# Differential Privacy

A mathematically rigorous definition and framework for quantifying privacy risk and enabling privacy-preserving data analysis.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## Instance-Optimal Moment Estimation in the Add-Remove Model of Differential Privacy

DevFeed: [Instance-Optimal Moment Estimation in the Add-Remove Model of Differential Privacy](<https://devfeed.tech/articles/instance-optimal-moment-estimation-in-the-add-remove-model-of-differential-privacy-145055.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25425>)

Author: Shokichi Takakura

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>)

Tags: [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [differential-privacy](<https://devfeed.tech/tags/differential-privacy.md>), [error](<https://devfeed.tech/tags/error.md>), [lower-bounds](<https://devfeed.tech/tags/lower-bounds.md>), [means](<https://devfeed.tech/tags/means.md>), [stat-ml](<https://devfeed.tech/tags/stat-ml.md>)

### AI overview

The paper studies pure differentially private estimation of empirical means, variances, and covariances in the add-remove model. It introduces instance-dependent error metrics for evaluating performance on individual datasets, proves sharp lower bounds for those metrics, and designs mechanisms that attain the bounds. The abstract reports asymptotically better utility than existing methods, while noting that prior mechanisms can be worst-case optimal without being optimal for individual datasets.

### Source excerpt

arXiv:2611.25425v1 Announce Type: cross Abstract: In this paper, we study the problem of estimating empirical means, variances, and covariances under pure differential privacy in the add-remove model. While existing mechanisms achieve optimal worst-case error, they may not be optimal for individual datasets. To measure the performance of mechanisms on individual datasets, we introduce instance-dependent error metrics and prove sharp lower bounds on them. Then, we design mechanisms that achieve these lower bounds and demonstrate that they achieve asymptotically better utility compared to existing methods.

## Braid: Sybil-Resistant Decentralized Identity with Trustless Key Recovery and Non-Transferable Anonymous Credentials

DevFeed: [Braid: Sybil-Resistant Decentralized Identity with Trustless Key Recovery and Non-Transferable Anonymous Credentials](<https://devfeed.tech/articles/braid-sybil-resistant-decentralized-identity-with-trustless-key-recovery-and-non-transferable-anonymous-credentials-145990.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2307.14679>)

Author: Rui Song, Tianyu Zheng, Shang Gao, Guyue Li, Xue Liu, Bin Xiao

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [identity and access management](<https://devfeed.tech/topics/identity-and-access-management.md>), [Cryptocurrency](<https://devfeed.tech/topics/cryptocurrency.md>), [fake account fraud](<https://devfeed.tech/topics/fake-account-fraud.md>), [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>)

Tags: [anonymous-credentials](<https://devfeed.tech/tags/anonymous-credentials.md>), [blockchain](<https://devfeed.tech/tags/blockchain.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [decentralized](<https://devfeed.tech/tags/decentralized.md>), [identity](<https://devfeed.tech/tags/identity.md>), [key](<https://devfeed.tech/tags/key.md>), [privacy](<https://devfeed.tech/tags/privacy.md>), [recovery](<https://devfeed.tech/tags/recovery.md>), [resistance](<https://devfeed.tech/tags/resistance.md>), [zero-knowledge](<https://devfeed.tech/tags/zero-knowledge.md>)

### AI overview

Braid links blockchain identifiers into growing associations presumed to belong to one holder, carrying that state across applications to make Sybil activity harder to sustain and support recovery after lost keys or partial compromise. Recovery relies on proving knowledge of association structures. Blockchain identifier-key bindings and credential ownership proofs are intended to prevent unauthorized credential transfers, while presentation remains zero-knowledge. The abstract contrasts this persistent state with prior approaches limited to campaign-scoped Sybil protection or stateless credential presentation. Reported times on consumer-grade devices are 2.41 seconds for identifier association and 3.31 seconds for credential presentation.

### Source excerpt

arXiv:2307.14679v4 Announce Type: replace Abstract: Decentralized identity frameworks grant users full sovereignty over their digital assets in the Web3 ecosystem. However, allowing arbitrary creation of identifiers makes the system susceptible to Sybil attacks and puts assets at risk when keys are lost or compromised. Moreover, the lack of identification prevents anonymous credential schemes from deterring malicious transfers. Existing solutions link identifiers to entities through trusted intermediaries, which are not always accessible and require costly offline interactions. We introduce Braid, a decentralized identity system with Sybil resistance, trustless key recovery, and non-transferable anonymous credentials. Braid creates blockchain-based bindings between identifiers and gradually combines identifiers belonging to the same holder into a unified associated identifier. As all identifiers within an association are presumed to belong to one individual, any fraudulent activity can be detected. The association grows larger as interactions increase, substantially reducing the likelihood of successful Sybil attacks. This mechanism allows holders to recover identifiers with lost keys or partial key compromise by proving knowledge of specific association structures. Additionally, Braid prevents unauthorized transfers through blockchain-based identifier-key bindings and proofs of ownership for credentials. Compared with representative prior approaches that provide only campaign-scoped Sybil protection or stateless anonymous-credential presentation, Braid carries association state across applications, enabling progressive Sybil resistance and trustless recovery while preserving zero-knowledge credential presentation. The evaluation shows that Braid bridges privacy-preserving credentials and accountable identity, achieving identifier association and credential presentation times of 2.41s and 3.31s on consumer-grade devices.

## Privacy-Utility Tradeoffs with Quantum Local Differential Privacy

DevFeed: [Privacy-Utility Tradeoffs with Quantum Local Differential Privacy](<https://devfeed.tech/articles/privacy-utility-tradeoffs-with-quantum-local-differential-privacy-144876.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2602.10510>)

Author: Theshani Nuradha, Sujeet Bhalerao, Felix Leditzky

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>), [Security & Privacy](<https://devfeed.tech/topics/security-privacy.md>), [Quantum Computing](<https://devfeed.tech/topics/quantum-computing.md>)

Tags: [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-it](<https://devfeed.tech/tags/cs-it.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [differential-privacy](<https://devfeed.tech/tags/differential-privacy.md>), [learning](<https://devfeed.tech/tags/learning.md>), [local](<https://devfeed.tech/tags/local.md>), [math-it](<https://devfeed.tech/tags/math-it.md>), [privacy](<https://devfeed.tech/tags/privacy.md>), [quant-ph](<https://devfeed.tech/tags/quant-ph.md>), [quantum](<https://devfeed.tech/tags/quantum.md>), [tradeoffs](<https://devfeed.tech/tags/tradeoffs.md>), [utility](<https://devfeed.tech/tags/utility.md>)

### AI overview

Under (ε,δ)-quantum local differential privacy, the paper finds that the depolarizing mechanism is optimal for generic state utility measured by fidelity and trace distance, while task-specific mechanisms can improve utility for learning observable expectations. It derives a high-probability sample lower bound using private quantum hypothesis-testing results and reports optimal sample complexity in the privacy and accuracy parameters, scaling as Θ((εβ)⁻²) for ε ∈ (0,1) and accuracy tolerance β. The abstract also introduces private classical shadows as a direction for future applications.

### Source excerpt

arXiv:2602.10510v2 Announce Type: replace-cross Abstract: When sensitive information is encoded in data, it is important to ensure the privacy of information when attempting to learn useful information from the data. There is a natural tradeoff whereby increasing privacy requirements may decrease the utility of a learning protocol. In the quantum setting of differential privacy, such tradeoffs between privacy and utility have so far remained largely unexplored. In this work, we study optimal privacy-utility tradeoffs for both generic and application-specific utility metrics when privacy is quantified by $(\varepsilon,\delta)$-quantum local differential privacy. In the generic setting, we focus on optimizing fidelity and trace distance between the original state and the privatized state. We show that the depolarizing mechanism achieves the optimal utility for given privacy requirements. We then study the specific application of learning the expectation of an observable with respect to an input state when only given access to privatized states. We derive a lower bound on the number of samples of privatized data required to achieve a fixed accuracy guarantee with high probability. To prove this result, we employ existing lower bounds on private quantum hypothesis testing, thus showcasing the first operational use of them. We also devise private mechanisms that achieve optimal sample complexity with respect to the privacy parameters and accuracy parameters, demonstrating that utility can be significantly improved for specific tasks in contrast to the generic setting. In addition, we show that the number of samples required to privately learn observable expectation values scales as $\Theta((\varepsilon \beta)^{-2})$, where $\varepsilon \in (0,1)$ is the privacy parameter and $\beta$ is the accuracy tolerance. We conclude by initiating the study of private classical shadows, which promise useful applications for private learning tasks.

