arXiv subjects

Farhad Farokhi

Publications and source records attributed to Farhad Farokhi.

At least 55 records · Page 3Linked to original sources

arXiv2020

Deconvoluting Kernel Density Estimation and Regression for Locally Differentially Private Data

Local differential privacy has become the gold-standard of privacy literature for gathering or releasing sensitive individual data points in a privacy-preserving manner. However, locally differential data can twist the probability density of the data because of the additive noise used to ensure privacy. In fact, the density of privacy-preserving data (no matter how many samples we gather) is always flatter in comparison with the density function of the original data points due to convolution with privacy-preserving noise density function. The effect is especially more pronounced when using slow-decaying privacy-preserving noises, such as the Laplace noise. This can result in under/over-estimation of the heavy-hitters. This is an important challenge facing social scientists due to the use of differential privacy in the 2020 Census in the United States. In this paper, we develop density estimation methods using smoothing kernels. We use the framework of deconvoluting kernel density estimators to remove the effect of privacy-preserving noise. This approach also allows us to adapt the results from non-parameteric regression with errors-in-variables to develop regression models based on locally differentially private data. We demonstrate the performance of the developed methods on financial and demographic datasets.

math.ST↗

arXiv2020

Non-Stochastic Private Function Evaluation

We consider private function evaluation to provide query responses based on private data of multiple untrusted entities in such a way that each cannot learn something substantially new about the data of others. First, we introduce perfect non-stochastic privacy in a two-party scenario. Perfect privacy amounts to conditional unrelatedness of the query response and the private uncertain variable of other individuals conditioned on the uncertain variable of a given entity. We show that perfect privacy can be achieved for queries that are functions of the common uncertain variable, a generalization of the common random variable. We compute the closest approximation of the queries that do not take this form. To provide a trade-off between privacy and utility, we relax the notion of perfect privacy. We define almost perfect privacy and show that this new definition equates to using conditional disassociation instead of conditional unrelatedness in the definition of perfect privacy. Then, we generalize the definitions to multi-party function evaluation (more than two data entities). We prove that uniform quantization of query responses, where the quantization resolution is a function of privacy budget and sensitivity of the query (cf., differential privacy), achieves function evaluation privacy.

cs.IT↗

arXiv2020

Structured preconditioning of conjugate gradients for path-graph network optimal control problems

A structured preconditioned conjugate gradient (PCG) solver is developed for the Newton steps in second-order methods for a class of constrained network optimal control problems. Of specific interest are problems with discrete-time dynamics arising from the path-graph interconnection of $N$ heterogeneous sub-systems. The computational complexity of each PGC step is shown to be $O(NT)$, where $T$ is the length of the time horizon. The proposed preconditioning involves a fixed number of block Jacobi iterations per PCG step. A decreasing analytic bound on the effective conditioning is given in terms of this number. The computations are decomposable across the spatial and temporal dimensions of the optimal control problem, into sub-problems of size independent of $N$ and $T$. Numerical results are provided for a mass-spring-damper chain.

eess.SY↗

arXiv2020

Online Stochastic Convex Optimization: Wasserstein Distance Variation

Distributionally-robust optimization is often studied for a fixed set of distributions rather than time-varying distributions that can drift significantly over time (which is, for instance, the case in finance and sociology due to underlying expansion of economy and evolution of demographics). This motivates understanding conditions on probability distributions, using the Wasserstein distance, that can be used to model time-varying environments. We can then use these conditions in conjunction with online stochastic optimization to adapt the decisions. We considers an online proximal-gradient method to track the minimizers of expectations of smooth convex functions parameterised by a random variable whose probability distributions continuously evolve over time at a rate similar to that of the rate at which the decision maker acts. We revisit the concepts of estimation and tracking error inspired by systems and control literature and provide bounds for them under strong convexity, Lipschitzness of the gradient, and bounds on the probability distribution drift. Further, noting that computing projections for a general feasible sets might not be amenable to online implementation (due to computational constraints), we propose an exact penalty method. Doing so allows us to relax the uniform boundedness of the gradient and establish dynamic regret bounds for tracking and estimation error. We further introduce a constraint-tightening approach and relate the amount of tightening to the probability of satisfying the constraints.

