arXiv subjects

Rachel Cummings

Publications and source records attributed to Rachel Cummings.

At least 19 recordsLinked to original sources

arXiv2026

Sample Complexity of Peer Prediction

Peer prediction seeks to incentivize agents to truthfully report an observed signal by rewarding joint sets of reports without observing a ground truth. Following the generalization of information-theoretic mutual information introduced in Kong and Schoenebeck (2019), we call a function of a joint distribution over signals a mutual information when it is non-negative and disincentivizes garbling reports for all information structures. An unbiased estimator for a mutual information takes some number of samples from the distribution and returns rewards for both agents, such that the expected reward is equal to the mutual information. We seek to characterize the set of mutual informations with unbiased estimators for a given number of samples. We show that for three or fewer sampled report pairs, the only mutual information with an unbiased estimator is trivially zero, and for four or five samples with a binary report space, the Determinant Mutual Information (DMI) of Kong (2024) is the unique mutual information (up to a scalar multiple). We further show that DMI ceases to be unique at six samples. We provide an improved estimator of DMI for any given number of samples and characterize its convergence rate. We also examine mutual information estimators that accept a randomized number of samples. First, we show that mutual information estimators on an ex-ante bounded number of samples (termed "stop-short estimators") can achieve a lower variance than an equivalent fixed-sample estimator (for DMI). Second, we introduce the class of scoring-rule-based mutual informations and identify in this family a mutual information that can be estimated with under three samples in expectation.

cs.IT↗

arXiv2026

A Complexity Measure for Active Learning in Multi-group Mean Estimation

We study a \emph{max-risk} objective for active learning in a multi-group mean estimation $d$-armed bandits: a learner adaptively allocates a budget of $T$ samples across $d$ groups to minimize the worst-case uncertainty index $\max_{k\in[d]}\sigma_k^2/n_k$, where $\sigma_k$ is the standard deviation of the distribution of arm $d$, and $n_k$ is the number of times arm $d$ is sampled. We develop a local minimax framework and prove the first general lower bound for this objective, valid for any finite-variance hypothesis class. The bound separates difficulty into three orthogonal factors: a \emph{budget} term, a \emph{heteroscedasticity} index measuring how unevenly the uncertainty is spread across arms, and a model-dependent complexity measure, the \emph{Variance Local Curvature} ($\mathrm{VLC}$), which captures how much information a local change of variance creates inside the hypothesis class. For smooth classes, the $\mathrm{VLC}$ is a reparametrization of a variance--Fisher information, with closed-form values for common families. Benchmarking against the strongest available upper bound shows near-optimality up to logarithmic factors in broad regimes, and pinpoints a systematic gap in highly heterogeneous instances. Our proof introduces two key ingredients: a loss-induced $\ell_1$ geometry on the decision space, and a representation-based instance generator that reduces hard-instance construction to an explicit random matrix calculation.

cs.LG↗

arXiv2026

ContinuousBench: Can Differentially Private Synthetic Text Improve Capabilities?

Differentially private (DP) text synthesis promises to unlock sensitive corpora for model training, but it remains unclear whether DP synthetic data transmits genuinely new knowledge and capabilities present only in those corpora. This is because existing evaluations rely on tasks that are nearly solvable without training, so strong benchmark performance does not establish that DP synthesis can substitute original data access. Thus, we introduce ContinuousBench, a continuously and automatically-regenerated benchmark that measures capability gain from DP synthetic text. Each quarter, a new release pairs a never-before-seen training corpus with a derived QA set, constructed to be: (1) unsolvable sans-corpus; and (2) learnable under DP, as the tested knowledge is supported by hundreds of independent records. Researchers produce DP synthetic data from the training corpus and run our standardized training and evaluation harness on their synthetic data to measure gains. We instantiate two tracks: Geminon, a procedurally-generated dataset about fictional creatures; and News, a stream of newly crawled public news articles. Although standard benchmarks are nearly saturated, on ContinuousBench we find that non-private synthesis transfers substantial knowledge from the original corpus, while state-of-the-art DP synthesis methods generally fail to do so, even at $\varepsilon=100$.

