In this paper we revisit the DP stochastic convex optimization (SCO) problem. For convex smooth losses, it is well-known that the canonical DP-SGD (stochastic gradient descent) achieves the optimal rate of $O\left(\frac{LR}{\sqrt{n}} + \frac{LR \sqrt{p \log(1/\delta)}}{\epsilon n}\right)$ under $(\epsilon, \delta)$-DP (Bassily et al. 2014), and also well-known that variants of DP-SGD can achieve the optimal rate in a single epoch (Feldman et al. 2019). However, the batch gradient complexity (i.e., number of adaptive optimization steps), which is important in applications like federated learning, is less well-understood. In particular, all prior work on DP-SCO requires $\Omega(n)$ batch gradient steps, multiple epochs, or convexity for privacy. We propose an algorithm, Accelerated-DP-SRGD (stochastic recursive gradient descent), which bypasses the limitations of past work: it achieves the optimal rate for DP-SCO (up to polylog factors), in a single epoch using $\sqrt{n}$ batch gradient steps with batch size $\sqrt{n}$, and can be made private for arbitrary (non-convex) losses via clipping. If the global minimizer is in the constraint set, we can further improve this to $n^{1/4}$ batch gradient steps with batch size $n^{3/4}$. To achieve this, our algorithm combines three key ingredients, a variant of stochastic recursive gradients (SRG), accelerated gradient descent, and correlated noise generation from DP continual counting.

We study the problem of computing the privacy parameters for DP machine learning when using privacy amplification via random batching and noise correlated across rounds via a correlation matrix $\textbf{C}$ (i.e., the matrix mechanism). Past work on this problem either only applied to banded $\textbf{C}$, or gave loose privacy parameters. In this work, we give a framework for computing near-exact privacy parameters for any lower-triangular, non-negative $\textbf{C}$. Our framework allows us to optimize the correlation matrix $\textbf{C}$ while accounting for amplification, whereas past work could not. Empirically, we show this lets us achieve smaller RMSE on prefix sums than the previous state-of-the-art (SOTA). We also show that we can improve on the SOTA performance on deep learning tasks. Our two main technical tools are (i) using Monte Carlo accounting to bypass composition, which was the main technical challenge for past work, and (ii) a ``balls-in-bins'' batching scheme that enables easy privacy analysis and is closer to practical random batching than Poisson sampling.

We propose a simple heuristic privacy analysis of noisy clipped stochastic gradient descent (DP-SGD) in the setting where only the last iterate is released and the intermediate iterates remain hidden. Namely, our heuristic assumes a linear structure for the model. We show experimentally that our heuristic is predictive of the outcome of privacy auditing applied to various training procedures. Thus it can be used prior to training as a rough estimate of the final privacy leakage. We also probe the limitations of our heuristic by providing some artificial counterexamples where it underestimates the privacy leakage. The standard composition-based privacy analysis of DP-SGD effectively assumes that the adversary has access to all intermediate iterates, which is often unrealistic. However, this analysis remains the state of the art in practice. While our heuristic does not replace a rigorous privacy analysis, it illustrates the large gap between the best theoretical upper bounds and the privacy auditing lower bounds and sets a target for further work to improve the theoretical privacy analyses.

As major progress in LLM-based long-form text generation enables paradigms such as retrieval-augmented generation (RAG) and inference-time scaling, safely incorporating private information into the generation remains a critical open question. We present InvisibleInk, a highly scalable long-form text generation framework satisfying rigorous differential privacy guarantees with respect to the sensitive reference texts. It interprets sampling from the LLM's next-token-distribution as the exponential mechanism over the LLM logits with two innovations. First, we reduce the privacy cost by isolating and clipping only the sensitive information in the model logits (relative to the public logits). Second, we improve text quality by sampling without any privacy cost from a small superset of the top-$k$ private tokens. Empirical evaluations demonstrate a consistent $8\times$ (or more) reduction in computation cost over state-of-the-art baselines to generate long-form private text of the same utility across privacy levels. InvisibleInk is able to generate, for the first time, high-quality private long-form text at less than $4\text{-}8\times$ times the computation cost of non-private generation, paving the way for its practical use. We open-source a pip-installable Python package (invink) for InvisibleInk at https://github.com/cerai-iitm/invisibleink.

DP training pipelines for modern neural networks are iterative and generate multiple checkpoints. However, all except the final checkpoint are discarded after training. In this work, we propose novel methods to utilize intermediate checkpoints to improve prediction accuracy and estimate uncertainty in DP predictions. First, we design a general framework that uses aggregates of intermediate checkpoints during training to increase the accuracy of DP ML techniques. Specifically, we demonstrate that training over aggregates can provide significant gains in prediction accuracy over the existing state-of-the-art for StackOverflow, CIFAR10 and CIFAR100 datasets. For instance, we improve the state-of-the-art DP StackOverflow accuracies to 22.74% (+2.06% relative) for epsilon=8.2, and 23.90% (+2.09%) for epsilon=18.9. Furthermore, these gains magnify in settings with periodically varying training data distributions. We also demonstrate that our methods achieve relative improvements of 0.54% and 62.6% in terms of utility and variance, on a proprietary, production-grade pCVR task. Lastly, we initiate an exploration into estimating the uncertainty (variance) that DP noise adds in the predictions of DP ML models. We prove that, under standard assumptions on the loss function, the sample variance from last few checkpoints provides a good approximation of the variance of the final model of a DP run. Empirically, we show that the last few checkpoints can provide a reasonable lower bound for the variance of a converged DP model. Crucially, all the methods proposed in this paper operate on a single training run of the DP ML technique, thus incurring no additional privacy cost.

We study a class of private learning problems in which the data is a join of private and public features. This is often the case in private personalization tasks such as recommendation or ad prediction, in which features related to individuals are sensitive, while features related to items (the movies or songs to be recommended, or the ads to be shown to users) are publicly available and do not require protection. A natural question is whether private algorithms can achieve higher utility in the presence of public features. We give a positive answer for multi-encoder models where one of the encoders operates on public features. We develop new algorithms that take advantage of this separation by only protecting certain sufficient statistics (instead of adding noise to the gradient). This method has a guaranteed utility improvement for linear regression, and importantly, achieves the state of the art on two standard private recommendation benchmarks, demonstrating the importance of methods that adapt to the private-public feature separation.

Personalization of machine learning (ML) predictions for individual users/domains/enterprises is critical for practical recommendation systems. Standard personalization approaches involve learning a user/domain specific \emph{embedding} that is fed into a fixed global model which can be limiting. On the other hand, personalizing/fine-tuning model itself for each user/domain — a.k.a meta-learning — has high storage/infrastructure cost. Moreover, rigorous theoretical studies of scalable personalization approaches have been very limited. To address the above issues, we propose a novel meta-learning style approach that models network weights as a sum of low-rank and sparse components. This captures common information from multiple individuals/users together in the low-rank part while sparse part captures user-specific idiosyncrasies. We then study the framework in the linear setting, where the problem reduces to that of estimating the sum of a rank-$r$ and a $k$-column sparse matrix using a small number of linear measurements. We propose a computationally efficient alternating minimization method with iterative hard thresholding — AMHT-LRS — to learn the low-rank and sparse part. Theoretically, for the realizable Gaussian data setting, we show that AMHT-LRS solves the problem efficiently with nearly optimal sample complexity. Finally, a significant challenge in personalization is ensuring privacy of each user’s sensitive data. We alleviate this problem by proposing a differentially private variant of our method that also is equipped with strong generalization guarantees.