math.OC↗

arXiv2020

Security Versus Privacy

Linear queries can be submitted to a server containing private data. The server provides a response to the queries systematically corrupted using an additive noise to preserve the privacy of those whose data is stored on the server. The measure of privacy is inversely proportional to the trace of the Fisher information matrix. It is assumed that an adversary can inject a false bias to the responses. The measure of the security, capturing the ease of detecting the presence of the false data injection, is the sensitivity of the Kullback-Leiber divergence to the additive bias. An optimization problem for balancing privacy and security is proposed and subsequently solved. It is shown that the level of guaranteed privacy times the level of security equals a constant. Therefore, by increasing the level of privacy, the security guarantees can only be weakened and vice versa. Similar results are developed under the differential privacy framework.

cs.CR↗

arXiv2020

Data and Model Dependencies of Membership Inference Attack

Machine learning (ML) models have been shown to be vulnerable to Membership Inference Attacks (MIA), which infer the membership of a given data point in the target dataset by observing the prediction output of the ML model. While the key factors for the success of MIA have not yet been fully understood, existing defense mechanisms such as using L2 regularization \cite{10shokri2017membership} and dropout layers \cite{salem2018ml} take only the model's overfitting property into consideration. In this paper, we provide an empirical analysis of the impact of both the data and ML model properties on the vulnerability of ML techniques to MIA. Our results reveal the relationship between MIA accuracy and properties of the dataset and training model in use. In particular, we show that the size of shadow dataset, the class and feature balance and the entropy of the target dataset, the configurations and fairness of the training model are the most influential factors. Based on those experimental findings, we conclude that along with model overfitting, multiple properties jointly contribute to MIA success instead of any single property. Building on our experimental findings, we propose using those data and model properties as regularizers to protect ML models against MIA. Our results show that the proposed defense mechanisms can reduce the MIA accuracy by up to 25\% without sacrificing the ML model prediction utility.

cs.LG↗

arXiv2020

Developing Non-Stochastic Privacy-Preserving Policies Using Agglomerative Clustering

We consider a non-stochastic privacy-preserving problem in which an adversary aims to infer sensitive information $S$ from publicly accessible data $X$ without using statistics. We consider the problem of generating and releasing a quantization $\hat{X}$ of $X$ to minimize the privacy leakage of $S$ to $\hat{X}$ while maintaining a certain level of utility (or, inversely, the quantization loss). The variables $S$ and $S$ are treated as bounded and non-probabilistic, but are otherwise general. We consider two existing non-stochastic privacy measures, namely the maximum uncertainty reduction $L_0(S \rightarrow \hat{X})$ and the refined information $I_*(S; \hat{X})$ (also called the maximin information) of $S$. For each privacy measure, we propose a corresponding agglomerative clustering algorithm that converges to a locally optimal quantization solution $\hat{X}$ by iteratively merging elements in the alphabet of $X$. To instantiate the solution to this problem, we consider two specific utility measures, the worst-case resolution of $X$ by observing $\hat{X}$ and the maximal distortion of the released data $\hat{X}$. We show that the value of the maximin information $I_*(S; \hat{X})$ can be determined by dividing the confusability graph into connected subgraphs. Hence, $I_*(S; \hat{X})$ can be reduced by merging nodes connecting subgraphs. The relation to the probabilistic information-theoretic privacy is also studied by noting that the G{á}cs-K{ö}rner common information is the stochastic version of $I_*$ and indicates the attainability of statistical indistinguishability.

