arXiv subjects

T. Tony Cai

Publications and source records attributed to T. Tony Cai.

At least 19 recordsLinked to original sources

arXiv2026

The Cost of Privacy: Rates of Convergence for Parameter Estimation with Differential Privacy

We study the minimax cost of $(\varepsilon,δ)$-differential privacy for mean estimation and Gaussian linear regression in low and high dimensions. For low-dimensional mean estimation, a resampling reduction to fingerprinting yields the privacy contribution $d^2\log(1/δ)/(n^2\varepsilon^2)$ in the stated polynomial-$δ$ regime. For low-dimensional regression, a tracing argument gives the contribution $d^2/(n^2\varepsilon^2)$ under an explicit approximate-DP remainder condition. For sparse mean estimation and sparse regression, a constant-weight packing and a private Fano lemma produce an effective privacy entropy of order $\min\{s\log(ed/s),[\log((e^\varepsilon-1)/δ)]_+\}$ for $δ>0$, up to universal constants and a fixed threshold; for pure DP it is $s\log(ed/s)$. Thus, when $δ$ is polynomially smaller than $\varepsilon$, the pure-DP dependence is retained up to polylogarithmic factors whenever the effective dimension is polylogarithmic in $n$, including regimes with $\varepsilon=o(1)$. Coordinatewise-clipping estimators for means and split-sample noisy-gradient estimators for regression attain the lower bounds up to explicit logarithmic factors. Simulations and data examples illustrate related implementations.

stat.ML↗

arXiv2026

Exact Asymptotic Efficiency under zCDP: Diameter-Constrained Information Geometry

Protecting individual privacy has become a central and urgent concern in modern data analysis, given the vast quantities of data now generated and processed. In this paper, we develop a systematic theory of exact asymptotic efficiency for regular parametric estimation under central zero-concentrated differential privacy. In the privacy regime, the governing object is a diameter-constrained information region: the set of information matrices generated by statistics with diameter at most one. For weighted quadratic loss, we show that the exact local minimax risk is an inverse information variational functional over this region. More generally, a mixed information region yields a unified efficiency constant across different regimes, covering the privacy regime and classical Fisher efficiency. A matching estimator releases the empirical mean of a nearly optimal bounded statistic with Gaussian noise and locally inverts its population moment map. Our theory differs from classical efficiency theory in its set-valued information geometry and loss-dependent efficient estimator. As examples, we provide closed-form constants and optimal procedures for various concrete models, including one-dimensional regular families, Gaussian means, categorical probability vectors, and regression models among others. The theory also transfers to Gaussian differential privacy through an exact parameter rescaling.

math.ST↗

arXiv2026

Score Attack: A Lower Bound Technique for Optimal Differentially Private Learning

Achieving optimal statistical performance while ensuring the privacy of personal data is a challenging yet crucial objective in modern data analysis. However, characterizing the optimality, particularly the minimax lower bound, under privacy constraints is technically difficult. To address this issue, we propose a novel approach called the score attack, which provides a lower bound on the differential-privacy-constrained minimax risk of parameter estimation. The score attack method is based on the tracing attack concept in differential privacy and can be applied to any statistical model with a well-defined score statistic. It can optimally lower bound the minimax risk of estimating unknown model parameters, up to a logarithmic factor, while ensuring differential privacy for a range of statistical problems. We demonstrate the effectiveness and optimality of this general method in various examples, such as the generalized linear model in both classical and high-dimensional sparse settings, the Bradley-Terry-Luce model for pairwise comparisons, and nonparametric regression over the Sobolev class.

math.ST↗

arXiv2026

Optimal Watermark Localization in Mixed-Source Large Language Model Texts

Watermarking provides a principled way to authenticate text generated by large language models (LLMs). In practice, however, the final text may be mixed-source, with watermark evidence surviving at only a subset of token positions after rewriting, insertion, deletion, or paraphrasing. Although prior work has studied global detection of watermark signals, when such signals can be localized remains unclear. We formulate watermark localization as a token-level multiple-testing problem based on pivotal statistics, with a latent indicator recording whether watermark dependence survives at each position. Under an asymptotic regime indexed by exponents for signal sparsity, next-token concentration, and effective-vocabulary growth, we derive a sharp boundary for global detection and phase transitions for discovery and classification within the class of coordinatewise pivot-based localization rules. We show that discovery is strictly harder than detection and that consistent classification is impossible across the parameter regime within this class. We then develop an adaptive thresholding method that does not require knowledge of the exponents or time-varying next-token distributions, but uses a data-driven estimate of the surviving watermark fraction. The method attains the optimal discovery boundary and near-optimal discovery power relative to homogeneous pivot-based rules. Simulations support the theoretical phase transitions, while experiments on model-generated texts demonstrate practical localization performance under common edit mechanisms.