## Data Poisoning Attacks to Locally Differentially Private Triangle Counting Protocols

DevFeed: [Data Poisoning Attacks to Locally Differentially Private Triangle Counting Protocols](<https://devfeed.tech/articles/data-poisoning-attacks-to-locally-differentially-private-triangle-counting-protocols-143131.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23701>)

Author: Tingxuan Han, Wei Tong, Zijian Wang, Sheng Zhong

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>), [Adversarial attacks](<https://devfeed.tech/topics/adversarial-attacks.md>)

Tags: [adversarial](<https://devfeed.tech/tags/adversarial.md>), [analytics](<https://devfeed.tech/tags/analytics.md>), [analyze](<https://devfeed.tech/tags/analyze.md>), [attack](<https://devfeed.tech/tags/attack.md>), [attacks](<https://devfeed.tech/tags/attacks.md>), [bias](<https://devfeed.tech/tags/bias.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [data](<https://devfeed.tech/tags/data.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [framework](<https://devfeed.tech/tags/framework.md>), [global](<https://devfeed.tech/tags/global.md>), [graph](<https://devfeed.tech/tags/graph.md>), [guarantees](<https://devfeed.tech/tags/guarantees.md>), [integrity](<https://devfeed.tech/tags/integrity.md>), [interactive](<https://devfeed.tech/tags/interactive.md>), [local](<https://devfeed.tech/tags/local.md>), [malicious](<https://devfeed.tech/tags/malicious.md>), [manipulate](<https://devfeed.tech/tags/manipulate.md>), [private](<https://devfeed.tech/tags/private.md>), [protocols](<https://devfeed.tech/tags/protocols.md>)

### AI overview

The paper finds that malicious participants can significantly distort locally differentially private triangle count estimates. Because locally randomized reports make malicious users difficult to distinguish, small protocol-compliant manipulations can be amplified during reconstruction and calibration into large estimation bias. The authors develop a gain-driven attack framework for both non-interactive and interactive protocols and report experiments across four real-world graph datasets. They also find that existing and straightforward defenses are insufficient, pointing to the need for defenses tailored to LDP subgraph counting.

### Source excerpt

arXiv:2611.23701v1 Announce Type: new Abstract: Local differential privacy (LDP) has become a crucial paradigm for privacy-preserving graph analytics, enabling graph analytics tasks such as subgraph counting without requiring a trusted curator. However, while LDP provides strong privacy guarantees, the integrity and robustness of LDP subgraph counting protocols in the presence of malicious participants remain poorly understood. In particular, since data are randomized locally before collection, it is inherently difficult to distinguish malicious users from honest ones based solely on their submitted reports, and small, protocol-compliant local manipulations can be systematically amplified through reconstruction and calibration into large global estimation bias. In this paper, we use LDP triangle counting as a representative task to characterize the vulnerability of LDP subgraph counting to data poisoning attacks. For both non-interactive and interactive LDP triangle counting protocols, we analyze their attack surfaces and develop a unified gain-driven framework that captures how adversarial users can manipulate calibrated subgraph count estimates, which we instantiate on representative protocols of both types. Extensive experiments on four real-world graph datasets demonstrate that these attacks can significantly distort triangle count estimates across different protocols and adversarial settings. Moreover, we show that existing and straightforward defenses are insufficient to mitigate these attacks, underscoring the need for robustness mechanisms tailored to LDP subgraph counting.

## Tight Lower Bounds for Differentially Private Continual Counting

DevFeed: [Tight Lower Bounds for Differentially Private Continual Counting](<https://devfeed.tech/articles/tight-lower-bounds-for-differentially-private-continual-counting-141452.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2609.17650>)

Author: Charlie Harrison, Ethan Leeman

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>), [Algorithms](<https://devfeed.tech/topics/algorithms.md>)

Tags: [algorithm](<https://devfeed.tech/tags/algorithm.md>), [binary-tree](<https://devfeed.tech/tags/binary-tree.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-ds](<https://devfeed.tech/tags/cs-ds.md>), [differential-privacy](<https://devfeed.tech/tags/differential-privacy.md>), [distribution](<https://devfeed.tech/tags/distribution.md>), [error](<https://devfeed.tech/tags/error.md>), [gaussian](<https://devfeed.tech/tags/gaussian.md>), [lower-bounds](<https://devfeed.tech/tags/lower-bounds.md>), [mechanism](<https://devfeed.tech/tags/mechanism.md>), [noise](<https://devfeed.tech/tags/noise.md>), [stream](<https://devfeed.tech/tags/stream.md>)

### AI overview

For differential privacy with 0 ≤ ε ≤ 1 and 0 ≤ δ ≤ 1/4, the paper proves asymptotically tight lower bounds for ℓ∞ error, mean squared error, and maximum per-coordinate expected squared error in continual counting. The bounds apply to arbitrary mechanisms, including ones given the full stream in advance, and match the blocked Binary Tree Mechanism using whichever of Laplace or Gaussian noise performs better. This establishes asymptotic optimality across the stated error measures and privacy regimes. The proof uses one hard distribution that tags possible target flipped elements with a score evaluated down a tree; a simple score modification handles all three error measures.

### Source excerpt

arXiv:2609.17650v3 Announce Type: replace-cross Abstract: The Binary Tree Mechanism is a standard algorithm for differentially private continual counting, but its complete asymptotic optimality under differential privacy has remained unresolved since its introduction. We resolve this question. For $0 \leq \varepsilon \leq 1$, we prove asymptotically tight lower bounds for $\ell_\infty$, mean squared, and maximum per-coordinate expected squared error simultaneously under all regimes of $0 \leq \delta \leq 1/4$. These bounds hold for arbitrary mechanisms, even when the entire stream is available in advance. Our lower bounds match the (blocked) Binary Tree Mechanism instantiated with the best of Laplace or Gaussian noise, establishing its asymptotic optimality. Our proof uses a single hard distribution where elements are ``tagged'' as potentially being the target flipped element via a score evaluated down a tree. A simple modification of the score allows the same framework to establish tight lower bounds for all three error measures.

## Differentially Private Non-convex Distributionally Robust Optimization

DevFeed: [Differentially Private Non-convex Distributionally Robust Optimization](<https://devfeed.tech/articles/differentially-private-non-convex-distributionally-robust-optimization-141543.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2602.16155>)

Author: Difei Xu, Meng Ding, Zebin Ma, Huanyi Xie, Youming Tao, Aicha Slaitane, Di Wang

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>)

Tags: [adversarial](<https://devfeed.tech/tags/adversarial.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [differential-privacy](<https://devfeed.tech/tags/differential-privacy.md>), [experimental](<https://devfeed.tech/tags/experimental.md>), [kl-divergence](<https://devfeed.tech/tags/kl-divergence.md>), [minimax](<https://devfeed.tech/tags/minimax.md>), [optimization](<https://devfeed.tech/tags/optimization.md>), [private](<https://devfeed.tech/tags/private.md>), [robust](<https://devfeed.tech/tags/robust.md>), [utility](<https://devfeed.tech/tags/utility.md>)

### AI overview

The paper develops differentially private methods for non-convex distributionally robust optimization (DRO). For general ψ-divergences, it reformulates DRO as a minimization problem and introduces DP Double-Spider, with a gradient-norm utility bound of O(1/√n + (√(d log(1/δ))/(nε))^(2/3)) under mild assumptions. For KL-divergence, it recasts the problem as compositional finite-sum optimization and proposes DP Recursive-Spider, whose bound is O((√(d log(1/δ))/(nε))^(2/3)), stated to match the best-known non-convex DP-ERM result. The abstract reports that experiments outperform existing approaches for DP minimax optimization.

### Source excerpt

arXiv:2602.16155v2 Announce Type: replace Abstract: Real-world deployments routinely face distribution shifts, group imbalances, and adversarial perturbations, under which the traditional Empirical Risk Minimization (ERM) framework can degrade severely. Distributionally Robust Optimization (DRO) addresses this issue by optimizing the worst-case expected loss over an uncertainty set of distributions, offering a principled approach to robustness. Meanwhile, as training data in DRO always involves sensitive information, safeguarding it against leakage under Differential Privacy (DP) is essential. In contrast to classical DP-ERM, DP-DRO has received much less attention due to its minimax optimization structure with uncertainty constraint. To bridge the gap, we provide a comprehensive study of DP-(finite-sum)-DRO with $\psi$-divergence and non-convex loss. First, we study DRO with general $\psi$-divergence by reformulating it as a minimization problem, and develop a novel $(\varepsilon, \delta)$-DP optimization method, called DP Double-Spider, tailored to this structure. Under mild assumptions, we show that it achieves a utility bound of $\mathcal{O}(\frac{1}{\sqrt{n}}+ (\frac{\sqrt{d \log (1/\delta)}}{n \varepsilon})^{2/3})$ in terms of the gradient norm, where $n$ denotes the data size and $d$ denotes the model dimension. We further improve the utility rate for specific divergences. In particular, for DP-DRO with KL-divergence, by transforming the problem into a compositional finite-sum optimization problem, we develop a DP Recursive-Spider method and show that it achieves a utility bound of $\mathcal{O}((\frac{\sqrt{d \log(1/\delta)}}{n\varepsilon})^{2/3} )$, matching the best-known result for non-convex DP-ERM. Experimentally, we demonstrate that our proposed methods outperform existing approaches for DP minimax optimization.

