem 1] In the final substitution step, the text says 'substituting U by U* in (17) we have I(bar{U};h(X)) >= ...', but the left-hand side should be I(U*;h(X)), not I(bar{U};h(X)). Please fix this typo.
  3. [Section IV] The paper states that the MNIST data set has |X| = 6000 images; the MNIST training set contains 60000 images. Please correct this number or clarify the subsample used.
  4. [Fig. 1] The label 'Lekagae matrix' in the figure should be 'Leakage matrix'.
  5. [Section III, Corollary 1] The proof of Corollary 1 is only a reference to [17, Theorem 3 and Proposition 3]; please state the common-information condition explicitly or provide the argument so that the corollary is understandable without consulting the prior paper.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; lower bounds are direct, independent applications of the prior peer-reviewed theorem [17], with no fitted parameter renamed as a prediction.

full rationale

The derivation chain of Theorem 1 is not circular. The upper bound I(U;f(X)) <= epsilon + H(f(X)|S) is proved in-paper from the exact expansion (4), and the utility bounds follow algebraically from (5), (17), and the theorem's upper bound. The lower bounds L1_h(epsilon) and L2_h(epsilon) are not fitted or re-derived from the target claim; they are imported by substituting (S, f(X)) in the peer-reviewed, parameter-free [17, Theorem 2], with the EFRL/ESFRL constructions established there. That prior theorem does not assume the present paper's conclusion, so the self-citation is independent support rather than a circular load-bearing step. The numerical experiment uses empirical distributions and evaluates the stated formulas; no fitted parameter is renamed as a prediction. The main concern in the paper is a correctness gap, not circularity: the theorem states 'For any 0 <= epsilon' while defining alpha = epsilon/H(S); for epsilon > H(S), L1_h(epsilon) can exceed H(f(X)), making the claimed lower bound impossible. That is an invalid domain statement, not a reduction of the result to its own inputs by construction. Therefore no circular step is identified.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The central bounds are imported from the authors' prior work [17] and not re-derived here. The paper relies on those lemmas as black boxes. No new entities or fitted parameters are introduced; the numerical example uses empirical distributions from MNIST but no fitted constants.

assumptions (5)
  • domain assumption Extended FRL and SFRL bounds from [17, Lemmas 4 and 5] hold for the pair (S, f(X)) with the stated constants.
    Theorem 1 lower bounds are asserted to follow from [17, Theorem 2] via these lemmas; no proof is given in this paper.
  • domain assumption Randomized response [34] can be tuned to achieve I(U;S)=epsilon exactly for the constructed U.
    Required to satisfy the leakage constraint with equality; the paper cites [17] but does not reproduce the construction.
  • domain assumption Common information and mutual information equality condition from [17, Theorem 3] / Wyner [35] implies tightness of the upper bound.
    Corollary 1 is stated without proof, citing [17].
  • domain assumption The joint distribution P_{S,X} is discrete on finite alphabets with arbitrary correlation.
    System model in Section II; needed for entropy and exact bounds.
  • domain assumption The semantic f(X) satisfies gamma1 ≤ I(f(X);h(X)) ≤ gamma2 < H(h(X)) and H(f(X)) ≤ gamma3.
    These constraints define the semantic design space in Section II and are used in Corollary 2.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Private Semantic Communications with Separate Blind Encoders." pith.science (2026). https://pith.science/paper/UMN2SVDV

@misc{pith2026241217658,
  author       = {Pith},
  title        = {Pith review of: Private Semantic Communications with Separate Blind Encoders},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UMN2SVDV}},
  note         = {Machine review of arXiv:2412.17658}
}
abstract