stat.ME↗

arXiv2026

Minimax and Adaptive Covariance Matrix Estimation under Differential Privacy

Estimating covariance matrices is fundamental to a wide range of statistical applications. This paper studies minimax and adaptive estimation of high-dimensional covariance matrices under $ρ$-zero-concentrated differential privacy ($ρ$-zCDP) over three nested classes: the pointwise-decay class $\mathcal{H}_α$, the row-tail class $\mathcal{G}_α$, and the separated-block class $\mathcal{F}_α$. We consider both squared operator norm loss and normalized squared Frobenius norm loss. For $\mathcal{H}_α$ and $\mathcal{G}_α$, we develop center--outer dyadic estimators tailored to the refined geometry of the two classes, while for $\mathcal{F}_α$, we develop a blockwise tridiagonal estimator. The resulting minimax-optimal rates reveal a nontrivial interplay among the smoothness $α$, the loss, the geometry of the covariance class, and the privacy constraint. In contrast to the non-private setting, privacy distinguishes covariance classes that share the same leading non-private rate and induces a polynomial dependence on the ambient dimension. We further develop procedures that adapt to the unknown decay parameter over all three covariance classes under both losses, at the cost of at most polylogarithmic factors. To establish minimax lower bounds, we develop a novel differentially private van Trees inequality that connects Fisher information with the $ρ$-zCDP constraint and may be useful for other private estimation problems. We also construct carefully designed prior distributions to obtain matching minimax lower bounds.

math.ST↗

arXiv2026

The Cost of Discretization in Functional Linear Regression: Minimax Rates and Adaptation

We study scalar-on-function linear regression when each covariate curve is observed only through finitely many noisy point evaluations. Our goal is to characterize the minimax estimation and prediction risks as joint functions of the number of trajectories $n$ and the within-trajectory resolution $m$. Working in a fixed trigonometric eigenbasis, with covariance eigenvalues decaying at rate $α$ and slope function of Sobolev smoothness $s$, we derive matching minimax upper and lower bounds under two canonical sampling schemes. Under an independent random design, the minimax prediction rate is $n^{-\frac{2α+2s}{2α+2s+1}} + (nm)^{-\frac{2α+2s}{4α+2s+1}}$. The first term is the fully observed functional linear regression benchmark, while the second term captures the cost of noisy point evaluations after amplification by the inverse covariance operator. Under a common design on an equally spaced grid, the shared sampling geometry introduces additional obstructions, and the minimax prediction rate becomes $n^{-\frac{2α+2s}{2α+2s+1}} + (nm)^{-\frac{2α+2s}{4α+2s+1}} + m^{-(2α+2s)} + m^{-4α}$. Here the third term represents discretization error induced by the fixed grid, whereas the fourth reflects the cost of identifying unknown eigenvalues from observations on a common grid. We further construct data-driven adaptive estimators that screen the covariance scale and threshold blockwise prediction energy, attaining these rates without prior knowledge of the eigenvalue sequence or the smoothness indices. The results reveal a sharp phase transition that depends on the sampling resolution under independent design and a richer phase diagram under common design. Numerical simulations and a real data example illustrate the theoretical findings.

math.ST↗

arXiv2026

A Van Trees Lower Bound for Fully Interactive Differentially Private Federated Learning

Federated differentially private protocols can communicate over many adaptive rounds and reuse each client's local samples. Existing lower bound arguments for federated DP are often restricted to noninteractive protocols or fresh batch decompositions, so the fundamental information-theoretic limit of estimation under fully interactive protocols remains unknown. We establish a federated van Trees inequality for parameter estimation under squared \ell_2 loss from any complete public transcript satisfying a clientwise zCDP constraint at the sample level. A scalar trace form covers homogeneous experiments, while a matrix form preserves directional Fisher geometry in heterogeneous experiments where different clients are informative in different subspaces. Together with existing upper bounds for the corresponding problems, these results identify the minimax rates for various statistical problems including mean estimation, linear regression, nonparametric regression, and functional mean estimation over the full class of interactive public-transcript protocols. For these problems, arbitrary public interaction and repeated sample reuse do not improve the rate over simpler restricted protocols. The key technical ingredient in our paper is a contraction inequality for the Fisher information in the transcript: each client's contribution is bounded both by the Fisher information in its local experiment and by its total privacy budget.

