Data imputation is an important data preparation task where the data analyst replaces missing or erroneous values to increase the expected accuracy of downstream analyses. The accuracy improvement of data imputation extends to private data analyses across distributed databases. However, existing data imputation methods violate the privacy of the data rendering the privacy protection in the downstream analyses obsolete. We conclude that private data analysis requires private data imputation. In this paper, we present the first optimized protocols for private data imputation. We consider the case of horizontally and vertically split data sets. Our optimization aims to reduce most of the computation to private set intersection (or at least oblivious programmable pseudo-random function) protocols which can be very efficiently computed. We show that private data imputation has -- on average across all evaluated datasets -- an accuracy advantage of 20\% in case of vertically split data and 5\% in case of horizontally split data over imputing data locally. In case of the worst data split we observed that imputing using our method resulted in an increase of up to 32.7 times in the quality of imputation over the vertically split data and 3.4 times in case of horizontally split data. Our protocols are very efficient and run in 2.4 seconds in case of vertically split data and 8.4 seconds in case of horizontally split data for 100,000 records evaluated in the 10 Gbps network setting, performing one data imputation.

Multi-party private set union (mPSU) allows multiple parties to compute the union of their private input sets without revealing any additional information. Existing efficient mPSU protocols can be categorized into symmetric key encryption (SKE)-based and public key encryption (PKE)-based approaches. However, neither type of mPSU protocol scales efficiently to a large number of parties, as they fail to fully utilize available computational resources, leaving participants idle during various stages of the protocol execution.

Secure multi-party computation has seen substantial performance improvements in recent years and is being increasingly used in commercial products. While a significant amount of work was dedicated to improving its efficiency under standard security models, the threat models do not account for information leakage from the output of secure function evaluation. Quantifying information disclosure about private inputs from observing the function outcome is the subject of this work. Motivated by the City of Boston gender pay gap studies, in this work, we focus on the computation of the average of salaries and quantify information disclosure about private inputs of one or more participants (the target) to an adversary via information-theoretic techniques. We study a number of distributions including log-normal, which is typically used for modeling salaries. We consequently evaluate information disclosure after repeated evaluation of the average function on overlapping inputs, as was done in the Boston gender pay study that ran multiple times, and provide recommendations for using the sum and average functions in secure computation applications. Our goal is to develop mechanisms that lower information disclosure about participants’ inputs to a desired level and provide guidelines for setting up real-world secure evaluation of this function.

Secure multi-party computation has seen substantial performance improvements in recent years and is being increasingly used in commercial products. While a significant amount of work was dedicated to improving its efficiency under standard security models, the threat models do not account for information leakage from the output of secure function evaluation. Quantifying information disclosure about private inputs from observing the function outcome is the subject of this work. Motivated by the City of Boston gender pay gap studies, in this work we focus on the computation of the average of salaries and quantify information disclosure about private inputs of one or more participants (the target) to an adversary via information-theoretic techniques. We study a number of distributions including log-normal, which is typically used for modeling salaries. We consequently evaluate information disclosure after repeated evaluation of the average function on overlapping inputs, as was done in the Boston gender pay study that ran multiple times, and provide recommendations for using the sum and average functions in secure computation applications. Our goal is to develop mechanisms that lower information disclosure about participants' inputs to a desired level and provide guidelines for setting up real-world secure evaluation of this function.

Biometric authentication eliminates the need for users to remember secrets and serves as a convenient mechanism for user authentication. Traditional implementations of biometric-based authentication store sensitive user biometry on the server and the server becomes an attractive target of attack and a source of large-scale unintended disclosure of biometric data. To mitigate the problem, we can resort to privacy-preserving computation and store only protected biometrics on the server. While a variety of secure computation techniques is available, our analysis of privacy-preserving biometric authentication constructions revealed that available solutions fall short of addressing the challenges of privacy-preserving biometric authentication. Thus, in this work we put forward new constructions to address the challenges.