In the task of differentially private (DP) continual counting, we receive a stream of increments and our goal is to output an approximate running total of these increments, without revealing too much about any specific increment. Despite its simplicity, differentially private continual counting has attracted significant attention both in theory and in practice. Existing algorithms for differentially private continual counting are either inefficient in terms of their space usage or add an excessive amount of noise, inducing suboptimal utility. The most practical DP continual counting algorithms add carefully correlated Gaussian noise to the values. The task of choosing the covariance for this noise can be expressed in terms of factoring the lower-triangular matrix of ones (which computes prefix sums). We present two approaches from this class (for different parameter regimes) that achieve near-optimal utility for DP continual counting and only require logarithmic or polylogarithmic space (and time). Our first approach is based on a space-efficient streaming matrix multiplication algorithm for a class of Toeplitz matrices. We show that to instantiate this algorithm for DP continual counting, it is sufficient to find a low-degree rational function that approximates the square root on a circle in the complex plane. We then apply and extend tools from approximation theory to achieve this. We also derive efficient closed-forms for the objective function for arbitrarily many steps, and show direct numerical optimization yields a highly practical solution to the problem. Our second approach combines our first approach with a recursive construction similar to the binary tree mechanism.

Differentially private learning algorithms inject noise into the learning process. While the most common private learning algorithm, DP-SGD, adds independent Gaussian noise in each iteration, recent work on matrix factorization mechanisms has shown empirically that introducing correlations in the noise can greatly improve their utility. We characterize the asymptotic learning utility for any choice of the correlation function, giving precise analytical bounds for linear regression and as the solution to a convex program for general convex functions. We show, using these bounds, how correlated noise provably improves upon vanilla DP-SGD as a function of problem parameters such as the effective dimension and condition number. Moreover, our analytical expression for the near-optimal correlation function circumvents the cubic complexity of the semi-definite program used to optimize the noise correlation matrix in previous work. We validate these theoretical results with experiments on private deep learning. Our work matches or outperforms prior work while being efficient both in terms of computation and memory.

Privacy amplification exploits randomness in data selection to provide tighter differential privacy (DP) guarantees. This analysis is key to DP-SGD's success in machine learning (ML), but, is not readily applicable to the newer state-of-the-art (SOTA) algorithms. This is because these algorithms, known as DP-FTRL, use the matrix mechanism to add correlated noise instead of independent noise as in DP-SGD. In this paper, we propose "MMCC'' (matrix mechanism conditional composition), the first algorithm to analyze privacy amplification via sampling for any generic matrix mechanism. MMCC is nearly tight in that it approaches a lower bound as $\epsilon\to0$. To analyze correlated outputs in MMCC, we prove that they can be analyzed as if they were independent, by conditioning them on prior outputs. Our "conditional composition theorem'' has broad utility: we use it to show that the noise added to binary-tree-DP-FTRL can asymptotically match the noise added to DP-SGD with amplification. Our algorithm also has practical empirical utility. We show that amplification leads to significant improvement in the privacy/utility trade-offs for DP-FTRL style algorithms for standard benchmark tasks.

We provide an improved analysis of standard differentially private gradient descent for linear regression under the squared error loss. Under modest assumptions on the input, we characterize the distribution of the iterate at each time step. Our analysis leads to new results on the algorithm’s accuracy: for a proper fixed choice of hyperparameters, the sample complexity depends only linearly on the dimension of the data. This matches the dimension-dependence of the (non-private) ordinary least squares estimator as well as that of recent private algorithms that rely on sophisticated adaptive gradient-clipping schemes (Varshney et al., 2022; Liu et al., 2023). Our analysis of the iterates’ distribution also allows us to construct confidence intervals for the empirical optimizer which adapt automatically to the variance of the algorithm on a particular data set. We validate our theorems through experiments on synthetic data.

We design differentially private regret-minimizing algorithms in the online convex optimization (OCO) framework. Unlike recent results, our algorithms and analyses do not require smoothness, thus yielding the first private regret bounds with an optimal leading-order term for non-smooth loss functions. Additionally, even for smooth losses, the resulting regret guarantees improve upon previous results in terms their dependence of dimension. Our results provide the best known rates for DP-OCO in all practical regimes of the privacy parameter, barring when it is exceptionally small. The principal innovation in our algorithm design is the use of sampling from strongly log-concave densities which satisfy the Log-Sobolev Inequality. The resulting concentration of measure allows us to obtain a better trade-off for the dimension factors than prior work, leading to improved results. Following previous works on DP-OCO, the proposed algorithm explicitly limits the number of switches via rejection sampling. Thus, independently of privacy constraints, the algorithm also provides improved results for online convex optimization with a switching budget.

Self-supervised learning (SSL) methods for large speech models have proven to be highly effective at ASR. With the interest in public deployment of large pre-trained models, there is a rising concern for unintended memorization and leakage of sensitive data points from the training data. In this paper, we apply differentially private (DP) pre-training to a SOTA Conformer-based encoder, and study its performance on a downstream ASR task assuming the fine-tuning data is public. This paper is the first to apply DP to SSL for ASR, investigating the DP noise tolerance of the BEST-RQ pre-training method. Notably, we introduce a novel variant of model pruning called gradient-based layer freezing that provides strong improvements in privacy-utility-compute trade-offs. Our approach yields a LibriSpeech test-clean/other WER (%) of 3.78/ 8.41 with ($10,1 \mathrm{e}-9$)-DP for extrapolation towards low dataset scales, and 2.81/5.89 with ($10,7.9 \mathrm{e}-11$)DP for extrapolation towards high scales.

We design, to the best of our knowledge, the first differentially private (DP) stream aggregation processing system at scale. Our system - Differential Privacy SQL Pipelines (DP-SQLP) - is built using a streaming framework similar to Spark streaming, and is built on top of the Spanner database and the F1 query engine from Google. Towards designing DP-SQLP we make both algorithmic and systemic advances, namely, we (i) design a novel (user-level) DP key selection algorithm that can operate on an unbounded set of possible keys, and can scale to one billion keys that users have contributed, (ii) design a preemptive execution scheme for DP key selection that avoids enumerating all the keys at each triggering time, and (iii) use algorithmic techniques from DP continual observation to release a continual DP histogram of user contributions to different keys over the stream length. We empirically demonstrate the efficacy by obtaining at least 16× reduction in error over meaningful baselines we consider. We implemented a streaming differentially private user impressions for Google Shopping with DP-SQLP. The streaming DP algorithms are further applied to Google Trends.