We study a semantic communication problem with a privacy constraint where an encoder consists of two separate parts, e.g., encoder 1 and encoder 2. The first encoder has access to information source $X=(X_1,\ldots,X_N)$ which is arbitrarily correlated with private data $S$. The private data is not accessible by encoder 1, however, the second encoder has access to it and the output of encoder 1. A user asks for a task $h(X)$ and the first encoder designs the semantic of the information source $f(X)$ to disclose. Due to the privacy constraints $f(X)$ can not be revealed directly to the user and the second encoder applies a statistical privacy mechanism to produce disclosed data $U$. Here, we assume that encoder 2 has no access to the task and the design of the disclosed data is based on the semantic and the private data. In this work, we propose a novel approach where $U$ is produced by solving a privacy-utility trade-off based on the semantic and the private data. We design $U$ utilizing different methods such as using extended versions of the Functional Representation Lemma and the Strong Functional Representation Lemma. We evaluate our design by computing the utility attained by the user. Finally, we study and compare the obtained bounds in a numerical example.

Figures

Figures reproduced from arXiv: 2412.17658 by the authors.

Figure 1
Figure 1. The encoder consists of two separate parties with different goals. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Comparing the lower and upper bounds obtained in Theorem 1. In [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 20 canonical work pages

  1. [17]

    On the privacy-utility trade-off with and without direct access to the private data,

    A. Zamani, T. J. Oechtering, and M. Skoglund, “On the privacy-utility trade-off with and without direct access to the private data,” IEEE Transactions on Information Theory , vol. 70, no. 3, pp. 2177–2200, 2024

    2024

  2. [1]

    Beyond transmitting bits: Context, semantics, and task-oriented communications,

    D. G ¨und¨uz, Z. Qin, I. E. Aguerri, H. S. Dhillon, Z. Yang, A. Yener, K. K. Wong, and C.-B. Chae, “Beyond transmitting bits: Context, semantics, and task-oriented communications,” IEEE Journal on Selected Areas in Communications, vol. 41, no. 1, pp. 5–41, 2023

    2023

  3. [2]

    Feist, Significance in language: A theory of semantics

    J. Feist, Significance in language: A theory of semantics . Taylor & Francis, 2022

    work page 2022

  4. [3]

    Goal- oriented and semantic communication in 6g ai-native networks: The 6g-goals approach,

    E. C. Strinati, P. Di Lorenzo, V . Sciancalepore, A. Aijaz, M. Kountouris, D. G ¨und¨uz, P. Popovski, M. Sana, P. A. Stavrou, B. Soret et al., “Goal- oriented and semantic communication in 6g ai-native networks: The 6g-goals approach,” arXiv preprint arXiv:2402.07573 , 2024

    arXiv 2024

  5. [4]

    Semantic communications with privacy constraints,

    A. Zamani, S. Daei, T. J. Oechtering, D. G ¨und¨uz, and M. Skoglund, “Semantic communications with privacy constraints,” in International Zurich Seminar on Information and Communication (IZS 2024). Pro- ceedings. ETH Z ¨urich, 2024, pp. 154–158

    work page 2024

  6. [5]

    Multi-task pri- vate semantic communication,

    A. Zamani, S. Daei, T. J. Oechtering, and M. Skoglund, “Multi-task pri- vate semantic communication,” in 2024 IEEE International Symposium on Information Theory (ISIT) , 2024, pp. 3480–3485

    work page 2024

  7. [6]

    Task-oriented multi- user semantic communications,

    H. Xie, Z. Qin, X. Tao, and K. B. Letaief, “Task-oriented multi- user semantic communications,” IEEE Journal on Selected Areas in Communications, vol. 40, no. 9, pp. 2584–2597, 2022

    2022

  8. [7]

    A unified multi-task semantic communication system with domain adaptation,

    G. Zhang, Q. Hu, Z. Qin, Y . Cai, and G. Yu, “A unified multi-task semantic communication system with domain adaptation,” in GLOBE- COM 2022 - 2022 IEEE Global Communications Conference , 2022, pp. 3971–3976

    work page 2022

Show all 35 references
  1. [8]

    Privacy-preserving task-oriented semantic communications against model inversion attacks,

    Y . Wang, S. Guo, Y . Deng, H. Zhang, and Y . Fang, “Privacy-preserving task-oriented semantic communications against model inversion attacks,” IEEE Transactions on Wireless Communications , pp. 1–1, 2024