This work studies compilation of honest-majority semi-honest secure multi-party protocols secure up to additive attacks to maliciously secure computation with abort. Prior work concentrated on arithmetic circuits composed of addition and multiplication gates, while many practical protocols rely on additional types of elementary operations or gates to achieve good performance. In this work we revisit the notion of security up to additive attacks in the presence of additional gates such as random element generation and opening. This requires re-evaluation of functions that can be securely evaluated, extending the notion of protocols secure up to additive attacks, and re-visiting the notion of delayed verification that points to weaknesses in its prior use and designing a mitigation strategy. We transform the computation using dual execution to achieve security in the malicious model with abort and experimentally evaluate the difference in performance of semi-honest and malicious protocols to demonstrate the low cost.

Secure multi-party computation has seen significant performance advances and increasing use in recent years. Techniques based on secret sharing offer attractive performance and are a popular choice for privacy-preserving machine learning applications. Traditional techniques operate over a field, while designing equivalent techniques for a ring Z_2^k can boost performance. In this work, we develop a suite of multi-party protocols for a ring in the honest majority setting starting from elementary operations to more complex with the goal of supporting general-purpose computation. We demonstrate that our techniques are substantially faster than their field-based equivalents when instantiated with a different number of parties and perform on par with or better than state-of-the-art techniques with designs customized for a fixed number of parties. We evaluate our techniques on machine learning applications and show that they offer attractive performance.

Motivated by the importance of floating-point computations, we study the problem of securely and accurately summing many floating-point numbers. Prior work has focused on security absent accuracy or accuracy absent security, whereas our approach achieves both of them. Specifically, we show how to implement floating-point superaccumulators using secure multi-party computation techniques, so that a number of participants holding secret shares of floating-point numbers can accurately compute their sum while keeping the individual values private.

Secure multi-party computation (SMC) is an emerging topic which has been drawing growing attention during recent decades. There are many examples which show importance of SMC constructions in practice, such as privacy-preserving decision making and machine learning, auctions, private set intersection, and others. In this tutorial, we provide a comprehensive coverage of SMC techniques, starting from precise definitions and fundamental techniques. Consequently, a significant portion of the tutorial focuses on recent advances in general SMC constructions. We cover garbled circuit evaluation (GCE) and linear secret sharing (LSS) which are commonly used for secure two-party and multi-party computation, respectively. The coverage includes both standard adversarial models: semi-honest and malicious. For GCE, we start with the original Yao's garbled circuits construction [30] for semi-honest adversaries and consequently cover its recent optimizations such as the "free XOR,'' the garbled row reduction, the half-gates optimization, and the use of AES NI techniques. We follow with a discussion of techniques for making GCE resilient to malicious behavior, which includes the cut-and-choose approach and additional techniques to deter known attacks in the presence of malicious participants. In addition, we include the-state-of-the-art protocols for oblivious transfer (OT) and OT extension in the presence of semi-honest and malicious users. For LSS, we start from standard solutions for the semi-honest adversarial model including [5, 28] and consequently move to recent efficient constructions for semi-honest and malicious adversarial models. The coverage includes different types of corruption thresholds (with and without honest majority), which imply different guarantees with respect to abort.

Recent compilers allow a general-purpose program (written in a conventional programming language) that handles private data to be translated into a secure distributed implementation of the corresponding functionality. The resulting program is then guaranteed to provably protect private data using secure multi-party computation techniques. The goals of such compilers are generality, usability, and efficiency, but the complete set of features of a modern programming language has not been supported to date by the existing compilers. In particular, recent compilers PICCO and the two-party ANSI C compiler strive to translate any C program into its secure multi-party implementation, but they currently lack support for pointers and dynamic memory allocation, which are important components of many C programs. In this work, we mitigate the limitation and add support for pointers to private data and consequently dynamic memory allocation to the PICCO compiler, enabling it to handle a more diverse set of programs over private data. Because doing so opens up a new design space, we investigate the use of pointers to private data (with known as well as private locations stored in them) in programs and report our findings. Aside from dynamic memory allocation, we examine other important topics associated with common pointer use such as reference by pointer/address, casting, and building various data structures in the context of secure multi-party computation. This results in enabling the compiler to automatically translate a user program that uses pointers to private data into its distributed implementation that provably protects private data throughout the computation. We empirically evaluate the constructions and report on the performance of representative programs.