We study the task of differentially private online convex optimization (OCO). In the online setting, the release of each distinct decision or iterate carries with it the potential for privacy loss. To limit such privacy leakage, we design an optimization-based OCO algorithm that explicitly limits the number of switches via objective perturbation and rejection sampling. This improves over known results in multiple aspects: an optimal leading-order regret term, in being efficiently implementable without requiring log-concave sampling subroutines, and in matching the non-private regret bound for sub-constant regimes of privacy parameters. Leveraging the fact that the algorithm is designed to explicitly minimize the number of switches of decisions, we show that the algorithm also obtains optimal regret bounds in the lazy OCO setting, where the learner is constrained to perform a limited number of switches. In addition, for one- and two-dimensional decision sets, we present a novel approach for differentially private online Lipschitz learning, where the loss functions are Lipschitz but not necessarily convex, that achieves the optimal regret bound matching known lower bounds.

In this paper we provide an algorithmic framework based on Langevin diffusion (LD) and its corresponding discretizations that allow us to simultaneously obtain: i) An algorithm for sampling from the exponential mechanism, whose privacy analysis does not depend on convexity, and which can be stopped at anytime without compromising privacy, and ii) tight uniform stability guarantees for the exponential mechanism. As a direct consequence, we obtain optimal excess empirical and population risk guarantees for (strongly) convex losses under both pure and approximate differential privacy (DP). The framework allows us to design a DP uniform sampler from a Rashomon set. Rashomon sets are widely used in interpretable and robust machine learning, understanding variable importance, and characterizing fairness.

We introduce new differentially private (DP) mechanisms for gradient-based machine learning (ML) with multiple passes (epochs) over a dataset, substantially improving the achievable privacy-utility-computation tradeoffs. We formalize the problem of DP mechanisms for adaptive streams with multiple participations and introduce a non-trivial extension of online matrix factorization DP mechanisms to our setting. This includes establishing the necessary theory for sensitivity calculations and efficient computation of optimal matrices. For some applications like $>\!\! 10,000$ SGD steps, applying these optimal techniques becomes computationally expensive. We thus design an efficient Fourier-transform-based mechanism with only a minor utility loss. Extensive empirical evaluation on both example-level DP for image classification and user-level DP for language modeling demonstrate substantial improvements over all previous methods, including the widely-used DP-SGD. Though our primary application is to ML, our main DP results are applicable to arbitrary linear queries and hence may have much broader applicability.

In the privacy-utility tradeoff of a model trained on benchmark language and vision tasks, remarkable improvements have been widely reported when the model is pretrained on public data. Some gain is expected as these models inherit the benefits of transfer learning, which is the standard motivation in non-private settings. However, the stark contrast in the gain of pretraining between non-private and private machine learning suggests that the gain in the latter is rooted in a fundamentally different cause. To explain this phenomenon, we hypothesize that the non-convex loss landscape of a model training necessitates the optimization algorithm to go through two phases. In the first, the algorithm needs to select a good “basin” in the loss landscape. In the second, the algorithm solves an easy optimization within that basin. The former is a harder problem to solve with private data, while the latter is harder to solve with public data due to a distribution shift or data scarcity. Guided by this intuition, we provide theoretical constructions that provably demonstrate the separation between private training with and without public pretraining. Further, systematic experiments on CIFAR10 and Librispeech provide supporting evidence for our hypothesis.

We study the problem of multi-task learning under user-level differential privacy, in which n users contribute data to m tasks, each involving a subset of users. One important aspect of the problem, that can significantly impact quality, is the distribution skew among tasks. Tasks that have much fewer data samples than others are more susceptible to the noise added for privacy. It is natural to ask whether algorithms can adapt to this skew to improve the overall utility. We give a systematic analysis of the problem, by studying how to optimally allocate a user’s privacy budget among tasks. We propose a generic algorithm, based on an adaptive reweighting of the empirical loss, and show that in the presence of distribution skew, this gives a quantifiable improvement of excess empirical risk. Experimental studies on recommendation problems that exhibit a long tail of small tasks, demonstrate that our methods significantly improve utility, achieving the state of the art on two standard benchmarks.

Matrix factorization (MF) mechanisms for differential privacy (DP) have substantially improved the state-of-the-art in privacy-utility-computation tradeoffs for ML applications in a variety of scenarios, but in both the centralized and federated settings there remain instances where either MF cannot be easily applied, or other algorithms provide better tradeoffs (typically, as $\epsilon$ becomes small). In this work, we show how MF can subsume prior state-of-the-art algorithms in both federated and centralized training settings, across all privacy budgets. The key technique throughout is the construction of MF mechanisms with banded matrices (lower-triangular matrices with at most $\hat{b}$ nonzero bands including the main diagonal). For cross-device federated learning (FL), this enables multiple-participations with a relaxed device participation schema compatible with practical FL infrastructure (as demonstrated by a production deployment). In the centralized setting, we prove that banded matrices enjoy the same privacy amplification results as the ubiquitous DP-SGD algorithm, but can provide strictly better performance in most scenarios---this lets us always at least match DP-SGD, and often outperform it

Differentially private (stochastic) gradient descent is the workhorse of DP private machine learning in both the convex and non-convex settings. Without privacy constraints, second-order methods, like Newton's method, converge faster than first-order methods like gradient descent. In this work, we investigate the prospect of using the second-order information from the loss function to accelerate DP convex optimization. We first develop a private variant of the regularized cubic Newton method of Nesterov and Polyak, and show that for the class of strongly convex loss functions, our algorithm has quadratic convergence and achieves the optimal excess loss. We then design a practical second-order DP algorithm for the unconstrained logistic regression problem. We theoretically and empirically study the performance of our algorithm. Empirical results show our algorithm consistently achieves the best excess loss compared to other baselines and is 10-40x faster than DP-GD/DP-SGD for challenging datasets.

We reconsider the challenge of non-convex optimization under differential privacy constraint. Building upon the previous variance-reduced algorithm SpiderBoost, we propose a novel framework that employs two types of gradient oracles: one that estimates the gradient at a single point and a more cost-effective option that calculates the gradient difference between two points. Our framework can ensure continuous accuracy of gradient estimations and subsequently enhances the rates of identifying second-order stationary points. Additionally, we consider a more challenging task by attempting to locate the global minima of a non-convex objective via the exponential mechanism without almost any assumptions. Our preliminary results suggest that the regularized exponential mechanism can effectively emulate previous empirical and population risk bounds, negating the need for smoothness assumptions for algorithms with polynomial running time. Furthermore, with running time factors excluded, the exponential mechanism demonstrates promising population risk bound performance, and we provide a nearly matching lower bound.

Training reliable deep learning models which avoid making overconfident but incorrect predictions is a longstanding challenge. This challenge is further exacerbated when learning has to be differentially private: protection provided to sensitive data comes at the price of injecting additional randomness into the learning process. In this work, we conduct a thorough empirical investigation of selective classifiers---that can abstain under uncertainty---under a differential privacy constraint. We find that some popular selective prediction approaches are ineffective in a differentially private setting because they increase the risk of privacy leakage. At the same time, we identify that a recent approach that only uses checkpoints produced by an off-the-shelf private learning algorithm stands out as particularly suitable under DP. Further, we show that differential privacy does not just harm utility but also degrades selective classification performance. To analyze this effect across privacy levels, we propose a novel evaluation mechanism which isolates selective prediction performance across model utility levels at full coverage. Our experimental results show that recovering the performance level attainable by non-private models is possible but comes at a considerable coverage cost as the privacy budget decreases.