cs.LG↗

arXiv2025

Nonparametric Bandits with Single-Index Rewards: Optimality and Adaptivity

Contextual bandits are a central framework for sequential decision-making, with applications ranging from recommendation systems to clinical trials. While nonparametric methods can flexibly model complex reward structures, they suffer from the curse of dimensionality. We address this challenge using a single-index model, which projects high-dimensional covariates onto a one-dimensional subspace while preserving nonparametric flexibility. We first develop a nonasymptotic theory for offline single-index regression for each arm, combining maximum rank correlation for index estimation with local polynomial regression. Building on this foundation, we propose a single-index bandit algorithm and establish its convergence rate. We further derive a matching lower bound, showing that the algorithm achieves minimax-optimal regret independent of the ambient dimension $d$, thereby overcoming the curse of dimensionality. We also establish an impossibility result for adaptation: without additional assumptions, no policy can adapt to unknown smoothness levels. Under a standard self-similarity condition, however, we construct a policy that remains minimax-optimal while automatically adapting to the unknown smoothness. Finally, as the dimension $d$ increases, our algorithm continues to achieve minimax-optimal regret, revealing a phase transition that characterizes the fundamental limits of single-index bandit learning.

math.ST↗

arXiv2025

The Cost of Adaptation under Differential Privacy: Optimal Adaptive Federated Density Estimation

Privacy-preserving data analysis has become a central challenge in modern statistics. At the same time, a long-standing goal in statistics is the development of adaptive procedures -- methods that achieve near-optimal performance across diverse function classes without prior knowledge of underlying smoothness or complexity. While adaptation is often achievable at no extra cost in the classical non-private setting, this naturally raises a fundamental question: to what extent is adaptation still possible under privacy constraints? We address this question in the context of density estimation under federated differential privacy (FDP), a framework that encompasses both central and local DP models. We establish sharp results that characterize the cost of adaptation under FDP for both global and pointwise estimation, revealing fundamental differences from the non-private case. We then propose an adaptive FDP estimator that achieves explicit performance guarantees by introducing a new noise mechanism, enabling one-shot adaptation via post-processing. This approach strictly improves upon existing adaptive DP methods. Finally, we develop new lower bound techniques that capture the limits of adaptive inference under privacy and may be of independent interest beyond this problem. Our findings reveal a sharp contrast between private and non-private settings. For global estimation, where adaptation can be achieved for free in the classical non-private setting, we prove that under FDP an intrinsic adaptation cost is unavoidable. For pointwise estimation, where a logarithmic penalty is already known to arise in the non-private setting, we show that FDP introduces an additional logarithmic factor, thereby compounding the cost of adaptation. Taken together, these results provide the first rigorous characterization of the adaptive privacy-accuracy trade-off.

math.ST↗

arXiv2025

Optimal Detection for Language Watermarks with Pseudorandom Collision

Text watermarking plays a crucial role in ensuring the traceability and accountability of large language model (LLM) outputs and mitigating misuse. While promising, most existing methods assume perfect pseudorandomness. In practice, repetition in generated text induces collisions that create structured dependence, compromising Type I error control and invalidating standard analyses. We introduce a statistical framework that captures this structure through a hierarchical two-layer partition. At its core is the concept of minimal units -- the smallest groups treatable as independent across units while permitting dependence within. Using minimal units, we define a non-asymptotic efficiency measure and cast watermark detection as a minimax hypothesis testing problem. Applied to Gumbel-max and inverse-transform watermarks, our framework produces closed-form optimal rules. It explains why discarding repeated statistics often improves performance and shows that within-unit dependence must be addressed unless degenerate. Both theory and experiments confirm improved detection power with rigorous Type I error control. These results provide the first principled foundation for watermark detection under imperfect pseudorandomness, offering both theoretical insight and practical guidance for reliable tracing of model outputs.