## Quantifying Membership Inference Risks in Genomic Data Sharing: The Role of Attacker's Auxiliary Information

DevFeed: [Quantifying Membership Inference Risks in Genomic Data Sharing: The Role of Attacker's Auxiliary Information](<https://devfeed.tech/articles/quantifying-membership-inference-risks-in-genomic-data-sharing-the-role-of-attacker-s-auxiliary-information-138801.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.18909>)

Author: Maryam Ghasemian, Lynette Hammond Gerido, Erman Ayday

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>), [Security & Privacy](<https://devfeed.tech/topics/security-privacy.md>), [sensitive data](<https://devfeed.tech/topics/sensitive-data.md>), [Data analysis](<https://devfeed.tech/topics/data-analysis.md>)

Tags: [attack](<https://devfeed.tech/tags/attack.md>), [auditing](<https://devfeed.tech/tags/auditing.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [differential-privacy](<https://devfeed.tech/tags/differential-privacy.md>), [false-positive-rate](<https://devfeed.tech/tags/false-positive-rate.md>), [framework](<https://devfeed.tech/tags/framework.md>), [genomic-data](<https://devfeed.tech/tags/genomic-data.md>), [local](<https://devfeed.tech/tags/local.md>), [mechanism](<https://devfeed.tech/tags/mechanism.md>), [power](<https://devfeed.tech/tags/power.md>), [pre-release](<https://devfeed.tech/tags/pre-release.md>), [precision](<https://devfeed.tech/tags/precision.md>), [principal-component-analysis](<https://devfeed.tech/tags/principal-component-analysis.md>), [privacy](<https://devfeed.tech/tags/privacy.md>), [project](<https://devfeed.tech/tags/project.md>), [public](<https://devfeed.tech/tags/public.md>), [recall](<https://devfeed.tech/tags/recall.md>), [research](<https://devfeed.tech/tags/research.md>), [sharing](<https://devfeed.tech/tags/sharing.md>)

### AI overview

The proposed framework audits genomic datasets for membership inference risk before publication. It maps data into a shared PCA space, applies epsilon-local differential privacy with Laplace noise using a derived sensitivity for PCA scores, and measures a distance-threshold attack at a 5% false-positive rate. It also reports precision, recall, and an Unbiased Reconstruction Robustness membership-neutrality diagnostic. Experiments on openSNP and 1000 Genomes Project data across dimensionalities and privacy budgets identify separation between the research dataset and public auxiliary data in PCA space as a stronger determinant of attack success than perturbation level alone. A GWAS-style likelihood ratio test provides a comparison baseline.

### Source excerpt

arXiv:2611.18909v1 Announce Type: cross Abstract: Genomic data sharing is essential for biomedical research but introduces significant privacy risks, including membership inference attacks, in which an adversary determines whether a specific individual's data was used in a study. Existing privacy protections are typically applied after a release format has already been chosen, leaving researchers without a principled method to assess dataset-specific risk beforehand. We propose a server-assisted, principal component analysis-based pre-release auditing framework that maps genomic data into a shared low-dimensional space, applies epsilon-local differential privacy via the Laplace mechanism with a formally derived sensitivity for principal component analysis scores, and estimates membership inference attack power at a controlled 5% false-positive rate using a distance-threshold attack. A membership-neutrality diagnostic based on Unbiased Reconstruction Robustness is reported alongside attack power, precision, and recall. Experiments using the openSNP and 1000 Genomes Project datasets across multiple dimensionalities and privacy budgets show that separation between the researcher's dataset and public auxiliary data in principal component analysis space is a dominant determinant of attack success, exceeding the effect of perturbation level alone. Comparison against a genome-wide association study-style likelihood ratio test baseline further contextualizes the measured risk. This framework enables researchers, data repositories, and institutional review bodies to quantify dataset-specific membership inference risk before genomic data publication, supporting more informed and privacy-aware biomedical data-sharing decisions.

## On the Price of Privacy for Language Identification and Generation

DevFeed: [On the Price of Privacy for Language Identification and Generation](<https://devfeed.tech/articles/on-the-price-of-privacy-for-language-identification-and-generation-139229.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2604.07238>)

Author: Xiaoyu Li, Andi Han, Jiaojiao Jiang, Junbin Gao

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>), [LLM Techniques](<https://devfeed.tech/topics/llm-techniques.md>)

Tags: [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-ds](<https://devfeed.tech/tags/cs-ds.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [differential-privacy](<https://devfeed.tech/tags/differential-privacy.md>), [generation](<https://devfeed.tech/tags/generation.md>), [language](<https://devfeed.tech/tags/language.md>), [language-learning](<https://devfeed.tech/tags/language-learning.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [privacy](<https://devfeed.tech/tags/privacy.md>)

### AI overview

In agnostic language learning, differential privacy affects generation more than identification. For identification, pure DP and approximate DP with δ >= n^−O(1) retain the non-private subexponential error rates exp(−r(n)) for every r(n)=o(n), at fixed ε. For generation, a public witness bound limits pure DP's cost to a Θ(min{1, ε}) factor in the exponent; on suitable two-language collections this rate is optimal. Without that bound, pure DP can no longer achieve an exponential rate, though every exp(−o(n)) rate remains achievable. On those collections, approximate DP achieves an exponential rate exactly when log(1/δ)=o(n).

### Source excerpt

arXiv:2604.07238v2 Announce Type: replace-cross Abstract: Large language models train on sensitive user data, so we ask what differential privacy (DP) costs in language learning. We study private language identification and generation in the agnostic statistical setting. For identification, privacy is free: at every fixed $\varepsilon$, pure DP and $(\varepsilon,\delta)$-DP with $\delta\ge n^{-O(1)}$ keep the non-private rate $\exp(-r(n))$ for every $r(n)=o(n)$. For generation, the price depends on what is public. With a public witness bound, pure DP loses at most a factor $\Theta(\min\{1,\varepsilon\})$ in the exponent, and on suitable two-language collections the optimal rate is $\exp(-\Theta(\min\{1,\varepsilon\}\,n))$. Without such a bound, pure DP can lose the exponential rate itself, already for two languages: every rate $\exp(-o(n))$ stays achievable, but no exponential rate is. On these collections, $(\varepsilon,\delta)$-DP keeps an exponential rate if and only if $\log(1/\delta)=o(n)$. Three facts about pure DP drive the proofs. It turns a robustness margin into an exponentially small error. It obeys a private form of Le Cam's two-point bound. And it cannot see the absence of a string.

## Evidence-Conditioned Completion and Synthesis of Electronic Health Records

DevFeed: [Evidence-Conditioned Completion and Synthesis of Electronic Health Records](<https://devfeed.tech/articles/evidence-conditioned-completion-and-synthesis-of-electronic-health-records-137460.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.19546>)

Author: Sicheng Shu, Parker Knight, Huitao Li, Yuan Luo, Yifan Peng, Shawn N. Murphy, Rui Yang, Chuan Hong

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>), [Representation learning](<https://devfeed.tech/topics/representation-learning.md>), [sensitive data](<https://devfeed.tech/topics/sensitive-data.md>), [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>)

Tags: [completion](<https://devfeed.tech/tags/completion.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [electronic-health-records](<https://devfeed.tech/tags/electronic-health-records.md>), [health](<https://devfeed.tech/tags/health.md>), [risk](<https://devfeed.tech/tags/risk.md>), [stat-ap](<https://devfeed.tech/tags/stat-ap.md>), [synthesis](<https://devfeed.tech/tags/synthesis.md>)

### AI overview

EHRDiB unifies EHR visit completion and synthesis with a shared multimodal representation and an evidence-conditioned latent diffusion bridge, distinguishing original missingness from withheld inputs. Across MIMIC-IV and numerical and discretized Duke Health data, it had the smallest utility gaps in seven of eight completion comparisons. For synthesis, AUPRC gaps were 45.2% smaller on ten MIMIC-IV principal-diagnosis targets and 53.3% smaller on numerical Duke Health data than with the best comparator. Calibrated thresholding produced the lowest marginal distribution error across all three datasets, with coverage above 0.95 under both output rules. In MIMIC-IV synthesis, membership-inference AUROC rose from 0.516 to 0.598 as mean target evidence fraction increased from 0.20 to 0.50; utility gaps narrowed relative to the lowest-evidence setting but not monotonically, indicating a measured disclosure-risk tradeoff.

### Source excerpt

arXiv:2611.19546v1 Announce Type: new Abstract: Patient information can guide electronic health record (EHR) generation, but how its amount shapes predictive utility and disclosure risk remains poorly characterized. Here we present EHRDiB, which unifies visit completion and synthesis by distinguishing original missingness from withheld inputs and combining a shared multimodal representation with an evidence-conditioned latent diffusion bridge. We compared EHRDiB with three completion and eight synthesis methods on MIMIC-IV and numerical and discretized Duke Health representations. Evaluation covered downstream predictive utility, generative diversity, and distributional fidelity for both tasks, and disclosure risk for synthesis. EHRDiB achieved the smallest utility gaps from real-data-trained classifiers in seven of eight completion comparisons. In synthesis, AUPRC gaps were 45.2% smaller for ten MIMIC-IV principal-diagnosis targets and 53.3% smaller for numerical Duke Health data relative to the best comparator. Synthesis with calibrated thresholding yielded the lowest marginal distribution error across all three datasets, with Coverage above 0.95 for both output rules. Across twelve evidence-fraction intervals in MIMIC-IV synthesis, membership-inference AUROC increased from 0.516 to 0.598 as the mean target evidence fraction rose from 0.20 to 0.50; utility gaps narrowed relative to the lowest evidence configuration but did not decrease monotonically. Our study provides a unified framework and empirical foundation for EHR generation that considers both its utility for clinical analysis and its information disclosure risk.