cs.LG↗

arXiv2026

Integrating Feature Correlation in Differential Privacy with Applications in DP-ERM

Standard differential privacy imposes uniform privacy constraints across all features, overlooking the inherent distinction between sensitive and insensitive features in practice. In this paper, we introduce a relaxed definition of differential privacy that accounts for such heterogeneity, allowing certain features to be treated as insensitive even when correlated with sensitive ones. We propose a correlation-aware framework, $\textsf{CorrDP}$, which relaxes privacy for insensitive features while accounting for their correlations with sensitive features, with the correlations quantified using total variation distance. We design algorithms for differentially private empirical risk minimization (DP-ERM) under the $\textsf{CorrDP}$ framework, incorporating distance-dependent noise into gradients for improved theoretical utility guarantees. When the correlation distance is unknown, we estimate it from the dataset and show that it achieves a comparable privacy-utility guarantee. We perform experiments on synthetic and real-world datasets and show that $\textsf{CorrDP}$-based DP-ERM algorithms consistently outperform the standard DP framework in the presence of insensitive features.

cs.LG↗

arXiv2025

"Having Confidence in My Confidence Intervals": How Data Users Engage with Privacy-Protected Wikipedia Data

In response to calls for open data and growing privacy threats, organizations are increasingly adopting privacy-preserving techniques such as differential privacy (DP) that inject statistical noise when generating published datasets. These techniques are designed to protect privacy of data subjects while enabling useful analyses, but their reception by data users is under-explored. We developed documentation that presents the noise characteristics of two Wikipedia pageview datasets: one using rounding (heuristic privacy) and another using DP (formal privacy). After incorporating expert feedback (n=5), we used these documents to conduct a task-based contextual inquiry (n=15) exploring how data users--largely unfamiliar with these methods--perceive, interact with, and interpret privacy-preserving noise during data analysis. Participants readily used simple uncertainty metrics from the documentation, but struggled when asked to compute confidence intervals across multiple noisy estimates. They were better able to devise simulation-based approaches for computing uncertainty with DP data compared to rounded data. Surprisingly, several participants incorrectly believed DP's stronger utility implied weaker privacy protections. Based on our findings, we offer design recommendations for documentation and tools to better support data users working with privacy-noised data.

cs.HC↗

arXiv2025

Beyond Laplace and Gaussian: Exploring the Generalized Gaussian Mechanism for Private Machine Learning

Differential privacy (DP) is obtained by randomizing a data analysis algorithm, which necessarily introduces a tradeoff between its utility and privacy. Many DP mechanisms are built upon one of two underlying tools: Laplace and Gaussian additive noise mechanisms. We expand the search space of algorithms by investigating the Generalized Gaussian (GG) mechanism, which samples the additive noise term $x$ with probability proportional to $e^{-\frac{| x |}{\sigma}^{\beta} }$ for some $\beta \geq 1$ (denoted $GG_{\beta, \sigma}(f,D)$). The Laplace and Gaussian mechanisms are special cases of GG for $\beta=1$ and $\beta=2$, respectively. We prove that the full GG family satisfies differential privacy and extend the PRV accountant to support privacy loss computation for these mechanisms. We then instantiate the GG mechanism in two canonical private learning pipelines, PATE and DP-SGD. Empirically, we explore PATE and DP-SGD with the GG mechanism across the computationally feasible values of $\beta$: $\beta \in [1,2]$ for DP-SGD and $\beta \in [1,4]$ for PATE. For both mechanisms, we find that $\beta=2$ (Gaussian) performs as well as or better than other values in their computational tractable domains.This provides justification for the widespread adoption of the Gaussian mechanism in DP learning.

cs.LG↗

arXiv2025

Differentially Private Space-Efficient Algorithms for Counting Distinct Elements in the Turnstile Model