Machine Learning (ML) models are ubiquitous in real-world applications and are a constant focus of research. Modern ML models have become more complex, deeper, and harder to reason about. At the same time, the community has started to realize the importance of protecting the privacy of the training data that goes into these models. Differential Privacy (DP) has become a gold standard for making formal statements about data anonymization. However, while some adoption of DP has happened in industry, attempts to apply DP to real world complex ML models are still few and far between. The adoption of DP is hindered by limited practical guidance of what DP protection entails, what privacy guarantees to aim for, and the difficulty of achieving good privacy-utility-computation trade-offs for ML models. Tricks for tuning and maximizing performance are scattered among papers or stored in the heads of practitioners, particularly with respect to the challenging task of hyperparameter tuning. Furthermore, the literature seems to present conflicting evidence on how and whether to apply architectural adjustments and which components are “safe” to use with DP. In this survey paper, we attempt to create a self-contained guide that gives an in-depth overview of the field of DP ML. We aim to assemble information about achieving the best possible DP ML model with rigorous privacy guarantees. Our target audience is both researchers and practitioners. Researchers interested in DP for ML will benefit from a clear overview of current advances and areas for improvement. We also include theory-focused sections that highlight important topics such as privacy accounting and convergence. For a practitioner, this survey provides a background in DP theory and a clear step-by-step guide for choosing an appropriate privacy definition and approach, implementing DP training, potentially updating the model architecture, and tuning hyperparameters. For both researchers and practitioners, consistently and fully reporting privacy guarantees is critical, so we propose a set of specific best practices for stating guarantees. With sufficient computation and a sufficiently large training set or supplemental nonprivate data, both good accuracy (that is, almost as good as a non-private model) and good privacy can often be achievable. And even when computation and dataset size are limited, there are advantages to training with even a weak (but still finite) formal DP guarantee. Hence, we hope this work will facilitate more widespread deployments of DP ML models.

Consider the following optimization problem: Given $n \times n$ matrices $A$ and $\Lambda$, maximize $⟨A, U\Lambda U^*⟩$ where $U$ varies over the unitary group $\mathrm{U}(n)$. This problem seeks to approximate $A$ by a matrix whose spectrum is the same as $\Lambda$ and, by setting $\Lambda$ to be appropriate diagonal matrices, one can recover matrix approximation problems such as PCA and rank-$k$ approximation. We study the problem of designing differentially private algorithms for this optimization problem in settings where the matrix $A$ is constructed using users’ private data. We give efficient and private algorithms that come with upper and lower bounds on the approximation error. Our results unify and improve upon several prior works on private matrix approximation problems. They rely on extensions of packing/covering number bounds for Grassmannians to unitary orbits which should be of independent interest.

We study the problem of differentially private linear regression where each of the data point is sampled from a fixed sub-Gaussian style distribution. We propose and analyze a one-pass mini-batch stochastic gradient descent method (DP-AMBSSGD) where points in each iteration are sampled without replacement. Noise is added for DP but the noise standard deviation is estimated online. Compared to existing $(\epsilon, \delta)$-DP techniques which have sub-optimal error bounds, DP-AMBSSGD is able to provide nearly optimal error bounds in terms of key parameters like dimensionality $d$, number of points $N$, and the standard deviation \sigma of the noise in observations. For example, when the $d$-dimensional covariates are sampled i.i.d. from the normal distribution, then the excess error of DP-AMBSSGD due to privacy is $\sigma^2 d/N(1+d/(\epsilon^2 N))$, i.e., the error is meaningful when number of samples N\geq d \log d which is the standard operative regime for linear regression. In contrast, error bounds for existing efficient methods in this setting are: $d^3/(\epsilon^2 N^2)$, even for $\sigma=0$. That is, for constant $\epsilon$, the existing techniques require $N=d^{1.5}$ to provide a non-trivial result.

In this paper, we revisit the problem of using in-distribution public data to improve the privacy/utility trade-offs for differentially private (DP) model training. (Here, public data refers to auxiliary data sets that have no privacy concerns.) We design a natural variant of DP mirror descent, where the DP gradients of the private/sensitive data act as the linear term, and the loss generated by the public data as the mirror map. We show that, for linear regression with feature vectors drawn from a non-isotropic sub-Gaussian distribution, our algorithm, PDA-DPMD (a variant of mirror descent), provides population risk guarantees that are asymptotically better than the best known guarantees under DP (without having access to public data), when the number of public data samples is sufficiently large. We further show that our algorithm has natural “noise stability” properties that control the variance due to noise added to ensure DP. We demonstrate the efficacy of our algorithm by showing privacy/utility trade-offs on four benchmark datasets (StackOverflow, WikiText-2, CIFAR-10, and EMNIST). We show that our algorithm not only significantly improves over traditional DP-SGD, which does not have access to public data, but to our knowledge is the first to improve over DP-SGD on models that have been pre-trained with public data.

Motivated by recent applications requiring differential privacy in the setting of adaptive streams, we investigate the question of optimal instantiations of the matrix mechanism in this setting. We prove fundamental theoretical results on the applicability of matrix factorizations to the adaptive streaming setting, and provide a new parameter-free fixed-point algorithm for computing optimal factorizations. We instantiate this framework with respect to concrete matrices which arise naturally in the machine learning setting, and train user-level differentially private models with the resulting optimal mechanisms, yielding significant improvements on a notable problem in federated learning with user-level differential privacy.

Large pretrained models can be fine-tuned with differential privacy to achieve performance approaching that of non-private models. A common theme in these results is the surprising observation that high-dimensional models can achieve favorable privacy-utility trade-offs. This seemingly contradicts known results on the model-size dependence of differentially private convex learning and raises the following research question: When does the performance of differentially private learning not degrade with increasing model size? We identify that the magnitudes of gradients projected onto subspaces is a key factor that determines performance. To precisely characterize this for private convex learning, we introduce a condition on the objective that we term restricted Lipschitz continuity and derive improved bounds for the excess empirical and population risks that are dimension- independent under additional conditions. We empirically show that in private fine-tuning of large language models, gradients obtained during fine-tuning are mostly controlled by a few principal components. This behavior is similar to conditions under which we obtain dimension-independent bounds in convex settings. Our theoretical and empirical results together provide a possible explanation for the recent success of large-scale private fine-tuning. Code to reproduce our results can be found at https://github.com/lxuechen/private-transformers/tree/main/examples/classification/spectral_analysis.

Because learning sometimes involves sensitive data, machine learning algorithms have been extended to offer differential privacy for training data. In practice, this has been mostly an afterthought, with privacy-preserving models obtained by re-running training with a different optimizer, but using the model architectures that already performed well in a non-privacy-preserving setting. This approach leads to less than ideal privacy/utility tradeoffs, as we show here. To improve these tradeoffs, prior work introduces variants of differential privacy that weaken the privacy guarantee proved to increase model utility. We show this is not necessary and instead propose that utility be improved by choosing activation functions designed explicitly for privacy-preserving training. A crucial operation in differentially private SGD is gradient clipping, which along with modifying the optimization path (at times resulting in not-optimizing a single objective function), may also introduce both significant bias and variance to the learning process. We empirically identify exploding gradients arising from ReLU may be one of the main sources of this. We demonstrate analytically and experimentally how a general family of bounded activation functions, the tempered sigmoids, consistently outperform the currently established choice: unbounded activation functions like ReLU. Using this paradigm, we achieve new state-of-the-art accuracy on MNIST, FashionMNIST, and CIFAR10 without any modification of the learning procedure fundamentals or differential privacy analysis. While the changes we make are simple in retrospect, the simplicity of our approach facilitates its implementation and adoption to meaningfully improve state-of-the-art machine learning while still providing strong guarantees in the original framework of differential privacy.