cs.CR↗

arXiv2020

The Cost of Privacy in Asynchronous Differentially-Private Machine Learning

We consider training machine learning models using Training data located on multiple private and geographically-scattered servers with different privacy settings. Due to the distributed nature of the data, communicating with all collaborating private data owners simultaneously may prove challenging or altogether impossible. In this paper, we develop differentially-private asynchronous algorithms for collaboratively training machine-learning models on multiple private datasets. The asynchronous nature of the algorithms implies that a central learner interacts with the private data owners one-on-one whenever they are available for communication without needing to aggregate query responses to construct gradients of the entire fitness function. Therefore, the algorithm efficiently scales to many data owners. We define the cost of privacy as the difference between the fitness of a privacy-preserving machine-learning model and the fitness of trained machine-learning model in the absence of privacy concerns. We prove that we can forecast the performance of the proposed privacy-preserving asynchronous algorithms. We demonstrate that the cost of privacy has an upper bound that is inversely proportional to the combined size of the training datasets squared and the sum of the privacy budgets squared. We validate the theoretical results with experiments on financial and medical datasets. The experiments illustrate that collaboration among more than 10 data owners with at least 10,000 records with privacy budgets greater than or equal to 1 results in a superior machine-learning model in comparison to a model trained in isolation on only one of the datasets, illustrating the value of collaboration and the cost of the privacy. The number of the collaborating datasets can be lowered if the privacy budget is higher.

cs.LG↗

arXiv2020

Distributionally-Robust Machine Learning Using Locally Differentially-Private Data

We consider machine learning, particularly regression, using locally-differentially private datasets. The Wasserstein distance is used to define an ambiguity set centered at the empirical distribution of the dataset corrupted by local differential privacy noise. The ambiguity set is shown to contain the probability distribution of unperturbed, clean data. The radius of the ambiguity set is a function of the privacy budget, spread of the data, and the size of the problem. Hence, machine learning with locally-differentially private datasets can be rewritten as a distributionally-robust optimization. For general distributions, the distributionally-robust optimization problem can relaxed as a regularized machine learning problem with the Lipschitz constant of the machine learning model as a regularizer. For linear and logistic regression, this regularizer is the dual norm of the model parameters. For Gaussian data, the distributionally-robust optimization problem can be solved exactly to find an optimal regularizer. This approach results in an entirely new regularizer for training linear regression models. Training with this novel regularizer can be posed as a semi-definite program. Finally, the performance of the proposed distributionally-robust machine learning training is demonstrated on practical datasets.

cs.LG↗

arXiv2020

An Explicit Formula for the Zero-Error Feedback Capacity of a Class of Finite-State Additive Noise Channels

It is known that for a discrete channel with correlated additive noise, the ordinary capacity with or without feedback both equal $ \log q-\mathcal{H} (Z) $, where $ \mathcal{H}(Z) $ is the entropy rate of the noise process $ Z $ and $ q $ is the alphabet size. In this paper, a class of finite-state additive noise channels is introduced. It is shown that the zero-error feedback capacity of such channels is either zero or $C_{0f} =\log q -h (Z) $, where $ h (Z) $ is the {\em topological entropy} of the noise process. A topological condition is given when the zero-error capacity is zero, with or without feedback. Moreover, the zero-error capacity without feedback is lower-bounded by $ \log q-2 h (Z) $. We explicitly compute the zero-error feedback capacity for several examples, including channels with isolated errors and a Gilbert-Elliot channel.