    2024

  2. [9]

    The role of fidelity in goal-oriented semantic communication: A rate distortion approach,

    P. A. Stavrou and M. Kountouris, “The role of fidelity in goal-oriented semantic communication: A rate distortion approach,” IEEE Transac- tions on Communications , vol. 71, no. 7, pp. 3918–3931, 2023

    2023

  3. [10]

    Goal-oriented single-letter codes for lossy joint source-channel coding,

    ——, “Goal-oriented single-letter codes for lossy joint source-channel coding,” in 2023 IEEE Information Theory Workshop (ITW) , 2023, pp. 64–69

    2023

  4. [11]

    6g networks: Beyond shannon towards semantic and goal-oriented communications,

    E. C. Strinati and S. Barbarossa, “6g networks: Beyond shannon towards semantic and goal-oriented communications,” Computer Networks, vol. 190, p. 107930, 2021

    2021

  5. [12]

    A theory of semantic communication,

    Y . Shao, Q. Cao, and D. G¨und¨uz, “A theory of semantic communication,” arXiv preprint arXiv:2212.01485 , 2022

    2022 arXiv

  6. [13]

    On perfect privacy,

    B. Rassouli and D. G ¨und¨uz, “On perfect privacy,” IEEE Journal on Selected Areas in Information Theory , vol. 2, no. 1, pp. 177–191, 2021

    2021

  7. [14]

    A design framework for strongly χ2-private data disclosure,

    A. Zamani, T. J. Oechtering, and M. Skoglund, “A design framework for strongly χ2-private data disclosure,” IEEE Transactions on Information Forensics and Security, vol. 16, pp. 2312–2325, 2021

    2021

  8. [15]

    Data disclosure with non-zero leakage and non-invertible leakage matrix,

    A. Zamani, T. J. Oechtering, and M. Skoglund, “Data disclosure with non-zero leakage and non-invertible leakage matrix,” IEEE Transactions on Information Forensics and Security , vol. 17, pp. 165–179, 2022

    2022

  9. [16]

    Secrecy by design with appli- cations to privacy and compression,

    Y . Y . Shkel, R. S. Blum, and H. V . Poor, “Secrecy by design with appli- cations to privacy and compression,” IEEE Transactions on Information Theory, vol. 67, no. 2, pp. 824–843, 2021

    2021

  10. [18]

    From the information bottleneck to the privacy funnel,

    A. Makhdoumi, S. Salamatian, N. Fawaz, and M. M ´edard, “From the information bottleneck to the privacy funnel,” in 2014 IEEE Information Theory Workshop, 2014, pp. 501–505

    2014

  11. [19]

    Calibrating noise to sensitivity in private data analysis,

    C. Dwork, F. McSherry, K. Nissim, and A. Smith, “Calibrating noise to sensitivity in private data analysis,” in Theory of cryptography conference. Springer, 2006, pp. 265–284

    2006

  12. [20]

    Information extraction under privacy constraints,

    S. Asoodeh, M. Diaz, F. Alajaji, and T. Linder, “Information extraction under privacy constraints,” Information, vol. 7, no. 1, 2016. [Online]. Available: https://www.mdpi.com/2078-2489/7/1/15

    2016

  13. [21]

    Privacy-utility tradeoff with nonspecific tasks: Robust privatization and minimum leakage,

    T.-Y . Liu and I.-H. Wang, “Privacy-utility tradeoff with nonspecific tasks: Robust privatization and minimum leakage,” in 2020 IEEE In- formation Theory Workshop (ITW) , 2021, pp. 1–5

    2020

  14. [22]

    On the lift, related privacy measures, and applications to privacy–utility trade-offs,

    M. A. Zarrabian, N. Ding, and P. Sadeghi, “On the lift, related privacy measures, and applications to privacy–utility trade-offs,” Entropy, vol. 25, no. 4, 2023. [Online]. Available: https://www.mdpi.com/ 1099-4300/25/4/679