math.ST↗

arXiv2025

Integrated Analysis for Electronic Health Records with Structured and Sporadic Missingness

Objectives: We propose a novel imputation method tailored for Electronic Health Records (EHRs) with structured and sporadic missingness. Such missingness frequently arises in the integration of heterogeneous EHR datasets for downstream clinical applications. By addressing these gaps, our method provides a practical solution for integrated analysis, enhancing data utility and advancing the understanding of population health. Materials and Methods: We begin by demonstrating structured and sporadic missing mechanisms in the integrated analysis of EHR data. Following this, we introduce a novel imputation framework, Macomss, specifically designed to handle structurally and heterogeneously occurring missing data. We establish theoretical guarantees for Macomss, ensuring its robustness in preserving the integrity and reliability of integrated analyses. To assess its empirical performance, we conduct extensive simulation studies that replicate the complex missingness patterns observed in real-world EHR systems, complemented by validation using EHR datasets from the Duke University Health System (DUHS). Results: Simulation studies show that our approach consistently outperforms existing imputation methods. Using datasets from three hospitals within DUHS, Macomss achieves the lowest imputation errors for missing data in most cases and provides superior or comparable downstream prediction performance compared to benchmark methods. Conclusions: We provide a theoretically guaranteed and practically meaningful method for imputing structured and sporadic missing data, enabling accurate and reliable integrated analysis across multiple EHR datasets. The proposed approach holds significant potential for advancing research in population health.

stat.AP↗

arXiv2025

Optimal Differentially Private Ranking from Pairwise Comparisons

Data privacy is a central concern in many applications involving ranking from incomplete and noisy pairwise comparisons, such as recommendation systems, educational assessments, and opinion surveys on sensitive topics. In this work, we propose differentially private algorithms for ranking based on pairwise comparisons. Specifically, we develop and analyze ranking methods under two privacy notions: edge differential privacy, which protects the confidentiality of individual comparison outcomes, and individual differential privacy, which safeguards potentially many comparisons contributed by a single individual. Our algorithms--including a perturbed maximum likelihood estimator and a noisy count-based method--are shown to achieve minimax optimal rates of convergence under the respective privacy constraints. We further demonstrate the practical effectiveness of our methods through experiments on both simulated and real-world data.

math.ST↗

arXiv2025

A Locally Adaptive Algorithm for Multiple Testing with Network Structure

Incorporating auxiliary information alongside primary data can significantly enhance the accuracy of simultaneous inference. However, existing multiple testing methods face challenges in efficiently incorporating complex side information, especially when it differs in dimension or structure from the primary data, such as network side information. This paper introduces a locally adaptive structure learning algorithm (LASLA), a flexible framework designed to integrate a broad range of auxiliary information into the inference process. Although LASLA is specifically motivated by the challenges posed by network-structured data, it also proves highly effective with other types of side information, such as spatial locations and multiple auxiliary sequences. LASLA employs a $p$-value weighting approach, leveraging structural insights to derive data-driven weights that prioritize the importance of different hypotheses. Our theoretical analysis demonstrates that LASLA asymptotically controls the false discovery rate (FDR) under independent or weakly dependent $p$-values, and achieves enhanced power in scenarios where the auxiliary data provides valuable side information. Simulation studies are conducted to evaluate LASLA's numerical performance, and its efficacy is further illustrated through two real-world applications.

stat.ME↗

arXiv2024

Federated PCA and Estimation for Spiked Covariance Matrices: Optimal Rates and Efficient Algorithm

Federated Learning (FL) has gained significant recent attention in machine learning for its enhanced privacy and data security, making it indispensable in fields such as healthcare, finance, and personalized services. This paper investigates federated PCA and estimation for spiked covariance matrices under distributed differential privacy constraints. We establish minimax rates of convergence, with a key finding that the central server's optimal rate is the harmonic mean of the local clients' minimax rates. This guarantees consistent estimation at the central server as long as at least one local client provides consistent results. Notably, consistency is maintained even if some local estimators are inconsistent, provided there are enough clients. These findings highlight the robustness and scalability of FL for reliable statistical inference under privacy constraints. To establish minimax lower bounds, we derive a matrix version of van Trees' inequality, which is of independent interest. Furthermore, we propose an efficient algorithm that preserves differential privacy while achieving near-optimal rates at the central server, up to a logarithmic factor. We address significant technical challenges in analyzing this algorithm, which involves a three-layer spectral decomposition. Numerical performance of the proposed algorithm is investigated using both simulated and real data.