cs.IT↗

arXiv2020

Modelling and Quantifying Membership Information Leakage in Machine Learning

Machine learning models have been shown to be vulnerable to membership inference attacks, i.e., inferring whether individuals' data have been used for training models. The lack of understanding about factors contributing success of these attacks motivates the need for modelling membership information leakage using information theory and for investigating properties of machine learning models and training algorithms that can reduce membership information leakage. We use conditional mutual information leakage to measure the amount of information leakage from the trained machine learning model about the presence of an individual in the training dataset. We devise an upper bound for this measure of information leakage using Kullback--Leibler divergence that is more amenable to numerical computation. We prove a direct relationship between the Kullback--Leibler membership information leakage and the probability of success for a hypothesis-testing adversary examining whether a particular data record belongs to the training dataset of a machine learning model. We show that the mutual information leakage is a decreasing function of the training dataset size and the regularization weight. We also prove that, if the sensitivity of the machine learning model (defined in terms of the derivatives of the fitness with respect to model parameters) is high, more membership information is potentially leaked. This illustrates that complex models, such as deep neural networks, are more susceptible to membership inference attacks in comparison to simpler models with fewer degrees of freedom. We show that the amount of the membership information leakage is reduced by $\mathcal{O}(\log^{1/2}(δ^{-1})ε^{-1})$ when using Gaussian $(ε,δ)$-differentially-private additive noises.

cs.LG↗

arXiv2020

Regularization Helps with Mitigating Poisoning Attacks: Distributionally-Robust Machine Learning Using the Wasserstein Distance

We use distributionally-robust optimization for machine learning to mitigate the effect of data poisoning attacks. We provide performance guarantees for the trained model on the original data (not including the poison records) by training the model for the worst-case distribution on a neighbourhood around the empirical distribution (extracted from the training dataset corrupted by a poisoning attack) defined using the Wasserstein distance. We relax the distributionally-robust machine learning problem by finding an upper bound for the worst-case fitness based on the empirical sampled-averaged fitness and the Lipschitz-constant of the fitness function (on the data for given model parameters) as regularizer. For regression models, we prove that this regularizer is equal to the dual norm of the model parameters. We use the Wine Quality dataset, the Boston Housing Market dataset, and the Adult dataset for demonstrating the results of this paper.

cs.LG↗

arXiv2020

Temporally Discounted Differential Privacy for Evolving Datasets on an Infinite Horizon

We define discounted differential privacy, as an alternative to (conventional) differential privacy, to investigate privacy of evolving datasets, containing time series over an unbounded horizon. We use privacy loss as a measure of the amount of information leaked by the reports at a certain fixed time. We observe that privacy losses are weighted equally across time in the definition of differential privacy, and therefore the magnitude of privacy-preserving additive noise must grow without bound to ensure differential privacy over an infinite horizon. Motivated by the discounted utility theory within the economics literature, we use exponential and hyperbolic discounting of privacy losses across time to relax the definition of differential privacy under continual observations. This implies that privacy losses in distant past are less important than the current ones to an individual. We use discounted differential privacy to investigate privacy of evolving datasets using additive Laplace noise and show that the magnitude of the additive noise can remain bounded under discounted differential privacy. We illustrate the quality of privacy-preserving mechanisms satisfying discounted differential privacy on smart-meter measurement time-series of real households, made publicly available by Ausgrid (an Australian electricity distribution company).

cs.CR↗

arXiv2019

Privacy-Preserving Public Release of Datasets for Support Vector Machine Classification

We consider the problem of publicly releasing a dataset for support vector machine classification while not infringing on the privacy of data subjects (i.e., individuals whose private information is stored in the dataset). The dataset is systematically obfuscated using an additive noise for privacy protection. Motivated by the Cramer-Rao bound, inverse of the trace of the Fisher information matrix is used as a measure of the privacy. Conditions are established for ensuring that the classifier extracted from the original dataset and the obfuscated one are close to each other (capturing the utility). The optimal noise distribution is determined by maximizing a weighted sum of the measures of privacy and utility. The optimal privacy-preserving noise is proved to achieve local differential privacy. The results are generalized to a broader class of optimization-based supervised machine learning algorithms. Applicability of the methodology is demonstrated on multiple datasets.