The turnstile continual release model of differential privacy captures scenarios where a privacy-preserving real-time analysis is sought for a dataset evolving through additions and deletions. In typical applications of real-time data analysis, both the length of the stream $T$ and the size of the universe $|U|$ from which data come can be extremely large. This motivates the study of private algorithms in the turnstile setting using space sublinear in both $T$ and $|U|$. In this paper, we give the first sublinear space differentially private algorithms for the fundamental problem of counting distinct elements in the turnstile streaming model. Our algorithm achieves, on arbitrary streams, $\tilde{O}_{\eta}(T^{1/3})$ space and additive error, and a $(1+\eta)$-relative approximation for all $\eta \in (0,1)$. Our result significantly improves upon the space requirements of the state-of-the-art algorithms for this problem, which is linear, approaching the known $\Omega(T^{1/4})$ additive error lower bound for arbitrary streams. Moreover, when a bound $W$ on the number of times an item appears in the stream is known, our algorithm provides $\tilde{O}_{\eta}(\sqrt{W})$ additive error, using $\tilde{O}_{\eta}(\sqrt{W})$ space. This additive error asymptotically matches that of prior work which required instead linear space. Our results address an open question posed by [Jain, Kalemaj, Raskhodnikova, Sivakumar, Smith, Neurips23] about designing low-memory mechanisms for this problem. We complement these results with a space lower bound for this problem, which shows that any algorithm that uses similar techniques must use space $\tilde{\Omega}(T^{1/3})$ on arbitrary streams.

cs.DS↗

arXiv2025

An active learning framework for multi-group mean estimation

We study a fundamental learning problem over multiple groups with unknown data distributions, where an analyst would like to learn the mean of each group. Moreover, we want to ensure that this data is collected in a relatively fair manner such that the noise of the estimate of each group is reasonable. In particular, we focus on settings where data are collected dynamically, which is important in adaptive experimentation for online platforms or adaptive clinical trials for healthcare. In our model, we employ an active learning framework to sequentially collect samples with bandit feedback, observing a sample in each period from the chosen group. After observing a sample, the analyst updates their estimate of the mean and variance of that group and chooses the next group accordingly. The analyst's objective is to dynamically collect samples to minimize the collective noise of the estimators, measured by the norm of the vector of variances of the mean estimators. We propose an algorithm, Variance-UCB, that sequentially selects groups according to an upper confidence bound on the variance estimate. We provide a general theoretical framework for providing efficient bounds on learning from any underlying distribution where the variances can be estimated reasonably. This framework yields upper bounds on regret that improve significantly upon all existing bounds, as well as a collection of new results for different objectives and distributions than those previously studied.

cs.LG↗

arXiv2025

ClusterSC: Advancing Synthetic Control with Donor Selection

In causal inference with observational studies, synthetic control (SC) has emerged as a prominent tool. SC has traditionally been applied to aggregate-level datasets, but more recent work has extended its use to individual-level data. As they contain a greater number of observed units, this shift introduces the curse of dimensionality to SC. To address this, we propose Cluster Synthetic Control (ClusterSC), based on the idea that groups of individuals may exist where behavior aligns internally but diverges between groups. ClusterSC incorporates a clustering step to select only the relevant donors for the target. We provide theoretical guarantees on the improvements induced by ClusterSC, supported by empirical demonstrations on synthetic and real-world datasets. The results indicate that ClusterSC consistently outperforms classical SC approaches.

cs.LG↗

arXiv2024

Differential Privacy Under Class Imbalance: Methods and Empirical Insights

Imbalanced learning occurs in classification settings where the distribution of class-labels is highly skewed in the training data, such as when predicting rare diseases or in fraud detection. This class imbalance presents a significant algorithmic challenge, which can be further exacerbated when privacy-preserving techniques such as differential privacy are applied to protect sensitive training data. Our work formalizes these challenges and provides a number of algorithmic solutions. We consider DP variants of pre-processing methods that privately augment the original dataset to reduce the class imbalance; these include oversampling, SMOTE, and private synthetic data generation. We also consider DP variants of in-processing techniques, which adjust the learning algorithm to account for the imbalance; these include model bagging, class-weighted empirical risk minimization and class-weighted deep learning. For each method, we either adapt an existing imbalanced learning technique to the private setting or demonstrate its incompatibility with differential privacy. Finally, we empirically evaluate these privacy-preserving imbalanced learning methods under various data and distributional settings. We find that private synthetic data methods perform well as a data pre-processing step, while class-weighted ERMs are an alternative in higher-dimensional settings where private synthetic data suffers from the curse of dimensionality.