We revisit the well-studied problem of differentially private empirical risk minimization (ERM). We show that for unconstrained convex generalized linear models (GLMs), one can obtain an excess empirical risk of $\tilde O\left(\sqrt{\rank}/\epsilon n\right)$, where $\rank$ is the rank of the feature matrix in the GLM problem, $n$ is the number of data samples, and $\epsilon$ is the privacy parameter. This bound is attained via differentially private gradient descent (DP-GD). Furthermore, via the \emph{first lower bound for unconstrained private ERM}, we show that our upper bound is tight. In sharp contrast to the constrained ERM setting, there is no dependence on the dimensionality of the ambient model space ($p$). (Notice that $\rank\leq \min\{n, p\}$.) Besides, we obtain an analogous excess population risk bound which depends on $\rank$ instead of $p$. For the smooth non-convex GLM setting (i.e., where the objective function is non-convex but preserves the GLM structure), we further show that DP-GD attains a dimension-independent convergence of $\tilde O\left(\sqrt{\rank}/\epsilon n\right)$ to a first-order-stationary-point of the underlying objective. Finally, we show that for convex GLMs, a variant of DP-GD commonly used in practice (which involves clipping the individual gradients) also exhibits the same dimension-independent convergence to the minimum of a well-defined objective. To that end, we provide a structural lemma that characterizes the effect of clipping on the optimization profile of DP-GD.

We revisit the problem of empirical risk minimziation (ERM) with differential privacy. We show that noisy AdaGrad, given appropriate knowledge and conditions on the subspace from which gradients can be drawn, achieves a regret comparable to traditional AdaGrad plus a well-controlled term due to noise. We show a convergence rate of $O(\tr(G_T)/T)$, where $G_T$ captures the geometry of the gradient subspace. Since $\tr(G_T)=O(\sqrt{T})$ we can obtain faster rates for convex and Lipschitz functions, compared to the $O(1/\sqrt{T})$ rate achieved by known versions of noisy (stochastic) gradient descent with comparable noise variance. In particular, we show that if the gradients lie in a known constant rank subspace, and assuming algorithmic access to an envelope which bounds decaying sensitivity, one can achieve faster convergence to an excess empirical risk of $\tilde O(1/\epsilon n)$, where $\epsilon$ is the privacy budget and $n$ the number of samples. Letting $p$ be the problem dimension, this result implies that, by running noisy Adagrad, we can bypass the DP-SGD bound $\tilde O(\sqrt{p}/\epsilon n)$ in $T=(\epsilon n)^{2/(1+2\alpha)}$ iterations, where $\alpha \geq 0$ is a parameter controlling gradient norm decay, instead of the rate achieved by SGD of $T=\epsilon^2n^2$. Our results operate with general convex functions in both constrained and unconstrained minimization. Along the way, we do a perturbation analysis of noisy AdaGrad, which is of independent interest. Our utility guarantee for the private ERM problem follows as a corollary to the regret guarantee of noisy AdaGrad.

We study the problem of differentially private (DP) matrix completion under user-level privacy. We design a joint differentially private variant of the popular Alternating-Least-Squares (ALS) method that achieves: i) (nearly) optimal sample complexity for matrix completion (in terms of number of items, users), and ii) the best known privacy/utility trade-off both theoretically, as well as on benchmark data sets. In particular, we provide the first global convergence analysis of ALS with noise introduced to ensure DP, and show that, in comparison to the best known alternative (the Private Frank-Wolfe algorithm by Jain et al. (2018)), our error bounds scale significantly better with respect to the number of items and users, which is critical in practical problems. Extensive validation on standard benchmarks demonstrate that the algorithm, in combination with carefully designed sampling procedures, is significantly more accurate than existing techniques, thus promising to be the first practical DP embedding model.

We consider training models with differential privacy (DP) using mini-batch gradients. The existing state-of-the-art, Differentially Private Stochastic Gradient Descent (DP-SGD), requires \emph{privacy amplification by sampling or shuffling} to obtain the best privacy/accuracy/computation trade-offs. Unfortunately, the precise requirements on exact sampling and shuffling can be hard to obtain in important practical scenarios, particularly federated learning (FL). We design and analyze a DP variant of Follow-The-Regularized-Leader (DP-FTRL) that compares favorably (both theoretically and empirically) to amplified DP-SGD, while allowing for much more flexible data access patterns. DP-FTRL does not use any form of privacy amplification.

Poisoning attacks have emerged as a significant security threat to machine learning algorithms. It has been demonstrated that adversaries who make small changes to the training set, such as adding specially crafted data points, can hurt the performance of the output model. Most of these attacks require the full knowledge of training data. This leaves open the possibility of achieving the same attack results using poisoning attacks that do not have the full knowledge of the clean training set.In this work, we initiate a theoretical study of the problem above. Specifically, for the case of feature selection with LASSO, we show that \emph{full information} adversaries (that craft poisoning examples based on the rest of the training data) are provably much more devastating compared to the optimal attacker that is \emph{oblivious} to the training set yet has access to the distribution of the data. Our separation result shows that the two settings of data-aware and data-oblivious are fundamentally different and we cannot hope to achieve the same attack or defense results in these scenarios.

We study personalization of supervised learning with user-level differential privacy. Consider a setting with many users, each of whom has a training data set drawn from their own distribution $P_i$. Assuming some shared structure among the problems $P_i$, can users collectively learn the shared structure---and solve their tasks better than they could individually---while preserving the privacy of their data? We formulate this question using joint, user-level differential privacy---that is, we control what is leaked about each user's entire data set. We provide algorithms that exploit popular non-private approaches in this domain like the Almost-No-Inner-Loop (ANIL) method, and give strong user-level privacy guarantees for our general approach. When the problems $P_i$ are linear regression problems with each user's regression vector lying in a common, unknown low-dimensional subspace, we show that our efficient algorithms satisfy nearly optimal estimation error guarantees. We also establish a general, information-theoretic upper bound via an exponential mechanism-based algorithm.

A private machine learning algorithm hides as much as possible about its training data while still preserving accuracy. In this work, we study whether a non-private learning algorithm can be made private by relying on an instance-encoding mechanism that modifies the training inputs before feeding them to a normal learner. We formalize both the notion of instance encoding and its privacy by providing two attack models. We first prove impossibility results for achieving a (stronger) model. Next, we demonstrate practical attacks in the second (weaker) attack model on InstaHide, a recent proposal by Huang, Song, Li and Arora [ICML’20] that aims to use instance encoding for privacy.