## Privacy-Preserving Clustering Algorithm and Parameter Selection for Collaborative Population Stratification Using Distributed PCA

DevFeed: [Privacy-Preserving Clustering Algorithm and Parameter Selection for Collaborative Population Stratification Using Distributed PCA](<https://devfeed.tech/articles/privacy-preserving-clustering-algorithm-and-parameter-selection-for-collaborative-population-stratification-using-distributed-pca-138805.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.18925>)

Author: Maryam Ghasemian, Lynette Hammond Gerido, Erman Ayday

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [clustering](<https://devfeed.tech/topics/clustering.md>), [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>), [Hyperparameter optimization](<https://devfeed.tech/topics/hyperparameter-optimization.md>)

Tags: [clustering](<https://devfeed.tech/tags/clustering.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [distributed](<https://devfeed.tech/tags/distributed.md>), [parameter](<https://devfeed.tech/tags/parameter.md>), [privacy](<https://devfeed.tech/tags/privacy.md>), [selection](<https://devfeed.tech/tags/selection.md>)

### AI overview

The framework selects clustering methods and hyperparameters from privacy-protected data for collaborative population stratification. Each party computes local top-r principal-component information; a server aggregates low-rank summaries to form a global basis without a public reference basis. It supports three locally differentially private pipelines: PCA then Noise, Noise then PCA, and Noise-Only. The server scores configurations with internal validation metrics on perturbed data, a post-processing step. Experiments on openSNP and 1000 Genomes report that PCA-based pipelines outperform Noise-Only in clustering utility, communication cost, and runtime, and have lower membership-inference attack power at small privacy budgets. These findings apply to the reported experimental setting.

### Source excerpt

arXiv:2611.18925v1 Announce Type: cross Abstract: We propose a privacy-preserving selection layer for collaborative population stratification workflows, enabling an honest-but-curious server to recommend a strong clustering algorithm-hyperparameter configuration from privacy protected shares. Instead of using a fixed workflow such as PCA followed by K-Means with a preset number of clusters, the framework compares multiple clustering methods and parameter settings under privacy constraints. To avoid relying on a public reference basis, we construct a shared low-dimensional representation through a distributed-PCA (DPCA) procedure in which each party computes local top-$r$ principal-component information and the server aggregates only low-rank summaries to recover a global basis. Building on this representation, the framework supports three LDP-protected pipelines: PCA then Noise, Noise then PCA, and Noise-Only. The server evaluates clustering methods and parameter settings using internal validation metrics and recommends a top-performing configuration from the perturbed data; this step is pure post-processing. Experiments on openSNP and 1000 Genomes show that the server-side selection layer reliably identifies strong settings from LDP-protected data. Across both datasets, the PCA-based pipelines outperform Noise-Only in clustering utility, communication cost, and runtime, and show lower membership-inference attack (MIA) power at small privacy budgets. Overall, they provide the best utility--scalability trade-off in our setting.

## A Maximum Entropy Implementation of Differential Privacy Under Linear Invariants

DevFeed: [A Maximum Entropy Implementation of Differential Privacy Under Linear Invariants](<https://devfeed.tech/articles/a-maximum-entropy-implementation-of-differential-privacy-under-linear-invariants-139238.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2607.22450>)

Author: Ryan Lafferty, Anindya Roy

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>), [Security, Privacy and Abuse Prevention](<https://devfeed.tech/topics/security-privacy-and-abuse-prevention.md>)

Tags: [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [differential-privacy](<https://devfeed.tech/tags/differential-privacy.md>), [entropy](<https://devfeed.tech/tags/entropy.md>), [implementation](<https://devfeed.tech/tags/implementation.md>), [invariants](<https://devfeed.tech/tags/invariants.md>), [linear](<https://devfeed.tech/tags/linear.md>), [stat-me](<https://devfeed.tech/tags/stat-me.md>)

### AI overview

The paper proposes a high-entropy differential privacy implementation for settings with fixed linear aggregates, or invariants, such as totals that cannot be perturbed. Since enforcing these constraints makes the noise vector dependent, the authors derive privacy guarantees for their implementation under the invariants. They say it preserves the constraints with probability one or exponentially close to one, and that the proof resolves an open question about null spaces of correlation matrices. The abstract also identifies sampling from normal mixture models under linear equality constraints as a broader application.

### Source excerpt

arXiv:2607.22450v2 Announce Type: replace Abstract: Differential privacy is the standard for ensuring data privacy and is widely used in major data publications, including reporting results from the U.S. decennial census. Common implementation of differential privacy uses independent Gaussian or Laplace noise addition to the database. However, there could be aggregate (linear) queries to the database that are excluded from the privacy budget, for example, state totals that can not be perturbed due to constitutional mandates. Any implementation of a differential privacy is required to honor these constraints, also referred to as invariants. Under aggregation constraints, the noise vector is no longer independent and the traditional differential privacy guarantees have to be re-evaluated. We propose a high entropy differential privacy implementation that maintains the aggregation invariants with probability one or exponentially close to one and derive the privacy guarantees for the implementation under the invariants. The theoretical proof also gives a solution to an open question about the null space of correlation matrices. Moreover, the methodology has general use in the context of sampling from normal mixture models under linear equality constraints.

## Online Allocation with Differential Privacy

DevFeed: [Online Allocation with Differential Privacy](<https://devfeed.tech/articles/online-allocation-with-differential-privacy-136029.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.17554>)

Author: Jianyi Yang, Xingyu Zhou, Mostafa Mushsharat, Shaolei Ren

Published: 2026-11-16T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>), [Operations research and optimization](<https://devfeed.tech/topics/operations-research-and-optimization.md>), [Security & Privacy](<https://devfeed.tech/topics/security-privacy.md>)

Tags: [adversarial](<https://devfeed.tech/tags/adversarial.md>), [ai-model-routing](<https://devfeed.tech/tags/ai-model-routing.md>), [algorithms](<https://devfeed.tech/tags/algorithms.md>), [battery](<https://devfeed.tech/tags/battery.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [differential-privacy](<https://devfeed.tech/tags/differential-privacy.md>), [online](<https://devfeed.tech/tags/online.md>), [performance](<https://devfeed.tech/tags/performance.md>), [privacy](<https://devfeed.tech/tags/privacy.md>)

### AI overview

The paper proposes PPOA, a meta-algorithm for private online allocation that aims to balance differential privacy with allocation performance under limited resources. It gives conditions for asymptotically near-optimal performance with Joint DP (JDP) or Local DP (LDP), and describes two instantiations: PPOA-DMD, which supports both JDP and LDP, and PPOA-FTRL, which supports JDP. The authors analyze both adversarial and stochastic settings, characterize privacy-performance trade-offs, and report numerical experiments on AI model routing in battery-powered edge systems. The supplied text does not include detailed guarantees or experimental results.

### Source excerpt

arXiv:2611.17554v1 Announce Type: new Abstract: We study private online allocation problems, where allocation decisions must satisfy differential privacy (DP) to protect sensitive user information while optimizing performance under limited resources. We first formalize differential privacy notions suited to different privacy requirements in online allocation and establish the fundamental performance limits of any algorithm under these definitions. Based on these insights, we propose PPOA, a privacy-preserving meta algorithm for online allocation, and establish sufficient conditions under which PPOA preserves Joint DP (JDP) or Local DP (LDP) while achieving asymptotically near-optimal performance. Building on these conditions, PPOA can be instantiated into a variety of concrete algorithms. In particular, we present the specific designs, which include PPOA-DMD that preserves JDP and LDP and PPOA-FTRL that preserves JDP. We analyze their performance under both adversarial and stochastic settings and characterize the fundamental trade-offs between DP and allocation performance. Finally, we demonstrate the superior performance of the proposed algorithms via numerical experiments on AI model routing in battery-powered edge systems.

## Tractable Importance Sampling for Monte Carlo Privacy Accounting

DevFeed: [Tractable Importance Sampling for Monte Carlo Privacy Accounting](<https://devfeed.tech/articles/tractable-importance-sampling-for-monte-carlo-privacy-accounting-135802.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.17596>)

Author: Galen Andrew, Arun Ganesh

Published: 2026-11-16T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>), [MCNP](<https://devfeed.tech/topics/mcnp.md>)

Tags: [accounting](<https://devfeed.tech/tags/accounting.md>), [complexity](<https://devfeed.tech/tags/complexity.md>), [convex](<https://devfeed.tech/tags/convex.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [differential-privacy](<https://devfeed.tech/tags/differential-privacy.md>), [distribution](<https://devfeed.tech/tags/distribution.md>), [distributions](<https://devfeed.tech/tags/distributions.md>), [estimator](<https://devfeed.tech/tags/estimator.md>), [factor](<https://devfeed.tech/tags/factor.md>), [failure](<https://devfeed.tech/tags/failure.md>), [gaussian](<https://devfeed.tech/tags/gaussian.md>), [guarantees](<https://devfeed.tech/tags/guarantees.md>), [matrix](<https://devfeed.tech/tags/matrix.md>), [monte-carlo](<https://devfeed.tech/tags/monte-carlo.md>), [privacy](<https://devfeed.tech/tags/privacy.md>), [probability](<https://devfeed.tech/tags/probability.md>), [protocol](<https://devfeed.tech/tags/protocol.md>), [root](<https://devfeed.tech/tags/root.md>), [sample](<https://devfeed.tech/tags/sample.md>), [sampling](<https://devfeed.tech/tags/sampling.md>), [stretch](<https://devfeed.tech/tags/stretch.md>)

### AI overview

For DP matrix factorization with privacy amplification by sampling, standard Monte Carlo accounting can require many samples to estimate the small failure probabilities associated with strict privacy guarantees. The abstract proposes shifted importance sampling with a stretch factor α > 1. It establishes a closed-form support bound B(α) < 1 that applies to both directions of the hockey-stick divergence, and shows B is log-convex, making selection of the optimal α a one-dimensional root-finding problem. The estimator fits the KL-Hoeffding verification protocol used in prior work; the stated sample complexity is reduced by a factor of exactly 1/B(α*).