cs.LG↗

arXiv2024

Models Matter: Setting Accurate Privacy Expectations for Local and Central Differential Privacy

Differential privacy is a popular privacy-enhancing technology that has been deployed both in industry and government agencies. Unfortunately, existing explanations of differential privacy fail to set accurate privacy expectations for data subjects, which depend on the choice of deployment model. We design and evaluate new explanations of differential privacy for the local and central models, drawing inspiration from prior work explaining other privacy-enhancing technologies. We find that consequences-focused explanations in the style of privacy nutrition labels that lay out the implications of differential privacy are a promising approach for setting accurate privacy expectations. Further, we find that while process-focused explanations are not enough to set accurate privacy expectations, combining consequences-focused explanations with a brief description of how differential privacy works leads to greater trust.

cs.CR↗

arXiv2024

Thompson Sampling Itself is Differentially Private

In this work we first show that the classical Thompson sampling algorithm for multi-arm bandits is differentially private as-is, without any modification. We provide per-round privacy guarantees as a function of problem parameters and show composition over $T$ rounds; since the algorithm is unchanged, existing $O(\sqrt{NT\log N})$ regret bounds still hold and there is no loss in performance due to privacy. We then show that simple modifications -- such as pre-pulling all arms a fixed number of times, increasing the sampling variance -- can provide tighter privacy guarantees. We again provide privacy guarantees that now depend on the new parameters introduced in the modification, which allows the analyst to tune the privacy guarantee as desired. We also provide a novel regret analysis for this new algorithm, and show how the new parameters also impact expected regret. Finally, we empirically validate and illustrate our theoretical findings in two parameter regimes and demonstrate that tuning the new parameters substantially improve the privacy-regret tradeoff.

cs.LG↗

arXiv2024

Centering Policy and Practice: Research Gaps around Usable Differential Privacy

As a mathematically rigorous framework that has amassed a rich theoretical literature, differential privacy is considered by many experts to be the gold standard for privacy-preserving data analysis. Others argue that while differential privacy is a clean formulation in theory, it poses significant challenges in practice. Both perspectives are, in our view, valid and important. To bridge the gaps between differential privacy's promises and its real-world usability, researchers and practitioners must work together to advance policy and practice of this technology. In this paper, we outline pressing open questions towards building usable differential privacy and offer recommendations for the field, such as developing risk frameworks to align with user needs, tailoring communications for different stakeholders, modeling the impact of privacy-loss parameters, investing in effective user interfaces, and facilitating algorithmic and procedural audits of differential privacy systems.

cs.CR↗

arXiv2024

ATTAXONOMY: Unpacking Differential Privacy Guarantees Against Practical Adversaries

Differential Privacy (DP) is a mathematical framework that is increasingly deployed to mitigate privacy risks associated with machine learning and statistical analyses. Despite the growing adoption of DP, its technical privacy parameters do not lend themselves to an intelligible description of the real-world privacy risks associated with that deployment: the guarantee that most naturally follows from the DP definition is protection against membership inference by an adversary who knows all but one data record and has unlimited auxiliary knowledge. In many settings, this adversary is far too strong to inform how to set real-world privacy parameters. One approach for contextualizing privacy parameters is via defining and measuring the success of technical attacks, but doing so requires a systematic categorization of the relevant attack space. In this work, we offer a detailed taxonomy of attacks, showing the various dimensions of attacks and highlighting that many real-world settings have been understudied. Our taxonomy provides a roadmap for analyzing real-world deployments and developing theoretical bounds for more informative privacy attacks. We operationalize our taxonomy by using it to analyze a real-world case study, the Israeli Ministry of Health's recent release of a birth dataset using DP, showing how the taxonomy enables fine-grained threat modeling and provides insight towards making informed privacy parameter choices. Finally, we leverage the taxonomy towards defining a more realistic attack than previously considered in the literature, namely a distributional reconstruction attack: we generalize Balle et al.'s notion of reconstruction robustness to a less-informed adversary with distributional uncertainty, and extend the worst-case guarantees of DP to this average-case setting.