math.ST↗

arXiv2024

Optimal Differentially Private PCA and Estimation for Spiked Covariance Matrices

Estimating a covariance matrix and its associated principal components is a fundamental problem in contemporary statistics. While optimal estimation procedures have been developed with well-understood properties, the increasing demand for privacy preservation introduces new complexities to this classical problem. In this paper, we study optimal differentially private Principal Component Analysis (PCA) and covariance estimation within the spiked covariance model. We precisely characterize the sensitivity of eigenvalues and eigenvectors under this model and establish the minimax rates of convergence for estimating both the principal components and covariance matrix. These rates hold up to logarithmic factors and encompass general Schatten norms, including spectral norm, Frobenius norm, and nuclear norm as special cases. We propose computationally efficient differentially private estimators and prove their minimax optimality for sub-Gaussian distributions, up to logarithmic factors. Additionally, matching minimax lower bounds are established. Notably, compared to the existing literature, our results accommodate a diverging rank, a broader range of signal strengths, and remain valid even when the sample size is much smaller than the dimension, provided the signal strength is sufficiently strong. Both simulation studies and real data experiments demonstrate the merits of our method.

math.ST↗

arXiv2024

Minimax And Adaptive Transfer Learning for Nonparametric Classification under Distributed Differential Privacy Constraints

This paper considers minimax and adaptive transfer learning for nonparametric classification under the posterior drift model with distributed differential privacy constraints. Our study is conducted within a heterogeneous framework, encompassing diverse sample sizes, varying privacy parameters, and data heterogeneity across different servers. We first establish the minimax misclassification rate, precisely characterizing the effects of privacy constraints, source samples, and target samples on classification accuracy. The results reveal interesting phase transition phenomena and highlight the intricate trade-offs between preserving privacy and achieving classification accuracy. We then develop a data-driven adaptive classifier that achieves the optimal rate within a logarithmic factor across a large collection of parameter spaces while satisfying the same set of differential privacy constraints. Simulation studies and real-world data applications further elucidate the theoretical analysis with numerical results.

math.ST↗

arXiv2024

Federated Nonparametric Hypothesis Testing with Differential Privacy Constraints: Optimal Rates and Adaptive Tests

Federated learning has attracted significant recent attention due to its applicability across a wide range of settings where data is collected and analyzed across disparate locations. In this paper, we study federated nonparametric goodness-of-fit testing in the white-noise-with-drift model under distributed differential privacy (DP) constraints. We first establish matching lower and upper bounds, up to a logarithmic factor, on the minimax separation rate. This optimal rate serves as a benchmark for the difficulty of the testing problem, factoring in model characteristics such as the number of observations, noise level, and regularity of the signal class, along with the strictness of the $(ε,δ)$-DP requirement. The results demonstrate interesting and novel phase transition phenomena. Furthermore, the results reveal an interesting phenomenon that distributed one-shot protocols with access to shared randomness outperform those without access to shared randomness. We also construct a data-driven testing procedure that possesses the ability to adapt to an unknown regularity parameter over a large collection of function classes with minimal additional cost, all while maintaining adherence to the same set of DP constraints.

math.ST↗

arXiv2024

Optimal Federated Learning for Nonparametric Regression with Heterogeneous Distributed Differential Privacy Constraints

This paper studies federated learning for nonparametric regression in the context of distributed samples across different servers, each adhering to distinct differential privacy constraints. The setting we consider is heterogeneous, encompassing both varying sample sizes and differential privacy constraints across servers. Within this framework, both global and pointwise estimation are considered, and optimal rates of convergence over the Besov spaces are established. Distributed privacy-preserving estimators are proposed and their risk properties are investigated. Matching minimax lower bounds, up to a logarithmic factor, are established for both global and pointwise estimation. Together, these findings shed light on the tradeoff between statistical accuracy and privacy preservation. In particular, we characterize the compromise not only in terms of the privacy budget but also concerning the loss incurred by distributing data within the privacy framework as a whole. This insight captures the folklore wisdom that it is easier to retain privacy in larger samples, and explores the differences between pointwise and global estimation under distributed privacy constraints.

math.ST↗