### Source excerpt

arXiv:2611.17596v1 Announce Type: cross Abstract: Evaluating tight Differential Privacy (DP) parameters for DP matrix factorization with privacy amplification by sampling requires analyzing high-dimensional, correlated Gaussian mixtures. Because integrating these mixture distributions to find the exact privacy failure probability ($\delta$) is analytically intractable, these mechanisms are typically analyzed using Monte Carlo accounting. However, standard sampling suffers from prohibitive variance when estimating small $\delta$ associated with strict privacy guarantees, which in turn requires a large sample complexity, making Monte Carlo accounting inefficient. We demonstrate that Monte Carlo accounting can be made much more efficient via importance sampling. We give a shifted importance sampling proposal distribution parameterized by a stretch factor $\alpha > 1$. We prove a closed-form bound $B(\alpha) < 1$ on the support of the importance-weighted estimator that holds identically for both directions of the hockey-stick divergence, show that $B$ is log-convex so that the optimal $\alpha$ is a one-dimensional root-finding problem, and show that the resulting estimator seamlessly plugs into the KL-Hoeffding verification protocol used by past work, with a sample complexity reduced by exactly a factor of $1/B(\alpha^*)$.

## Optimal Privacy-Utility Trade-Offs in LDP: Functional and Geometric Perspectives

DevFeed: [Optimal Privacy-Utility Trade-Offs in LDP: Functional and Geometric Perspectives](<https://devfeed.tech/articles/optimal-privacy-utility-trade-offs-in-ldp-functional-and-geometric-perspectives-133252.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2605.02319>)

Author: Seung-Hyun Nam, Hyun-Young Park, Si-Hyeon Lee

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>), [Operations research and optimization](<https://devfeed.tech/topics/operations-research-and-optimization.md>)

Tags: [bayesian](<https://devfeed.tech/tags/bayesian.md>), [channel](<https://devfeed.tech/tags/channel.md>), [convex](<https://devfeed.tech/tags/convex.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-it](<https://devfeed.tech/tags/cs-it.md>), [differential-privacy](<https://devfeed.tech/tags/differential-privacy.md>), [dpi](<https://devfeed.tech/tags/dpi.md>), [information](<https://devfeed.tech/tags/information.md>), [linear-programming](<https://devfeed.tech/tags/linear-programming.md>), [math-it](<https://devfeed.tech/tags/math-it.md>), [minimax](<https://devfeed.tech/tags/minimax.md>), [optimization](<https://devfeed.tech/tags/optimization.md>), [privacy](<https://devfeed.tech/tags/privacy.md>), [symmetry](<https://devfeed.tech/tags/symmetry.md>), [utility](<https://devfeed.tech/tags/utility.md>)

### AI overview

The paper presents a unified framework for characterizing optimal privacy-utility trade-offs and locally differentially private channels across statistical decision problems. It identifies properties of Bayesian and minimax risks--including data processing, quasi-convexity or additivity, concavity, and symmetry invariance--that reduce the optimization domain. A geometric characterization links maximal channels under the Blackwell order to a finite-dimensional polytope, making computation possible through vertex enumeration or linear programming. With transitive symmetries, the trade-off can instead be found by comparing finitely many values. The framework also covers information-theoretic measure maximization; the authors report recovering or strengthening known results and deriving exact expressions for some previously unaddressed tasks.

### Source excerpt

arXiv:2605.02319v2 Announce Type: replace Abstract: Local differential privacy (LDP) has emerged as a gold-standard framework for privacy-preserving data analysis. However, characterizing the optimal privacy-utility trade-off (PUT) and the corresponding optimal LDP channels remains largely fragmented, relying on problem-specific, case-by-case analyses. In this work, we develop a unified theoretical framework that systematically characterizes the optimal PUT and optimal LDP channels for general privacy-preserving statistical decision-making problems. We first identify key functional properties of Bayesian and minimax risks as functions of the LDP channel, including the data processing inequality (DPI), direct-sum quasi-convexity (or additivity), concavity, and symmetry invariance. Leveraging these properties, we reduce the optimization domain required to compute the optimal PUT. To this end, we reinterpret the existing result implying the one-to-one correspondence between maximal LDP channels under the Blackwell order and a finite-dimensional polytope, through an alternative convex geometric derivation. This result renders the optimal PUT computationally tractable via vertex enumeration or linear programming. Furthermore, when the underlying problem exhibits symmetries characterized by a transitive group action, we show that the optimal PUT can be computed by comparing finitely many values, rather than optimizing over a high dimensional polytope. Our framework applies broadly beyond risk minimization, encompassing the maximization of information-theoretic measures. We demonstrate the efficacy of our theoretical framework by recovering or strengthening several known results, and deriving exact analytic expressions for the optimal PUTs in specific tasks that were previously unaddressed.

## DP-MacAdam: Differentially Private Mechanism with Adaptive Clipping and Adaptive Momentum

DevFeed: [DP-MacAdam: Differentially Private Mechanism with Adaptive Clipping and Adaptive Momentum](<https://devfeed.tech/articles/dp-macadam-differentially-private-mechanism-with-adaptive-clipping-and-adaptive-momentum-132802.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2606.05435>)

Author: Naima Tasnim, Lalitha Sankar, Oliver Kosut

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [clipping](<https://devfeed.tech/tags/clipping.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [mechanism](<https://devfeed.tech/tags/mechanism.md>), [momentum](<https://devfeed.tech/tags/momentum.md>), [privacy-preserving-machine-learning](<https://devfeed.tech/tags/privacy-preserving-machine-learning.md>), [private](<https://devfeed.tech/tags/private.md>)

### AI overview

DP-MacAdam combines adaptive clipping from AdaClip with Adam-like momentum from DP-Adam, using shared empirical gradient mean and variance estimates for both. The authors report a bias-free variance estimate and experiments showing improved model utility over DP-SGD, AdaClip, and DP-Adam, without manual clipping-threshold tuning. The supplied text does not give experimental details or quantify the gains.

### Source excerpt

arXiv:2606.05435v2 Announce Type: replace Abstract: Differentially private stochastic gradient descent (DP-SGD) has become the standard framework for privacy-preserving machine learning, yet its reliance on a fixed gradient clipping threshold to limit sensitivity remains a significant practical limitation. Adaptive clipping algorithms such as AdaClip shift and scale the gradient prior to clipping and adding noise so that the clipped gradient yields a more informative descent direction. The shift and scaling parameters are selected adaptively based on the empirical mean and variance. However, in existing adaptive clipping algorithms, these empirical estimates have not been also used for momentum to accelerate training itself. On the other hand, DP-Adam is an algorithm that exploits Adam-like momentum updates based on the gradient mean and variance to accelerate training, but does not exploit these estimates for adaptive clipping. In this work, we propose Differentially Private Mechanism with Adaptive Clipping and Adaptive Momentum (DP-MacAdam), a novel algorithm that combines these two approaches so as to use the same mean and variance estimates for both clipping and momentum. We perform an analysis showing that DP-MacAdam estimates the gradient variances in a bias-free manner. In addition, we empirically evaluate the privacy and accuracy of DP-MacAdam, demonstrating that it achieves improved model utility compared to DP-SGD, AdaClip, and DP-Adam baselines, without requiring manual tuning of the clipping threshold.

## Privacy-Preserving Range Counting over Local Graph Statistics

DevFeed: [Privacy-Preserving Range Counting over Local Graph Statistics](<https://devfeed.tech/articles/privacy-preserving-range-counting-over-local-graph-statistics-132701.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.16483>)

Author: Gengda Zhao, Yizhang He, Dong Wen, Kai Wang, Ruijia Wu, Jianwei Wang

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Databases](<https://devfeed.tech/sources/arxiv-databases.md>)

Topics: [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>), [Graphs](<https://devfeed.tech/topics/graphs.md>), [Data science and engineering](<https://devfeed.tech/topics/data-science-and-engineering.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [adversary](<https://devfeed.tech/tags/adversary.md>), [aes](<https://devfeed.tech/tags/aes.md>), [bias](<https://devfeed.tech/tags/bias.md>), [boundary](<https://devfeed.tech/tags/boundary.md>), [bounded](<https://devfeed.tech/tags/bounded.md>), [cs-db](<https://devfeed.tech/tags/cs-db.md>), [graph](<https://devfeed.tech/tags/graph.md>), [local](<https://devfeed.tech/tags/local.md>), [privacy](<https://devfeed.tech/tags/privacy.md>), [range](<https://devfeed.tech/tags/range.md>), [statistics](<https://devfeed.tech/tags/statistics.md>)

### AI overview

The paper proposes an edge-local-DP framework for estimating ranges of vertex statistics while limiting disclosure of sensitive graph edges: each vertex randomizes its report before sending it to an untrusted curator. It combines scalar statistic estimators, directly privatized range indicators, and adaptive estimator selection. Its Bounded Conditional Range Inversion (BCRI) uses reusable bounded geometric reports for unbiased range estimates; a boundary-conditioned version has expected L2 loss no greater than the unbounded inverse, and histogram preprocessing enables constant-time queries. The reports also yield a nonnegative unbiased variance estimate. Adaptive Estimator Selection (AES) privately predicts bias, variance, and covariance to combine BCRI with a conditional estimator from the same transcript. The abstract reports experiments showing accuracy, efficiency, and workload scalability across diverse graph tasks.