Differentially private (DP) machine learning allows us to train models on private data while limiting data leakage. DP formalizes this data leakage through a cryptographic game, where an adversary must predict if a model was trained on a dataset D, or a dataset D′ that differs in just one example. If observing the training algorithm does not meaningfully increase the adversary's odds of successfully guessing which dataset the model was trained on, then the algorithm is said to be differentially private. Hence, the purpose of privacy analysis is to upper bound the probability that any adversary could successfully guess which dataset the model was trained on.In our paper, we instantiate this hypothetical adversary in order to establish lower bounds on the probability that this distinguishing game can be won. We use this adversary to evaluate the importance of the adversary capabilities allowed in the privacy analysis of DP training algorithms.For DP-SGD, the most common method for training neural networks with differential privacy, our lower bounds are tight and match the theoretical upper bound. This implies that in order to prove better upper bounds, it will be necessary to make use of additional assumptions. Fortunately, we find that our attacks are significantly weaker when additional (realistic) restrictions are put in place on the adversary's capabilities. Thus, in the practical setting common to many real-world deployments, there is a gap between our lower bounds and the upper bounds provided by the analysis: differential privacy is conservative and adversaries may not be able to leak as much information as suggested by the theoretical bound.

Differentially Private Stochastic Gradient Descent (DP-SGD) forms a fundamental building block in many applications for learning over sensitive data. Two standard approaches, privacy amplification by subsampling, and privacy amplification by shuffling, permit adding lower noise in DP-SGD than via na\"{\i}ve schemes. A key assumption in both these approaches is that the elements in the data set can be uniformly sampled, or be uniformly permuted --- constraints that may become prohibitive when the data is processed in a decentralized or distributed fashion. In this paper, we focus on conducting iterative methods like DP-SGD in the setting of federated learning (FL) wherein the data is distributed among many devices (clients). Our main contribution is the \emph{random check-in} distributed protocol, which crucially relies only on randomized participation decisions made locally and independently by each client. It has privacy/accuracy trade-offs similar to privacy amplification by subsampling/shuffling. However, our method does not require server-initiated communication, or even knowledge of the population size. To our knowledge, this is the first privacy amplification tailored for a distributed learning framework, and it may have broader applicability beyond FL. Along the way, we improve the privacy guarantees of amplification by shuffling and show that, in practical regimes, this improvement allows for similar privacy and utility using data from an order of magnitude fewer users.

We revisit the problem of counting the number of distinct elements $\dist$ in a data stream $D$, over a domain $[u]$. We propose an $(\epsilon,\delta)$-differentially private algorithm that approximates $\dist$ within a factor of $(1\pm\gamma)$, and with additive error of $O(\sqrt{\ln(1/\delta)}/\epsilon)$, using space $O(\ln(\ln(u)/\gamma)/\gamma^2)$. We improve on the prior work at least quadratically and up to exponentially, in terms of both space and additive error. Our additive error guarantee is optimal up to a factor of $O(\sqrt{\ln(1/\delta)})$, and the space bound is optimal up to a factor of $O\left(\min\left\{\ln\left(\frac{\ln(u)}{\gamma}\right), \frac{1}{\gamma^2}\right\}\right)$. We assume the existence of an ideal uniform random hash function, and ignore the space required to store it. We later relax this requirement by assuming pseudorandom functions and appealing to a computational variant of differential privacy, SIM-CDP. Our algorithm is built on top of the celebrated Flajolet-Martin (FM) sketch. We show that FM-sketch is differentially private as is, as long as there are $\approx \sqrt{\ln(1/\delta)}/(\epsilon\gamma)$ distinct elements in the data set. Along the way, we prove a structural result showing that the maximum of $k$ i.i.d. random variables is statistically close (in the sense of $\epsilon$-differential privacy) to the maximum of $(k+1)$ i.i.d. samples from the same distribution, as long as $k=\Omega\left(\frac{1}{\epsilon}\right)$. Finally, experiments show that our algorithms introduces error within an order of magnitude of the non-private analogues for streams with thousands of distinct elements, even while providing strong privacy guarantee ($\eps\leq 1$).

We present new practical local differentially private heavy hitters algorithms achieving optimal or near-optimal worst-case error and running time -- TreeHist and Bitstogram. In both algorithms, server running time is $\tilde O(n)$ and user running time is $\tilde O(1)$, hence improving on the prior state-of-the-art result of Bassily and Smith [STOC 2015] requiring $O(n^{5/2})$ server time and $O(n^{3/2})$ user time. With a typically large number of participants in local algorithms (in the millions), this reduction in time complexity, in particular at the user side, is crucial for making locally private heavy hitters algorithms usable in practice. We implemented Algorithm TreeHist to verify our theoretical analysis and compared its performance with the performance of Google's RAPPOR code.

We study differentially private (DP) algorithms for stochastic convex optimization (SCO). In this problem the goal is to approximately minimize the population loss given i.i.d.~samples from a distribution over convex and Lipschitz loss functions. A long line of existing work on private convex optimization focuses on the empirical loss and derives asymptotically tight bounds on the excess empirical loss. However a significant gap exists in the known bounds for the population loss. We show that, up to logarithmic factors, the optimal excess population loss for DP algorithms is equal to the larger of the optimal non-private excess population loss, and the optimal excess empirical loss of DP algorithms. This implies that, contrary to intuition based on private ERM, private SCO has asymptotically the same rate of $1/\sqrt{n}$ as non-private SCO in the parameter regime most common in practice. The best previous result in this setting gives rate of $1/n^{1/4}$. Our approach builds on existing differentially private algorithms and relies on the analysis of algorithmic stability to ensure generalization.

Sensitive statistics are often collected across sets of users, with repeated collection of reports done over time. For example, trends in users’ private preferences or software usage may be monitored via such reports. We study the collection of such statistics in the local differential privacy (LDP) model, and describe an algorithm whose privacy cost is polylogarithmic in the number of changes to a user's value. More fundamentally—by building on anonymity of the users’ reports—we also demonstrate how the privacy cost of our LDP algorithm can actually be much lower when viewed in the central model of differential privacy. We show, via a new and general privacy amplification technique, that any permutation-invariant algorithm satisfying ε-local differential privacy will satisfy -central differential privacy. By this, we explain how the high noise and overhead of LDP protocols is a consequence of them being significantly more private in the central model. As a practical corollary, our results imply that several LDP-based industrial deployments may have much lower privacy cost than their advertised ε would indicate—at least if reports are anonymized.

Building useful predictive models often involves learning from sensitive data. Training models with differential privacy can guarantee the privacy of such sensitive data. For convex optimization tasks, several differentially private algorithms are known, but none has yet been deployed in practice. In this work, we make two major contributions towards practical differentially private convex optimization. First, we present Approximate Minima Perturbation, a novel algorithm that can leverage any off-the-shelf optimizer. We show that it can be employed without any hyperparameter tuning, thus making it an attractive technique for practical deployment. Second, we perform an extensive empirical evaluation of the state-of-the-art algorithms for differentially private convex optimization, on a range of publicly available benchmark datasets, and real-world datasets obtained through an industrial collaboration. We release open-source implementations of all the differentially private convex optimization algorithms considered, and benchmarks on as many as nine public datasets, four of which are high-dimensional.