cs.CR↗

arXiv2024

Integrating Differential Privacy and Contextual Integrity

In this work, we propose the first framework for integrating Differential Privacy (DP) and Contextual Integrity (CI). DP is a property of an algorithm that injects statistical noise to obscure information about individuals represented within a database. CI defines privacy as information flow that is appropriate to social context. Analyzed together, these paradigms outline two dimensions on which to analyze privacy of information flows: descriptive and normative properties. We show that our new integrated framework provides benefits to both CI and DP that cannot be attained when each definition is considered in isolation: it enables contextually-guided tuning of the epsilon parameter in DP, and it enables CI to be applied to a broader set of information flows occurring in real-world systems, such as those involving PETs and machine learning. We conclude with a case study based on the use of DP in the U.S. Census Bureau.

cs.CR↗

arXiv2023

Mean Estimation with User-level Privacy under Data Heterogeneity

A key challenge in many modern data analysis tasks is that user data are heterogeneous. Different users may possess vastly different numbers of data points. More importantly, it cannot be assumed that all users sample from the same underlying distribution. This is true, for example in language data, where different speech styles result in data heterogeneity. In this work we propose a simple model of heterogeneous user data that allows user data to differ in both distribution and quantity of data, and provide a method for estimating the population-level mean while preserving user-level differential privacy. We demonstrate asymptotic optimality of our estimator and also prove general lower bounds on the error achievable in the setting we introduce.

cs.CR↗

arXiv2023

Advancing Differential Privacy: Where We Are Now and Future Directions for Real-World Deployment

In this article, we present a detailed review of current practices and state-of-the-art methodologies in the field of differential privacy (DP), with a focus of advancing DP's deployment in real-world applications. Key points and high-level contents of the article were originated from the discussions from "Differential Privacy (DP): Challenges Towards the Next Frontier," a workshop held in July 2022 with experts from industry, academia, and the public sector seeking answers to broad questions pertaining to privacy and its implications in the design of industry-grade systems. This article aims to provide a reference point for the algorithmic and design decisions within the realm of privacy, highlighting important challenges and potential research directions. Covering a wide spectrum of topics, this article delves into the infrastructure needs for designing private systems, methods for achieving better privacy/utility trade-offs, performing privacy attacks and auditing, as well as communicating privacy with broader audiences and stakeholders.

cs.CR↗

arXiv2023

Differentially Private Synthetic Control

Synthetic control is a causal inference tool used to estimate the treatment effects of an intervention by creating synthetic counterfactual data. This approach combines measurements from other similar observations (i.e., donor pool ) to predict a counterfactual time series of interest (i.e., target unit) by analyzing the relationship between the target and the donor pool before the intervention. As synthetic control tools are increasingly applied to sensitive or proprietary data, formal privacy protections are often required. In this work, we provide the first algorithms for differentially private synthetic control with explicit error bounds. Our approach builds upon tools from non-private synthetic control and differentially private empirical risk minimization. We provide upper and lower bounds on the sensitivity of the synthetic control query and provide explicit error bounds on the accuracy of our private synthetic control algorithms. We show that our algorithms produce accurate predictions for the target unit, and that the cost of privacy is small. Finally, we empirically evaluate the performance of our algorithm, and show favorable performance in a variety of parameter regimes, as well as providing guidance to practitioners for hyperparameter tuning.

cs.LG↗