Abstract Computation based on genomic data is becoming increasingly popular today, be it for medical or other purposes. Non-medical uses of genomic data in a computation often take place in a server-mediated setting where the server offers the ability for joint genomic testing between the users. Undeniably, genomic data is highly sensitive, which in contrast to other biometry types, discloses a plethora of information not only about the data owner, but also about his or her relatives. Thus, there is an urgent need to protect genomic data. This is particularly true when the data is used in computation for what we call recreational non-health-related purposes. Towards this goal, in this work we put forward a framework for server-aided secure two-party computation with the security model motivated by genomic applications. One particular security setting that we treat in this work provides stronger security guarantees with respect to malicious users than the traditional malicious model. In particular, we incorporate certified inputs into secure computation based on garbled circuit evaluation to guarantee that a malicious user is unable to modify her inputs in order to learn unauthorized information about the other user’s data. Our solutions are general in the sense that they can be used to securely evaluate arbitrary functions and offer attractive performance compared to the state of the art. We apply the general constructions to three specific types of genomic tests: paternity, genetic compatibility, and ancestry testing and implement the constructions. The results show that all such private tests can be executed within a matter of seconds or less despite the large size of one’s genomic data.

The emergence and wide availability of remote storage service providers prompted work in the security community that allows clients to verify integrity and availability of the data that they outsourced to a not fully trusted remote storage server at a relatively low cost. Most recent solutions to this problem allow clients to read and update (i.e., insert, modify, or delete) stored data blocks while trying to lower the overhead associated with verifying the integrity of the stored data. In this work, we develop a novel scheme, performance of which favorably compares with the existing solutions. Our solution additionally enjoys a number of new features, such as a natural support for operations on ranges of blocks, revision control, and support for multiple user access to shared content. The performance guarantees that we achieve stem from a novel data structure called a balanced update tree and removing the need for interaction during update operations in addition to communicating the updates themselves.

The increasing availability and use of biometric data leads to situations when sensitive biometric data is to be handled by entities who may not be fully trusted or otherwise are not authorized to have full access to such data. This calls for mechanisms of provably protecting biometric data while still allowing the computation to take place. Our focus is on privacy-preserving matching of two fingerprints (authentication or identification purposes) using traditional minutia-based representation of fingerprints that leads to the most discriminative fingerprint comparisons. Unlike previous work in the security literature, we would like to focus on algorithms that are guaranteed to find the maximum number of minutiae that can be paired together between two fingerprints leading to more accurate comparisons. To address this problem, we formulate it as a flow network problem and reduce it to finding maximum matching size in bipartite graphs. The resulting problem is in turn reduced to computing the rank of a (non-invertible) matrix, formed as a randomized adjacency matrix of the bipartite graph. We then provide data-oblivious algorithms for matrix rank computation and consecutively finding maximum matching size in a bipartite graph and also extend the algorithms to solve the problem of accurate fingerprint matching. These algorithms lead to their secure counterparts using standard secure two-party or multi-party techniques. Lastly, we implement secure fingerprint matching in the secure two-party computation setting using garbled circuit evaluation. Our experimental results demonstrate that the techniques are efficient, leading to performance similar to that of other fastest secure fingerprint matching techniques, despite higher complexity of our solution that higher accuracy demands. These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

This work treats the problem of designing data-oblivious algorithms for classical and widely used graph problems. A data-oblivious algorithm is defined as having the same sequence of operations regardless of the input data and data-independent memory accesses. Such algorithms are suitable for secure processing in outsourced and similar environments, which serves as the main motivation for this work. We provide data-oblivious algorithms for breadth-first search, single-source single-destination shortest path, minimum spanning tree, and maximum flow, the asymptotic complexities of which are optimal, or close to optimal, for dense graphs.

The emergence and availability of remote storage providers prompted work in the security community that allows a client to verify integrity and availability of the data she outsourced to an untrusted remove storage server at a relatively low cost. Most recent solutions to this problem allow the client to read and update (insert, modify, or delete) stored data blocks while trying to lower the overhead associated with verifying data integrity. In this work we develop a novel and efficient scheme, computation and communication overhead of which is orders of magnitude lower than those of other state-of-the-art schemes. Our solution has a number of new features such as a natural support for operations on ranges of blocks, and revision control. The performance guarantees that we achieve stem from a novel data structure, termed balanced update tree, and removing the need to verify update operations.