Preserving privacy of users is a key requirement of web-scale data mining applications and systems such as web search, recommender systems, crowdsourced platforms, and analytics applications, and has witnessed a renewed focus in light of recent data breaches and new regulations such as GDPR. In this tutorial, we will first present an overview of privacy breaches over the last two decades and the lessons learned, key regulations and laws, and evolution of privacy techniques leading to differential privacy definition / techniques. Then, we will focus on the application of privacy-preserving data mining techniques in practice, by presenting case studies such as Apple's differential privacy deployment for iOS / macOS, Google's RAPPOR, LinkedIn Salary, and Microsoft's differential privacy deployment for collecting Windows telemetry. We will conclude with open problems and challenges for the data mining / machine learning community, based on our experiences in industry.

Many commonly used learning algorithms work by iteratively updating an intermediate solution using one or a few data points in each iteration. Analysis of differential privacy for such algorithms often involves ensuring privacy of each step and then reasoning about the cumulative privacy cost of the algorithm. This is enabled by composition theorems for differential privacy that allow releasing of all the intermediate results. In this work, we demonstrate that for contractive iterations, not releasing the intermediate results strongly amplifies the privacy guarantees. We describe several applications of this new analysis technique to solving convex optimization problems via noisy stochastic gradient descent. For example, we demonstrate that a relatively small number of non-private data points from the same distribution can be used to close the gap between private and non-private convex optimization. In addition, we demonstrate that we can achieve guarantees similar to those obtainable using the privacy-amplification-by-sampling technique in several natural settings where that technique cannot be applied.

We provide the first provably joint differentially private algorithm with formal utility guarantees for the problem of user-level privacy-preserving collaborative filtering. Our algorithm is based on the Frank-Wolfe method, and it consistently estimates the underlying preference matrix as long as the number of users $m$ is $\omega(n^{5/4})$, where $n$ is the number of items, and each user provides her preference for at least $\sqrt{n}$ randomly selected items. Along the way, we provide an optimal differentially private algorithm for singular vector computation, based on the celebrated Oja’s method, that provides significant savings in terms of space and time while operating on sparse matrices. We also empirically evaluate our algorithm on a suite of datasets, and show that it consistently outperforms the state-of-the-art private algorithms.

We design differentially private learning algorithms that are agnostic to the learning model assuming access to limited amount of unlabeled public data. First, we give a new differentially private algorithm for answering a sequence of $m$ online classification queries (given by a sequence of $m$ unlabeled public feature vectors) based on a private training set. Our private algorithm follows the paradigm of subsample-and-aggregate, in which any generic non-private learner is trained on disjoint subsets of the private training set, then for each classification query, the votes of the resulting classifiers ensemble are aggregated in a differentially private fashion. Our private aggregation is based on a novel combination of distance-to-instability framework [Smith & Thakurta 2013] and the sparse-vector technique [Dwork et al. 2009, Hardt & Talwar 2010]. We show that our algorithm makes a conservative use of the privacy budget. In particular, if the underlying non-private learner yields classification error at most $\alpha\in (0, 1)$, then our construction answers more queries, by at least a factor of $1/\alpha$ in some cases, than what is implied by a straightforward application of the advanced composition theorem for differential privacy. Next, we apply the knowledge transfer technique to construct a private learner that outputs a classifier, which can be used to answer unlimited number of queries. In the PAC model, we analyze our construction and prove upper bounds on the sample complexity for both the realizable and the non-realizable cases. As in non-private sample complexity, our bounds are completely characterized by the VC dimension of the concept class.

Property testers form an important class of sublinear-time algorithms. In the standard property testing model, an algorithm accesses the input function $f :\mathcal{D} \mapsto {\cal R}$ via an oracle. With very few exceptions, all property testers studied in this model rely on the oracle to provide function values at all queried domain points. However, in many realistic situations, the oracle may be unable to reveal the function values at some domain points due to privacy concerns, or when some of the values get erased by mistake or by an adversary. The testers do not learn anything useful about the function by querying those erased points. Moreover, the knowledge of a tester may enable an adversary to erase some of the values so as to increase the query complexity of the tester arbitrarily or, in some cases, make the tester entirely useless. In this work, we initiate a study of property testers that are resilient to the presence of adversarially erased function values. An $\alpha$-erasure-resilient $\varepsilon$-tester is given parameters $\alpha \in [0,1),\varepsilon\in (0,1)$, along with oracle access to a function $f$ such that at most an $\alpha$ fraction of function values have been erased. The tester does not know whether a value is erased until it queries the corresponding domain point. The tester has to accept with high probability if there is a way to assign values to the erased points such that the resulting function satisfies the desired property $\mathcal{P}$. It has to reject with high probability if, for every assignment of values to the erased points, the resulting function has to be changed in at least an $\varepsilon$ fraction of the nonerased domain points to satisfy $\mathcal{P}$. Erasure-resilient testing generalizes the standard property testing model of Rubinfeld and Sudan [ SIAM J. Comput., 25 (1996), pp. 252--271] and Goldreich, Goldwasser, and Ron [ J. ACM, 45 (1998), pp. 653--750]. Compared to the tolerant testing model of Parnas, Ron, and Rubinfeld [ J. Comput. System Sci., 6 (2006), pp. 1012--1042], our model places less stringent requirements on the tester. We design erasure-resilient property testers for a large class of properties. For some properties, it is possible to obtain erasure-resilient testers by simply using standard testers as a black box. However, for some more challenging properties, all existing algorithms are more likely to query certain points in the domain. If these points are erased, the algorithms break. We give efficient erasure-resilient testers for several important classes of such properties of functions including monotonicity, the Lipschitz property, and convexity. Finally, we show a separation between the standard and erasure-resilient testing. Specifically, we describe a property that can be $\varepsilon$-tested with $O(1/\varepsilon)$ queries in the standard model, whereas testing it in the erasure-resilient model requires a number of queries polynomial in the input size.

We present new practical local differentially private heavy hitters algorithms achieving optimal or near-optimal worst-case error -- TreeHist and Bitstogram. In both algorithms, server running time is $\tilde O(n)$ and user running time is $\tilde O(1)$, hence improving on the prior state-of-the-art result of Bassily and Smith [STOC 2015] requiring $\tilde O(n^{5/2})$ server time and $\tilde O(n^{3/2})$ user time. With a typically large number of participants in local algorithms ($n$ in the millions), this reduction in time complexity, in particular at the user side, is crucial for the use of such algorithms in practice. We implemented Algorithm TreeHist to verify our theoretical analysis and compared its performance with the performance of Google's RAPPOR code.

Recent large-scale deployments of differentially private algorithms employ the local model for privacy (sometimes called PRAM or randomized response), where data are randomized on each individual's device before being sent to a server that computes approximate, aggregate statistics. The server need not be trusted for privacy, leaving data control in users' hands. For an important class of convex optimization problems (including logistic regression, support vector machines, and the Euclidean median), the best known locally differentially-private algorithms are highly interactive, requiring as many rounds of back and forth as there are users in the protocol. We ask: how much interaction is necessary to optimize convex functions in the local DP model? Existing lower bounds either do not apply to convex optimization, or say nothing about interaction. We provide new algorithms which are either noninteractive or use relatively few rounds of interaction. We also show lower bounds on the accuracy of an important class of noninteractive algorithms, suggesting a separation between what is possible with and without interaction.