    2023

  15. [23]

    Robust privatization with non-specific tasks and the optimal privacy-utility tradeoff,

    T.-Y . Liu, I. Wang et al. , “Robust privatization with non-specific tasks and the optimal privacy-utility tradeoff,” arXiv preprint arXiv:2010.10081, 2020

    2010 arXiv

  16. [24]

    Utility-privacy tradeoffs in databases: An information-theoretic approach,

    L. Sankar, S. R. Rajagopalan, and H. V . Poor, “Utility-privacy tradeoffs in databases: An information-theoretic approach,” IEEE Transactions on Information Forensics and Security , vol. 8, no. 6, pp. 838–852, 2013

    2013

  17. [25]

    Optimal utility-privacy trade-off with total variation distance as a privacy measure,

    B. Rassouli and D. G ¨und¨uz, “Optimal utility-privacy trade-off with total variation distance as a privacy measure,” IEEE Transactions on Information Forensics and Security , vol. 15, pp. 594–603, 2020

    2020

  18. [26]

    Tunable measures for information leakage and applications to privacy-utility tradeoffs,

    J. Liao, O. Kosut, L. Sankar, and F. P. Calmon, “Tunable measures for information leakage and applications to privacy-utility tradeoffs,” IEEE Transactions on Information Theory , vol. 65, no. 12, pp. 8043–8066, Dec 2019

    2019

  19. [27]

    Latent feature disclosure under perfect sample privacy,

    B. Rassouli, F. Rosas, and D. G ¨und¨uz, “Latent feature disclosure under perfect sample privacy,” in 2018 IEEE International Workshop on Information Forensics and Security , Dec 2018, pp. 1–7

    2018

  20. [28]

    Estimation efficiency under privacy constraints,

    S. Asoodeh, M. Diaz, F. Alajaji, and T. Linder, “Estimation efficiency under privacy constraints,” IEEE Transactions on Information Theory , vol. 65, no. 3, pp. 1512–1534, March 2019

    2019

  21. [29]

    Privacy against statistical inference,

    F. P. Calmon and N. Fawaz, “Privacy against statistical inference,” in 2012 50th Annual Allerton Conference on Communication, Control, and Computing, Oct 2012, pp. 1401–1408

    2012

  22. [30]

    Information- theoretic approaches to privacy in estimation and control,

    E. Nekouei, T. Tanaka, M. Skoglund, and K. H. Johansson, “Information- theoretic approaches to privacy in estimation and control,” Annual Reviews in Control, 2019

    2019

  23. [31]

    An operational measure of in- formation leakage,

    I. Issa, S. Kamath, and A. B. Wagner, “An operational measure of in- formation leakage,” in 2016 Annual Conference on Information Science and Systems, March 2016, pp. 234–239

    2016

  24. [32]

    Active privacy-utility trade- off against inference in time-series data sharing,

    E. Erdemir, P. L. Dragotti, and D. G ¨und¨uz, “Active privacy-utility trade- off against inference in time-series data sharing,” IEEE Journal on Selected Areas in Information Theory , vol. 4, pp. 159–173, 2023

    2023

  25. [33]

    Statistical privacy mechanism design using separation technique,

    A. Zamani, T. J. Oechtering, and M. Skoglund, “Statistical privacy mechanism design using separation technique,” in 2024 32nd European Signal Processing Conference (EUSIPCO) . IEEE, 2024, pp. 735–739

    2024

  26. [34]

    Randomized response: A survey technique for eliminating evasive answer bias,

    S. L. Warner, “Randomized response: A survey technique for eliminating evasive answer bias,” Journal of the American Statistical Association , vol. 60, no. 309, pp. 63–69, 1965

    1965

  27. [35]

    The wire-tap channel,

    A. D. Wyner, “The wire-tap channel,” The Bell System Technical Journal, vol. 54, no. 8, pp. 1355–1387, 1975

    1975

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.