Secure computation on private data has been an active area of research for many years and has received a renewed interest with the emergence of cloud computing. In recent years, substantial progress has been made with respect to the efficiency of the available techniques and several implementations have appeared. The available tools, however, lacked a convenient mechanism for implementing a general-purpose}program in a secure computation framework suitable for execution in not fully trusted environments. This work fulfills this gap and describes a system, called PICCO, for converting a program written in an extension of C into its distributed secure implementation and running it in a distributed environment. The C extension preserves all current features of the programming language and allows variables to be marked as private and be used in general-purpose computation. Secure distributed implementation of compiled programs is based on linear secret sharing, achieving efficiency and information-theoretical security. Our experiments also indicate that many programs can be evaluated very efficiently on private data using PICCO.

Secure sketches and fuzzy extractors enable the use of biometric data in cryptographic applications by correcting errors in noisy biometric readings and producing cryptographic materials suitable for authentication, encryption, and other purposes. Such constructions work by producing a public sketch, which is later used to reproduce the original biometric and all derived information exactly from a noisy biometric reading. It has been previously shown that release of multiple sketches associated with a single biometric presents security problems for certain constructions. We continue the analysis to demonstrate that all other constructions are also prone to similar problems and cannot be safely reused even in the presence of very weak adversaries.

Cloud computing services are becoming more prevalent and readily available today, bringing to us economies of scale and making large-scale computation feasible. Security and privacy considerations, however, stand in the way of fully utilizing the benefits of such services and architectures. In this work we address the problem of secure outsourcing of large-scale biometric experiments to a cloud or grid in a way that the client can verify that with very high probability the task was computed correctly. We conduct thorough theoretical analysis of the proposed techniques and provide implementation results that indicate that our solution imposes modest overhead.

One of the concerns patients have when confronted with a medical condition is which physician to trust. Any recommendation system that seeks to answer this question must ensure that any sensitive medical information collected by the system is properly secured. In this article, we codify these privacy concerns in a privacy-friendly framework and present two architectures that realize it: the Secure Processing Architecture (SPA) and the Anonymous Contributions Architecture (ACA). In SPA, patients submit their ratings in a protected form without revealing any information about their data and the computation of recommendations proceeds over the protected data using secure multiparty computation techniques. In ACA, patients submit their ratings in the clear, but no link between a submission and patient data can be made. We discuss various aspects of both architectures, including techniques for ensuring reliability of computed recommendations and system performance, and provide their comparison.

Privacy-preserving set operations and set intersection in particular are a popular research topic. Despite a large body of literature, the great majority of the available solutions are two-party protocols and are not composable. In this work we design a comprehensive suite of secure multi-party protocols for set and multiset operations that are composable, do not assume any knowledge of the sets by the parties carrying out the secure computation, and can be used for secure outsourcing. All of our protocols have communication and computation complexity of O(m log m) for sets or multisets of size m, which compares favorably with prior work. Furthermore, we are not aware of any results that realize composable operations. Our protocols are secure in the information theoretic sense and are designed to minimize the round complexity.

Today biometric data propagate more heavily into our lives. With more ubiquitous use of such data, computations over biometrics become more prevalent as well. While it is well understood that privacy of biometric data must be protected, often computations over biometric data involve untrusted participants or servers, let it be a cross check between different agencies who are not permitted to share the data or a researcher testing a new biometric matching algorithm on a large scale that forces the computation to be placed on a grid. Unarguably, it would be desirable to secure computation over sensitive biometric data in such environments. Currently, no secure techniques for outsourcing biometric comparisons or searching are readily available, and this work makes the first step at designing solutions for secure outsourcing iris identification to one or more untrusted servers. We develop new solutions for the single-server (i.e., non-interactive) and multiple-server settings that use significantly different techniques. Furthermore, we carry out extensive experimentation on a database of iris codes to both validate the findings and achieve efficiency improvements.