### Source excerpt

arXiv:2611.16483v1 Announce Type: new Abstract: Range counting over vertex statistics supports contact analysis, community profiling, and reputation screening. However, releasing exact range counts can disclose sensitive edges to an adversary with background knowledge. We study this problem under edge local differential privacy (edge LDP), where each vertex randomizes its report before sending it to an untrusted curator. We develop a layered framework that connects statistic estimation, direct range estimation, and adaptive estimator selection. Statistic-estimation methods provide complementary scalar estimators with explicit loss analyses, while indicator-estimation methods privatize range indicators directly. We next propose Bounded Conditional Range Inversion (BCRI), which derives unbiased range estimates from reusable bounded geometric reports through an explicit local inverse. Its boundary-conditioned estimator achieves no larger expected L2 loss than the unbounded inverse, and histogram preprocessing supports constant-time queries. The same reports also provide a nonnegative, unbiased variance estimate for every answer. At the center of the framework, Adaptive Estimator Selection (AES) privately predicts bias, variance, and covariance to combine BCRI with a conditional estimator from the same transcript. Experiments on diverse real-world graph tasks demonstrate the accuracy, efficiency, and workload scalability of the proposed framework.

## Differential Privacy for Euclidean Jordan Algebra with Applications to Private Symmetric Cone Programming

DevFeed: [Differential Privacy for Euclidean Jordan Algebra with Applications to Private Symmetric Cone Programming](<https://devfeed.tech/articles/differential-privacy-for-euclidean-jordan-algebra-with-applications-to-private-symmetric-cone-programming-128485.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2509.16915>)

Author: Zhao Song, Jianfei Xue, Lichen Zhang

Published: 2026-11-11T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>)

Tags: [algebra](<https://devfeed.tech/tags/algebra.md>), [algorithms](<https://devfeed.tech/tags/algorithms.md>), [application](<https://devfeed.tech/tags/application.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-ds](<https://devfeed.tech/tags/cs-ds.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [differential-privacy](<https://devfeed.tech/tags/differential-privacy.md>), [gaussian](<https://devfeed.tech/tags/gaussian.md>), [linear-programming](<https://devfeed.tech/tags/linear-programming.md>), [math-oc](<https://devfeed.tech/tags/math-oc.md>), [matrices](<https://devfeed.tech/tags/matrices.md>), [optimization](<https://devfeed.tech/tags/optimization.md>)

### AI overview

The paper proposes a generic Gaussian mechanism for differentially private functions whose outputs lie in a Euclidean Jordan algebra, with sensitivity measured in ℓ₂, ℓ₁, or ℓ∞ norms. For symmetric-matrix outputs, these correspond to considering Frobenius, nuclear, or spectral norm sensitivity. It combines this mechanism with multiplicative weights to derive private algorithms for symmetric cone programs under various settings, including semidefinite programming; the authors say this resolves an open question posed by Hsu, Roth, Roughgarden, and Ullman at ICALP 2014.

### Source excerpt

arXiv:2509.16915v2 Announce Type: replace-cross Abstract: In this paper, we study differentially private mechanisms for functions whose outputs lie in a Euclidean Jordan algebra. Euclidean Jordan algebras capture many important mathematical structures and form the foundation of linear programming, second-order cone programming, and semidefinite programming. Our main contribution is a generic Gaussian mechanism for such functions, with sensitivity measured in $\ell_2$, $\ell_1$, and $\ell_\infty$ norms. Notably, this framework includes the important case where the function outputs are symmetric matrices, and sensitivity is measured in the Frobenius, nuclear, or spectral norm. We further derive private algorithms for solving symmetric cone programs under various settings, using a combination of the multiplicative weights update method and our generic Gaussian mechanism. As an application, we present differentially private algorithms for semidefinite programming, resolving a major open question posed by [Hsu, Roth, Roughgarden, and Ullman, ICALP 2014].

## SnapAudit: Active Auditing of Differentially Private In-Context Learning via Snapshot-Based Simulation

DevFeed: [SnapAudit: Active Auditing of Differentially Private In-Context Learning via Snapshot-Based Simulation](<https://devfeed.tech/articles/snapaudit-active-auditing-of-differentially-private-in-context-learning-via-snapshot-based-simulation-128879.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2511.13502>)

Author: Yuyang Xia, Ruixuan Liu, Li Xiong

Published: 2026-11-11T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [aggregation](<https://devfeed.tech/tags/aggregation.md>), [also](<https://devfeed.tech/tags/also.md>), [analysis](<https://devfeed.tech/tags/analysis.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [attacks](<https://devfeed.tech/tags/attacks.md>), [audit](<https://devfeed.tech/tags/audit.md>), [auditing](<https://devfeed.tech/tags/auditing.md>), [audits](<https://devfeed.tech/tags/audits.md>), [based](<https://devfeed.tech/tags/based.md>), [bootstrap](<https://devfeed.tech/tags/bootstrap.md>), [clock](<https://devfeed.tech/tags/clock.md>), [contain](<https://devfeed.tech/tags/contain.md>), [context](<https://devfeed.tech/tags/context.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [data](<https://devfeed.tech/tags/data.md>), [learning](<https://devfeed.tech/tags/learning.md>), [private](<https://devfeed.tech/tags/private.md>), [simulation](<https://devfeed.tech/tags/simulation.md>), [snapshot](<https://devfeed.tech/tags/snapshot.md>)

### AI overview

SnapAudit audits differentially private in-context learning more efficiently by separating near-deterministic clean inference from stochastic noise. It samples a limited set of clean model outputs, then bootstraps many noisy trials, using an empirical Bernstein correction to account for finite-sample uncertainty. For embedding-based mechanisms, it searches for maximally separable audit signals. The abstract reports tighter, more stable privacy estimates than passive auditing and a 4.3-5.5x reduction in wall-clock time across four pipelines and five privacy budgets; snapshots can be reused across noise configurations. It also reports two privacy flaws: Gaussian noise calibrations can underestimate leakage at large privacy budgets, and an embedding mechanism's sensitivity analysis fails when there is one partition, resulting in undersized noise.

### Source excerpt

arXiv:2511.13502v4 Announce Type: replace Abstract: In-context learning (ICL) enables LLMs to adapt to tasks using demonstrations that may contain sensitive data. Differentially private (DP) ICL mechanisms mitigate this risk by adding noise during aggregation. However, verifying their claimed privacy bounds through repeated end-to-end membership-inference attacks is costly and yields unstable estimates. We propose SnapAudit, an active auditing framework that decomposes a DP-ICL pipeline into a deterministic clean-inference stage and a stochastic DP-noise stage, and audits the full pipeline by combining a small snapshot of the former with bootstrap simulation of the latter. Because clean LLM outputs are near-deterministic at low temperature, tens of thousands of clean LLM calls suffice to approximate the snapshot distribution; SnapAudit then bootstraps hundreds of thousands of noisy trials at negligible additional cost, controlling finite-sample uncertainty via an empirical Bernstein correction. For embedding-based mechanisms, a multi-sweep search constructs maximally separable audit signals. Compared with passive auditing, SnapAudit yields tighter, more stable privacy estimates that closely match theoretical guarantees. Across privacy budgets, SnapAudit reuses snapshots across noise configurations, whereas passive auditing requires fresh end-to-end trials. Across four DP-ICL pipelines and five privacy budgets, this reduces total wall-clock auditing time by $4.3$--$5.5\times$. SnapAudit also uncovers two flaws in existing DP-ICL designs: (i) classical Gaussian noise calibrations underestimate leakage at large privacy budgets, allowing empirical leakage to exceed the theoretical bound; (ii) an embedding-aggregation mechanism's sensitivity analysis is incorrect when the number of partitions equals one, leading to undersized noise and a privacy violation.

## Optimal Byzantine-resilient Federated Learning with User-level Differential Privacy

DevFeed: [Optimal Byzantine-resilient Federated Learning with User-level Differential Privacy](<https://devfeed.tech/articles/optimal-byzantine-resilient-federated-learning-with-user-level-differential-privacy-125279.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.10898>)

Author: Ming Xiang, Stratis Ioannidis, Edmund Yeh, Carlee Joe-Wong, Lili Su

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Federated Learning](<https://devfeed.tech/topics/federated-learning.md>), [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>)

Tags: [adversarial](<https://devfeed.tech/tags/adversarial.md>), [aggregation](<https://devfeed.tech/tags/aggregation.md>), [algorithm](<https://devfeed.tech/tags/algorithm.md>), [analysis](<https://devfeed.tech/tags/analysis.md>), [central](<https://devfeed.tech/tags/central.md>), [client](<https://devfeed.tech/tags/client.md>), [clients](<https://devfeed.tech/tags/clients.md>), [convergence](<https://devfeed.tech/tags/convergence.md>), [convex](<https://devfeed.tech/tags/convex.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [data](<https://devfeed.tech/tags/data.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [differential-privacy](<https://devfeed.tech/tags/differential-privacy.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [federated-learning](<https://devfeed.tech/tags/federated-learning.md>), [gradient](<https://devfeed.tech/tags/gradient.md>), [guarantees](<https://devfeed.tech/tags/guarantees.md>), [guards](<https://devfeed.tech/tags/guards.md>), [information](<https://devfeed.tech/tags/information.md>), [lower-bound](<https://devfeed.tech/tags/lower-bound.md>), [math-oc](<https://devfeed.tech/tags/math-oc.md>), [privacy](<https://devfeed.tech/tags/privacy.md>)

### AI overview

The paper establishes an information-theoretic lower bound for Byzantine-resilient empirical risk minimization under user-level differential privacy with heterogeneous clients, then proposes ByzULDP, a central-DP method using momentum on both clients and server. Client momentum reduces stochastic gradient variance during robust aggregation. The authors say ordinary nonlinear Byzantine-resilient aggregation combined with central-DP noise may fail to ensure convergence near a stationary point; server momentum uses the way client updates determine user-level privacy sensitivity to support convergence. ByzULDP achieves an order-optimal bound for strongly convex objectives, with error terms decaying as O(1/T) for strongly convex and O(1/√T) for non-convex objectives. Numerical experiments on real-world datasets across privacy budgets support the analysis.