We present a nearly optimal differentially private version of the well known LASSO estimator. Our algorithm provides privacy protection with respect to each training data item. The excess risk of our algorithm, compared to the non-private version, is $\widetilde{O}(1/n^{2/3})$, assuming all the input data has bounded $\ell_\infty$ norm. This is the first differentially private algorithm that achieves such a bound without the polynomial dependence on $p$ under no addition assumption on the design matrix. In addition, we show that this error bound is nearly optimal amongst all differentially private algorithms.

Convex empirical risk minimization is a basic tool in machine learning and statistics. We provide new algorithms and matching lower bounds for differentially private convex empirical risk minimization assuming only that each data point's contribution to the loss function is Lipschitz and that the domain of optimization is bounded. We provide a separate set of algorithms and matching lower bounds for the setting in which the loss functions are known to also be strongly convex. Our algorithms run in polynomial time, and in some cases even match the optimal nonprivate running time (as measured by oracle complexity). We give separate algorithms (and lower bounds) for (ε, 0)and (ε, δ)-differential privacy; perhaps surprisingly, the techniques used for designing optimal algorithms in the two cases are completely different. Our lower bounds apply even to very simple, smooth function families, such as linear and quadratic functions. This implies that algorithms from previous work can be used to obtain optimal error rates, under the additional assumption that the contributions of each data point to the loss function is smooth. We show that simple approaches to smoothing arbitrary loss functions (in order to apply previous techniques) do not yield optimal error rates. In particular, optimal algorithms were not previously known for problems such as training support vector machines and the high-dimensional median.

In this paper, we study the problem of differentially private risk minimization where the goal is to provide differentially private algorithms that have small excess risk. In particular we address the following open problem: \emphIs it possible to design computationally efficient differentially private risk minimizers with excess risk bounds that do not explicitly depend on dimensionality (p) and do not require structural assumptions like restricted strong convexity? In this paper, we answer the question in the affirmative for a variant of the well-known \emphoutput and \emphobjective perturbation algorithms [Chaudhuri et al., 2011]. In particular, we show that in generalized linear model, variants of both output and objective perturbation algorithms have no \em explicit dependence on p. Our results assume that the underlying loss function is a 1-Lipschitz convex function and we show that the excess risk depends only on L_2 norm of the true risk minimizer and that of training points. Next, we present a novel privacy preserving algorithm for risk minimization over simplex in the generalized linear model, where the loss function is a doubly differentiable convex function. Assuming that the training points have bounded L_∞-norm, our algorithm provides risk bound that has only \em logarithmic dependence on p. We also apply our technique to the online learning setting and obtain a regret bound with similar logarithmic dependence on p. In contrast, the existing differentially private online learning methods incur O(\sqrtp) dependence.

We consider the problem of privately releasing a low dimensional approximation to a set of data records, represented as a matrix A in which each row corresponds to an individual and each column to an attribute. Our goal is to compute a subspace that captures the covariance of A as much as possible, classically known as principal component analysis (PCA). We assume that each row of A has ℓ2 norm bounded by one, and the privacy guarantee is defined with respect to addition or removal of any single row. We show that the well-known, but misnamed, randomized response algorithm, with properly tuned parameters, provides nearly optimal additive quality gap compared to the best possible singular subspace of A. We further show that when ATA has a large eigenvalue gap -- a reason often cited for PCA -- the quality improves significantly. Optimality (up to logarithmic factors) is proved using techniques inspired by the recent work of Bun, Ullman, and Vadhan on applying Tardos's fingerprinting codes to the construction of hard instances for private mechanisms for 1-way marginal queries. Along the way we define a list culling game which may be of independent interest.

We design differentially private algorithms for statistical model selection. Given a data set and a large, discrete collection of “models”, each of which is a family of probability distributions, the goal is to determine the model that best “fits” the data. This is a basic problem in many areas of statistics and machine learning. We consider settings in which there is a well-defined answer, in the following sense: Suppose that there is a \emphnonprivate model selection procedure f, which is the reference to which we compare our performance. Our differentially private algorithms output the correct value f(D) whenever f is \emphstable on the input data set D. We work with two notions, \emphperturbation stability and \emphsub-sampling stability. We give two classes of results: generic ones, that apply to any function with discrete output set; and specific algorithms for the problem of sparse linear regression. The algorithms we describe are efficient and in some cases match the optimal \emphnon-private asymptotic sample complexity. Our algorithms for sparse linear regression require analyzing the stability properties of the popular LASSO estimator. We give sufficient conditions for the LASSO estimator to be robust to small changes in the data set, and show that these conditions hold with high probability under essentially the same stochastic assumptions that are used in the literature to analyze convergence of the LASSO.

In this paper, we consider the problem of differentially private learning where access to the training features is through a kernel function only. Existing work on this problem is restricted to translation invariant kernels only, where (approximate) training features are available explicitly. In fact, for general class of kernel functions and in general setting of releasing different private predictor (\w^*), the problem is impossible to solve \citeCMS11. In this work, we relax the problem setting into three different easier but practical settings. In our first problem setting, we consider an interactive model where the user sends its test set to a trusted learner who sends back differentially private predictions over the test points. This setting is prevalent in modern online systems like search engines, ad engines etc. In the second model, the learner sends back a differentially private version of the optimal parameter vector \w^* but requires access to a small subset of unlabeled test set beforehand. This also is a practical setting that involves two parties interacting through trusted third party. Our third model is similar to the traditional model, where learner is oblivious to the test set and needs to send a differentially private version of \w^*, but the kernels are restricted to efficiently computable functions over low-dimensional vector spaces. For each of the models, we derive differentially private learning algorithms with provable “utlity” or error bounds. Moreover, we show that our methods can also be applied to the traditional setting of \cite Rubinstein09, CMS11. Here, our sample complexity bounds have only O(d^1/3) dependence on the dimensionality d while existing methods require O(d^1/2) samples to achieve same generalization error.

We provide a general technique for making online learning algorithms differentially private, in both the full information and bandit settings. Our technique applies to algorithms that aim to minimize a \emph{convex} loss function which is a sum of smaller convex loss terms, one for each data point. We modify the popular \emph{mirror descent} approach, or rather a variant called \emph{follow the approximate leader}. The technique leads to the first nonprivate algorithms for private online learning in the bandit setting. In the full information setting, our algorithms improve over the regret bounds of previous work. In many cases, our algorithms (in both settings) matching the dependence on the input length, $T$, of the \emph{optimal nonprivate} regret bounds up to logarithmic factors in $T$. Our algorithms require logarithmic space and update time.

It is often highly valuable for organizations to have their data analyzed by external agents. However, any program that computes on potentially sensitive data risks leaking information through its output. Differential privacy provides a theoretical framework for processing data while protecting the privacy of individual records in a dataset. Unfortunately, it has seen limited adoption because of the loss in output accuracy, the difficulty in making programs differentially private, lack of mechanisms to describe the privacy budget in a programmer's utilitarian terms, and the challenging requirement that data owners and data analysts manually distribute the limited privacy budget between queries.

Discovering frequent patterns from data is a popular exploratory technique in datamining. However, if the data are sensitive (e.g., patient health records, user behavior records) releasing information about significant patterns or trends carries significant risk to privacy. This paper shows how one can accurately discover and release the most significant patterns along with their frequencies in a data set containing sensitive information, while providing rigorous guarantees of privacy for the individuals whose information is stored there.