Hierarchies arise in the context of access control whenever the user population can be modeled as a set of partially ordered classes (represented as a directed graph). A user with access privileges for a class obtains access to objects stored at that class and all descendant classes in the hierarchy. The problem of key management for such hierarchies then consists of assigning a key to each class in the hierarchy so that keys for descendant classes can be obtained via efficient key derivation. We propose a solution to this problem with the following properties: (1) the space complexity of the public information is the same as that of storing the hierarchy; (2) the private information at a class consists of a single key associated with that class; (3) updates (i.e., revocations and additions) are handled locally in the hierarchy; (4) the scheme is provably secure against collusion; and (5) each node can derive the key of any of its descendant with a number of symmetric-key operations bounded by the length of the path between the nodes. Whereas many previous schemes had some of these properties, ours is the first that satisfies all of them. The security of our scheme is based on pseudorandom functions, without reliance on the Random Oracle Model. Another substantial contribution of this work is that we are able to lower the key derivation time at the expense of modestly increasing the public storage associated with the hierarchy. Insertion of additional, so-called shortcut, edges, allows to lower the key derivation to a small constant number of steps for graphs that are total orders and trees by increasing the total number of edges by a small asymptotic factor such as O (log * n ) for an n -node hierarchy. For more general access hierarchies of dimension d , we use a technique that consists of adding dummy nodes and dimension reduction. The key derivation work for such graphs is then linear in d and the increase in the number of edges is by the factor O (log d − 1 n ) compared to the one-dimensional case. Finally, by making simple modifications to our scheme, we show how to handle extensions proposed by Crampton [2003] of the standard hierarchies to “limited depth” and reverse inheritance.

Combinatorial group testing, given a set C of individuals ("customers"), consists of applying group tests on subsets of C for the purpose of identifying which members of C are infected (or, more generally, defective in some way). The outcome of a group test reveals only the presence or absence of infection(s) in that group, but a number of group tests exactly identifies all infected members.

Online privacy is an increasingly important problem, as many services are now offered in a digital form. Privacy (or the lack thereof) is of a special concern in subscriptions to large data repositories with heterogeneous information, where the service provider can easily profile its users and sell that information to third parties. In this work we present the design and implementation of a system that closely resembles the current practice of subscriptions to many services such as newspapers, digital libraries, music collections, etc., but at the same time offers anonymous access to the service. As with current practice, in our solution a user subscribes to the service obtaining access to it for a certain period of time, at the end of which the subscription expires.

The problem of key management for access control systems has been well-studied, and the literature contains several schemes for hierarchy-based and temporal-based access control. The problem of key management in such systems is how to assign keys to users such that each user is able to compute and have access to the appropriate resources while minimizing computation and storage requirements. In the current paper, we consider key management schemes for geo-spatial access control. That is, the access control policy assigns to a user a specific geographic area, and the user consequently obtains access to her area or information about it.In this work, the geography is modeled as an m × n grid of cells (let m ≥ n). Each cell has its own key associated with it, and a user who wants to access the content of a cell needs to obtain its key. Each user obtains access to a rectangular area (or a finite collection of such rectangles) and is able compute keys corresponding to the cells that comprise her area.Our main result is an efficient scheme with the following properties: (i) each user obtains a small constant number of secret keys that permit access to an arbitrary rectangular sub-grid, (ii) computation to derive the key of a specific cell in that rectangle consists of a constant number of efficient operations, and (iii) the server needs to maintain O(mn(log log m)2 log* m) public information accessible to all users. The public storage requirement is the worst-case bound and can be improved if the grid is partitioned into regions where the cells of a region share the same key.

Access hierarchies are useful in many applications and are modeled as a set of access classes organized by a partial order. A user who obtains access to a class in such a hierarchy is entitled to access objects stored at that class, as well as objects stored at its descendant classes. Efficient schemes for this framework assign only one key to a class and use key derivation to permit access to descendant classes. Ideally, the key derivation uses simple primitives such as cryptographic hash computations and modular additions. A straightforward key derivation time is then linear in the length of the path between the user's class and the class of the object that the user wants to access. Recently, work presented in [2] has given an efficient solution that significantly lowers this key derivation time, while using only hash functions and modular additions. Two fastkey-derivation techniques in that paper were given for trees, achieving O(log log n) and O(1) key derivation times, respectively, where n is the number of access classes. The present paper presents efficient key derivation techniques for hierarchies that are not trees, using a scheme that is very different from the above-mentioned paper. The construction we give in the present paper is recursive and uses the onedimensional case solution as its base. It makes a novel use of the notion of the dimension d of an access graph, and provides a solution through which no key derivation requires more than 2d+1 hash function computations, even for "unbalanced" hierarchies whose depth is linear in their number of access classes n. The significance of this result is strengthened by the fact that many access graphs have a low d value (e.g., trees correspond to the case d = 2). Our scheme has the desirable property (as did [2] for trees) that addition and deletion of edges and nodes in the access hierarchy can be "contained".