### Source excerpt

arXiv:2611.10898v1 Announce Type: new Abstract: Real-world deployment of federated learning (FL) requires resilience against adversarial clients and data privacy. Byzantine fault captures the worst-case behavior of adversarial clients, while differential privacy (DP) provides statistical guarantees against privacy loss. In particular, user-level DP guards a client's entire contribution and provides more realistic protection against information leakage. Despite intensive efforts, the minimax-optimal convergence bound of Byzantine-resilient FL with user-level DP remains open. We establish an information-theoretic lower bound for any Byzantine-resilient empirical risk minimization problem under user-level DP and heterogeneous clients. We propose ByzULDP, a central DP algorithm featuring a two-level momentum across the client and server. Client-side momentum mitigates stochastic gradient variance during robust aggregation. Unlike the fault-free case, naively combining non-linear Byzantine-resilient aggregation with central DP noise can be insufficient to guarantee convergence to a bounded neighborhood of a stationary point. Our key contribution is a server-side momentum that leverages a property specific to user-level DP, where privacy sensitivity depends on how client updates are constructed, to empower convergence. ByzULDP achieves the order-optimal convergence bound for strongly-convex objectives, with vanishing error terms decaying at the optimal $O(1/T)$ and $O(1/\sqrt{T})$ rates for strongly-convex and non-convex objectives, respectively. We corroborate our analysis with numerical experiments on real-world datasets across diverse privacy budgets.

## Optimal Rates for Adaptive Private $k$-PCA

DevFeed: [Optimal Rates for Adaptive Private $k$-PCA](<https://devfeed.tech/articles/optimal-rates-for-adaptive-private-k-pca-125550.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.12419>)

Author: Johanna D\\"ungler, Amartya Sanyal

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [algorithm](<https://devfeed.tech/tags/algorithm.md>), [computer-science](<https://devfeed.tech/tags/computer-science.md>), [covariance](<https://devfeed.tech/tags/covariance.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [differential-privacy](<https://devfeed.tech/tags/differential-privacy.md>), [efficiency](<https://devfeed.tech/tags/efficiency.md>), [intrinsic](<https://devfeed.tech/tags/intrinsic.md>), [linear-algebra](<https://devfeed.tech/tags/linear-algebra.md>), [lower-bounds](<https://devfeed.tech/tags/lower-bounds.md>), [matrices](<https://devfeed.tech/tags/matrices.md>), [noise](<https://devfeed.tech/tags/noise.md>), [privacy](<https://devfeed.tech/tags/privacy.md>), [private](<https://devfeed.tech/tags/private.md>), [sample](<https://devfeed.tech/tags/sample.md>), [stat-ml](<https://devfeed.tech/tags/stat-ml.md>), [variance](<https://devfeed.tech/tags/variance.md>)

### AI overview

The proposed differentially private stochastic PCA algorithm jointly updates a full dxk subspace and claims optimal sample complexity in both d and k, retaining adaptive privacy noise and support for any target dimension k≤d from the earlier deflation-based \kdppca. The authors also say it removes a prior utility guarantee's minimum-sample requirement with exponential eigengap dependence and improves spectral-parameter dependence to match known lower bounds for the spiked covariance model. Joint updates simplify the algorithm, reduce hyperparameters, and may improve efficiency with block linear algebra libraries. These are claims in the supplied abstract; it provides no empirical results or detailed conditions for the guarantees.

### Source excerpt

arXiv:2611.12419v1 Announce Type: new Abstract: Given $n$ i.i.d. random matrices $A_i \in \mathbb{R}^{d \times d}$ with common expectation $\Sigma$, the goal of \emph{Differentially Private Stochastic PCA} is to identify a $k$-dimensional subspace capturing the leading variance directions of $\Sigma$, while preserving differential privacy~(DP) for each individual sample $A_i$. \citet{dungler2025iterative} introduced \kdppca, the first algorithm to simultaneously (1) achieve sample complexity $n=\widetilde O(d)$ for sub-Gaussian data, (2) adapt its privacy noise to the intrinsic randomness of the data, and (3) extend seamlessly to any target dimension $k\le d$. However, its sample complexity has a suboptimal dependence on $k$. We propose the first algorithm that achieves optimal sample complexity in both $d$ and $k$, while retaining properties (2) and (3) of \kdppca. Additionally, prior utility guarantees require a minimum sample size with exponential dependence on the eigengap. Our method removes this dependence. It also improves the dependence on spectral parameters, matching known lower bounds for the spiked covariance model. Unlike deflation-based approaches like \kdppca, our algorithm updates the full $d\times k$ subspace jointly rather than one eigenvector at a time. This non-deflation structure simplifies the algorithm, reduces the number of hyper-parameters, and improves computational efficiency, particularly when using block linear algebra libraries.

## Enforcing Demographic Coherence: A Harms Aware Framework for Reasoning about Private Data Release

DevFeed: [Enforcing Demographic Coherence: A Harms Aware Framework for Reasoning about Private Data Release](<https://devfeed.tech/articles/enforcing-demographic-coherence-a-harms-aware-framework-for-reasoning-about-private-data-release-125651.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2502.02709>)

Author: Mark Bun, Marco Carmosino, Palak Jain, Gabriel Kaptchuk, Satchit Sivakumar

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>), [arXiv: Databases](<https://devfeed.tech/sources/arxiv-databases.md>)

Topics: [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>), [enterprise data protection](<https://devfeed.tech/topics/enterprise-data-protection.md>), [Adversarial attacks](<https://devfeed.tech/topics/adversarial-attacks.md>)

Tags: [adversarial](<https://devfeed.tech/tags/adversarial.md>), [algorithms](<https://devfeed.tech/tags/algorithms.md>), [assessment](<https://devfeed.tech/tags/assessment.md>), [attack](<https://devfeed.tech/tags/attack.md>), [attacks](<https://devfeed.tech/tags/attacks.md>), [breaches](<https://devfeed.tech/tags/breaches.md>), [coherence](<https://devfeed.tech/tags/coherence.md>), [confidence](<https://devfeed.tech/tags/confidence.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-db](<https://devfeed.tech/tags/cs-db.md>), [data](<https://devfeed.tech/tags/data.md>), [differential-privacy](<https://devfeed.tech/tags/differential-privacy.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [formal](<https://devfeed.tech/tags/formal.md>), [framework](<https://devfeed.tech/tags/framework.md>), [gap](<https://devfeed.tech/tags/gap.md>), [guarantees](<https://devfeed.tech/tags/guarantees.md>), [intuition](<https://devfeed.tech/tags/intuition.md>), [literature](<https://devfeed.tech/tags/literature.md>), [patterns](<https://devfeed.tech/tags/patterns.md>), [private](<https://devfeed.tech/tags/private.md>), [release](<https://devfeed.tech/tags/release.md>)

### AI overview

The paper proposes demographic coherence as a way to assess privacy risks from individual inferences that conflict with demographic patterns in the data. It aims to bridge formal privacy guarantees and attacks that may be criticized as unrealistic, and focuses on confidence-rated predictors that can be distilled from data-informed processes, so risks can arise even without an explicit attack. The framework offers a condition for analyzing data releases and suggests experimental evaluations to assess risks. The authors prove that differential privacy implies demographic coherence, while some demographically coherent algorithms are not differentially private; thus their condition is weaker than differential privacy.

### Source excerpt

arXiv:2502.02709v2 Announce Type: replace-cross Abstract: The technical literature about data privacy largely consists of two complementary approaches: formal definitions of conditions sufficient for privacy preservation and attacks that demonstrate privacy breaches. Differential privacy is an accepted standard in the former sphere. However, differential privacy's powerful adversarial model and worst-case guarantees may make it too stringent in some situations, especially when achieving it comes at a significant cost to data utility. Meanwhile, privacy attacks aim to expose real and worrying privacy risks associated with existing data release processes but often face criticism for being unrealistic. Moreover, the literature on attacks generally does not identify what properties are necessary to defend against them. We address the gap between these approaches by introducing demographic coherence, a condition inspired by privacy attacks that we argue is necessary for data privacy. This condition captures privacy violations arising from inferences about individuals that are incoherent with respect to the demographic patterns in the data. Our framework focuses on confidence rated predictors, which can in turn be distilled from almost any data-informed process. Thus, we capture privacy threats that exist even when no attack is explicitly being carried out. Our framework not only provides a condition with respect to which data release algorithms can be analysed but suggests natural experimental evaluation methodologies that could be used to build practical intuition and make tangible assessment of risks. Finally, we argue that demographic coherence is weaker than differential privacy: we prove that every differentially private data release is also demographically coherent, and that there are demographically coherent algorithms which are not differentially private.