cs.CR↗

arXiv2019

Noiseless Privacy

In this paper, we define noiseless privacy, as a non-stochastic rival to differential privacy, requiring that the outputs of a mechanism (i.e., function composition of a privacy-preserving mapping and a query) can attain only a few values while varying the data of an individual (the logarithm of the number of the distinct values is bounded by the privacy budget). Therefore, the output of the mechanism is not fully informative of the data of the individuals in the dataset. We prove several guarantees for noiselessly-private mechanisms. The information content of the output about the data of an individual, even if an adversary knows all the other entries of the private dataset, is bounded by the privacy budget. The zero-error capacity of memory-less channels using noiselessly private mechanisms for transmission is upper bounded by the privacy budget. The performance of a non-stochastic hypothesis-testing adversary is bounded again by the privacy budget. Finally, assuming that an adversary has access to a stochastic prior on the dataset, we prove that the estimation error of the adversary for individual entries of the dataset is lower bounded by a decreasing function of the privacy budget. In this case, we also show that the maximal information leakage is bounded by the privacy budget. In addition to privacy guarantees, we prove that noiselessly-private mechanisms admit composition theorem and post-processing does not weaken their privacy guarantees. We prove that quantization operators can ensure noiseless privacy if the number of quantization levels is appropriately selected based on the sensitivity of the query and the privacy budget. Finally, we illustrate the privacy merits of noiseless privacy using multiple datasets in energy and transport.

cs.IT↗

arXiv2019

A Fundamental Bound on Performance of Non-Intrusive Load Monitoring with Application to Smart Meter Privacy

We prove that the expected estimation error of non-intrusive load monitoring algorithms is lower bounded by the trace of the inverse of the cross-correlation matrix between the derivatives of the load profiles of the appliances. We use this fundamental bound to develop privacy-preserving policies. Particularly, we devise a load-scheduling policy by maximizing the lower bound on the expected estimation error of non-intrusive load monitoring algorithms.

eess.SP↗

arXiv2019

Differential Privacy for Evolving Almost-Periodic Datasets with Continual Linear Queries: Application to Energy Data Privacy

For evolving datasets with continual reports, the composition rule for differential privacy (DP) dictates that the scale of DP noise must grow linearly with the number of the queries, or that the privacy budget must be split equally between all the queries, so that the privacy budget across all the queries remains bounded and consistent with the privacy guarantees. To avoid this drawback of DP, we consider datasets containing almost periodic time series, composed of periodic components and noisy variations on top that are independent across periods. Our interest in these datasets is motivated by that, for reporting on private periodic time series, we do not need to divide the privacy budget across the entire, possibly infinite, horizon. Instead, for periodic time series, we generate DP reports for the first period and report the same DP reports periodically. In practice, however, exactly periodic time series do not exist as the data always contains small variations due to random or uncertain events. For instance, the energy consumption of a household may repeat the same daily pattern with slight variations due to minor changes to the habits of the individuals. The underlying periodic pattern is a function of the private information of the households. It might be desired to protect the privacy of households by not leaking information about the recurring patterns while the individual daily variations are almost noise-like with little to no privacy concerns (depending on the situation). Motivated by this, we define DP for almost periodic datasets and develop a Laplace mechanism for responding to linear queries. We provide statistical tools for testing the validity of almost periodicity assumption. We use multiple energy datasets containing smart-meter measurements of households to validate almost periodicity assumption. We generate DP aggregate reports and investigate their utility.

cs.CR↗

arXiv2019

Taking a Lesson from Quantum Particles for Statistical Data Privacy

Privacy is under threat from artificial intelligence revolution fueled by unprecedented abundance of data. Differential privacy, an established candidate for privacy protection, is susceptible to adversarial attacks, acts conservatively, and leads to miss-implementations because of lacking systematic methods for setting its parameters (known as the privacy budget). An alternative is information-theoretic privacy using entropy with the drawback of requiring prior distribution of the private data. Here, by using the Fisher information, information-theoretic privacy framework is extended to avoid unnecessary assumptions on the private data. The optimal privacy-preserving additive noise, extracted by minimizing the Fisher information, must follow the time-independent Schrodinger's equation. A fundamental trade-off between privacy and utility is also proved, reminiscent of the Heisenberg uncertainty principle.

cs.CR↗