The problem of key management in an access hierarchy has elicited much interest in the literature. The hierarchy is modeled as a set of partially ordered classes (represented as a directed graph), and a user who obtains access (i.e., a key) to a certain class can also obtain access to all descendant classes of her class through key derivation. Our solution to the above problem has the following properties: (i) only hash functions are used for a node to derive a descendant's key from its own key; (ii) the space complexity of the public information is the same as that of storing the hierarchy; (iii) the private information at a class consists of a single key associated with that class; (iv) updates (revocations, additions, etc.) are handled locally in the hierarchy; (v) the scheme is provably secure against collusion; and (vi) key derivation by a node of its descendant's key is bounded by the number of bit operations linear in the length of the path between the nodes. Whereas many previous schemes had some of these properties, ours is the first that satisfies all of them. Moreover, for trees (and other recursively decomposable hierarchies), we are the first to achieve a worst- and average-case number of bit operations for key derivation that is exponentially better than the of a balanced hierarchy (double-exponentially better if the hierarchy is unbalanced, i.e., tall and skinny); this is achieved with only a constant increase in the space for the hierarchy. We also show how with simple modifications our scheme can handle extensions proposed by Crampton of the standard hierarchies to limited depth and reverse inheritance [13]. The security of our scheme relies only on the use of pseudo-random functions.

In this work we address the problem of portable and flexible privacy-preserving access rights for large online data repositories. Privacy-preserving access control means that the service provider can neither learn what access rights a customer has nor link a request to access an item to a particular customer, thus maintaining privacy of both customer activity and customer access rights. Flexible access rights allow any customer to choose any subset of items from the repository and correspondingly be charged only for the items selected. And portability of access rights means that the rights themselves can be stored on small devices of limited storage space and computational capabilities, and therefore the rights must be enforced using the limited resources available.Our main results are solutions to the problem that utilize minimal perfect hash functions and order-preserving minimal perfect hash functions. None of them use expensive cryptography, all require very little space, and they are therefore suitable for computationally weak and space-limited devices such as smartcards, sensors, etc. Performance of the schemes is measured as the probability of false positives (i.e., the probability that access to an unpurchased item will be permitted) for a given storage space bound. Using our techniques, for a data repository of size n and subscription order of m ll n items, we achieve a probability of false positives of m-c using only O(cm) bits of storage space, where c is an adjustable parameter (a constant or otherwise) that can be set to provide the desired performance. This is the first time that such provable bounds are established for this problem, and we believe the techniques we use are of more general interest through the unusual use we make of perfect hashing.

When customers need to each be given portable access rights to a subset of documents from a large universe of n available documents, it is often the case that the space available for representing each customer's access rights is limited to much less than n, say it is no more than m bits. This is the case when, e.g., limited-capacity inexpensive cards are used to store the access rights to huge multimedia document databases. How does one represent subsets of a huge set of n elements, when only m bits are available and m is much smaller than n? We use an approach reminiscent of Bloom filters, by assigning to each document a subset of the m bits: If that document is in a customer's subset then we set the corresponding bits to 1 on the customer's card. This guarantees that each customer gets the documents he paid for, but it also gives him access to documents he did not pay for ("false positives"). We want to do so in a manner that minimizes the expected total false positives under various deterministic and probabilistic models: In the former model we assume k customers whose respective subsets are known a priori, whereas in the latter we assume (more realistically) that each document has a probability of being included in a customer's subset. We cannot use randomly assigned bits for each document (in the way Bloom filters do), rather we need to consider the a priori knowledge (deterministic or probabilistic) we are given in each model in order to better assign a subset of the m available bits to each of the n documents. We analyze and give e#cient schemes for this problem.

In this work, we collect and analyze all of the IP and TCP headers of packets seen on a network that either violate existing standards or should not appear in modern internets. Our goal is to determine the reason that these packets appear on the network and evaluate what proportion of such packets could cause actual damage. Thus, we examine and divide the unusual packets obtained during our experiments into several categories based on their type and possible cause and show the results.