## GSynPEF: A Privacy Evaluation Framework for Synthetic Graphs

DevFeed: [GSynPEF: A Privacy Evaluation Framework for Synthetic Graphs](<https://devfeed.tech/articles/gsynpef-a-privacy-evaluation-framework-for-synthetic-graphs-125895.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.10185>)

Author: Yu Duan, Luca Baldesi, Carter T. Butts, Athina Markopoulou

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>), [graph theory](<https://devfeed.tech/topics/graph-theory.md>)

Tags: [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-si](<https://devfeed.tech/tags/cs-si.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [framework](<https://devfeed.tech/tags/framework.md>), [graphs](<https://devfeed.tech/tags/graphs.md>), [privacy](<https://devfeed.tech/tags/privacy.md>), [synthetic](<https://devfeed.tech/tags/synthetic.md>)

### AI overview

GSynPEF empirically evaluates privacy and utility when a real graph is provided and one or more synthetic graphs are observed. It connects statistical and deep-learning graph synthesizers to privacy metrics and evaluates multiple outputs to account for variation. The framework supports an Exact Solution scenario, where synthesizer behavior is observable, and a Monte Carlo Approximation for black-box synthesizers. Experiments across four real-world datasets and seven synthesizers assess different attack goals, such as inferring an edge or node, and attacker capabilities, including access to one or multiple synthetic graphs. The paper also examines how graph statistics relate to synthesizer design choices to inform synthesizer selection and development.

### Source excerpt

arXiv:2611.10185v1 Announce Type: cross Abstract: It is often desirable to generate and release synthetic graphs, which resemble real-world data, in a privacy-preserving way. While substantial progress has been made in designing private graph synthesizers (PGS), opportunities remain for more explicit connections between synthesizer models and privacy evaluation, and better accounting for output variation. This paper proposes GSynPEF, a practical framework for empirical evaluation of privacy and utility in settings for which a real-world graph is given as input and output synthetic graph(s) are observed. Our framework links a broad class of graph synthesizers (both statistical and deep-learning based) with privacy metrics, expanding evaluation dimensions to practical utility and risk, through empirical analysis of multiple actual outputs. The framework accommodates both an Exact Solution scenario, where synthesizer behavior is directly observable, and a Monte Carlo Approximation scenario where only synthetic outputs from a black box synthesizer are available. Extensive experiments on four real-world datasets and seven widely used graph synthesizers demonstrate framework's versatility, enabling privacy evaluation for different attack goals (e.g., to infer an edge or node) and capabilities (e.g., observing one or multiple synthetic graphs). We analyze evaluation results by examining the relationship between graph statistics and synthesizer design choices, providing insights for data curators in PGS selection and researchers for improved PGS design.

## Sampling-Free Privacy Accounting for Matrix Mechanisms under Random Allocation

DevFeed: [Sampling-Free Privacy Accounting for Matrix Mechanisms under Random Allocation](<https://devfeed.tech/articles/sampling-free-privacy-accounting-for-matrix-mechanisms-under-random-allocation-126874.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2601.21636>)

Author: Jan Schuchardt, Nikita Kalinin

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>), [Operations research and optimization](<https://devfeed.tech/topics/operations-research-and-optimization.md>)

Tags: [accounting](<https://devfeed.tech/tags/accounting.md>), [composition](<https://devfeed.tech/tags/composition.md>), [conditional](<https://devfeed.tech/tags/conditional.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [dynamic-programming](<https://devfeed.tech/tags/dynamic-programming.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [matrix](<https://devfeed.tech/tags/matrix.md>), [model](<https://devfeed.tech/tags/model.md>), [model-training](<https://devfeed.tech/tags/model-training.md>), [monte-carlo](<https://devfeed.tech/tags/monte-carlo.md>), [privacy](<https://devfeed.tech/tags/privacy.md>), [private](<https://devfeed.tech/tags/private.md>), [probability](<https://devfeed.tech/tags/probability.md>), [programming](<https://devfeed.tech/tags/programming.md>), [random](<https://devfeed.tech/tags/random.md>), [research](<https://devfeed.tech/tags/research.md>), [sampling](<https://devfeed.tech/tags/sampling.md>), [small](<https://devfeed.tech/tags/small.md>), [stat-ml](<https://devfeed.tech/tags/stat-ml.md>)

### AI overview

The paper presents sampling-free privacy bounds for differentially private model training with matrix factorization under random allocation. It uses Rényi divergence bounds, computed efficiently with dynamic programming, and conditional composition to strengthen guarantees when ε is small, where Rényi bounds over-approximate privacy loss. The framework covers banded and non-banded matrices. Compared with the cited sampling-based Monte Carlo approach, it avoids guarantees that hold only with high probability or require random abstention, and avoids a sample requirement inversely proportional to δ. Numerical comparisons report efficacy across a broad range of matrix mechanisms; the abstract gives no specific benchmark results.

### Source excerpt

arXiv:2601.21636v4 Announce Type: replace-cross Abstract: We study privacy amplification for differentially private model training with matrix factorization under random allocation (also known as the balls-in-bins model). Recent work by Choquette-Choo et al. (2025) proposes a sampling-based Monte Carlo approach to compute amplification parameters in this setting. However, their guarantees either only hold with some high probability or require random abstention by the mechanism. Furthermore, the required number of samples for ensuring $(\epsilon,\delta)$-DP is inversely proportional to $\delta$. In contrast, we develop sampling-free bounds based on R\'enyi divergence and conditional composition. The former is facilitated by a dynamic programming formulation to efficiently compute the bounds. The latter complements it by offering stronger privacy guarantees for small $\epsilon$, where R\'enyi divergence bounds inherently lead to an over-approximation. Our framework applies to arbitrary banded and non-banded matrices. Through numerical comparisons, we demonstrate the efficacy of our approach across a broad range of matrix mechanisms used in research and practice.

## PLDP-IDS: Personalized Local Differential Privacy for Infinite Data Stream

DevFeed: [PLDP-IDS: Personalized Local Differential Privacy for Infinite Data Stream](<https://devfeed.tech/articles/pldp-ids-personalized-local-differential-privacy-for-infinite-data-stream-125670.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.11467>)

Author: Leilei Du, Yikun Hu, Xu Zhou, Yang Cao, Peng Cheng, Wangze Ni, Kenli Li

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Databases](<https://devfeed.tech/sources/arxiv-databases.md>)

Topics: [Differential Privacy](<https://devfeed.tech/topics/differential-privacy.md>), [streaming data analytics](<https://devfeed.tech/topics/streaming-data-analytics.md>)

Tags: [bias](<https://devfeed.tech/tags/bias.md>), [cs-db](<https://devfeed.tech/tags/cs-db.md>), [differential-privacy](<https://devfeed.tech/tags/differential-privacy.md>), [frequency-estimation](<https://devfeed.tech/tags/frequency-estimation.md>), [fusion](<https://devfeed.tech/tags/fusion.md>), [groups](<https://devfeed.tech/tags/groups.md>), [ops](<https://devfeed.tech/tags/ops.md>), [privacy](<https://devfeed.tech/tags/privacy.md>), [real-world](<https://devfeed.tech/tags/real-world.md>), [sampling](<https://devfeed.tech/tags/sampling.md>), [scheduling](<https://devfeed.tech/tags/scheduling.md>), [streaming](<https://devfeed.tech/tags/streaming.md>), [streaming-data](<https://devfeed.tech/tags/streaming-data.md>), [synthetic](<https://devfeed.tech/tags/synthetic.md>)

### AI overview

The paper proposes personalized continual frequency estimation under local differential privacy, letting each user choose a privacy window and budget. It identifies two challenges: heterogeneous requirements make one sampling size ineffective across groups, and combining estimates from higher-budget groups can introduce bias when privacy preferences correlate with data. OPS guides PLPD and PLPA; PLPD+ and PLPA+ add group-aware population scheduling and bias-aware cross-group fusion while preserving personalized w-event LDP. In experiments on real and synthetic streams, PLPA+ lowers average mean squared error by 52.3% and 40.8%, respectively, versus a direct personalized extension of LDP-IDS, and is about five times more accurate than PLPA under strong privacy-data correlation.

### Source excerpt

arXiv:2611.11467v1 Announce Type: new Abstract: Streaming data collection under local differential privacy (LDP) enables real-time analytics without a trusted curator. Existing $w$-event LDP mechanisms, however, impose a uniform privacy window and budget on all users, forcing everyone to adopt the most stringent requirement. We study personalized continual frequency estimation under LDP, where each user specifies her own window size and privacy budget, and identify two challenges. First, a single sampling size cannot balance sampling and perturbation errors for privacy groups with heterogeneous requirements. Second, borrowing estimates from groups with larger budgets reduces variance but introduces bias when privacy preferences correlate with the data. We first propose Optimal Population Selection (OPS) and two OPS-guided mechanisms, PLPD and PLPA. We then develop PLPD+ and PLPA+, which combine group-aware population scheduling with bias-aware cross-group fusion while preserving personalized w-event LDP. Experiments on real-world and synthetic streams show that PLPA+ reduces the AMSE of the direct personalized extension of LDP-IDS by 52.3% and 40.8% on average, respectively, and remains about $5\times$ more accurate than PLPA under strong privacy--data correlation.

[Next page](<https://devfeed.tech/topics/differential-privacy.md?cursor=WyIyMDI2LTExLTEwVDA1OjAwOjAwKzAwOjAwIiwgIjNlMDM3M2JkLWJiOTAtNDY5ZC04YWQ3LTIyMWUwN2FlYThiMCJd>)