We describe initial research results investigating the impact of a new Computing Fellows program. The program provides computing-related peer mentoring and teaching in undergraduate courses across disciplines. Results suggest that the program positively contributes to both fellows and students engagement with computing.

We consider privacy-preserving learning in the context of online learning. Insettings where data instances arrive sequentially in streaming fashion, incremental trainingalgorithms such as stochastic gradient descent (SGD) can be used to learn and updateprediction models. When labels are costly to acquire, active learning methods can beused to select samples to be labeled from a stream of unlabeled data. These labeled datasamples are then used to update the machine learning models. Privacy-preserving onlinelearning can be used to update predictors on data streams containing sensitive information.The differential privacy framework quantifies the privacy risk in such settings. This workproposes a differentially private online active learning algorithm using stochastic gradientdescent (SGD) to retrain the classifiers. We propose two methods for selecting informativesamples. We incorporated this into a general-purpose web application that allows a non-expert user to evaluate the privacy-aware classifier and visualize key privacy-utility tradeoffs.Our application supports linear support vector machines and logistic regression and enablesan analyst to configure and visualize the effect of using differentially private online activelearning versus a non-private counterpart. The application is useful for comparing theprivacy/utility tradeoff of different algorithms, which can be useful to decision makers inchoosing which algorithms and parameters to use. Additionally, we use the application toevaluate our SGD-based solution and to show that it generates predictions with a superiorprivacy-utility tradeoff than earlier methods.

Modern day security technologies, such as passwords, authentication protocols, firewalls, and access-control Mechanisms, are preventive in nature. That is, they stop unauthorized parties before they are able to access data, information and services or violate system policies. However, the dramatically increased scale and complexity of Internet commerce, social networking, remote work, distance learning, and myriad other forms of social, economic, and intellectual engagement online with both strangers and friends has increased the awareness that these preventive mechanisms are inadequate in certain circumstances. The result is a growing interest in accountability mechanisms to complement preventive measures. In this survey of the concept of accountability in information systems, the authors focus on systems in which policy violations are punished; that is, the actors are held accountable for their actions. As there is no accepted definition of the precise meaning of accountability in terms of such systems, the authors provide the reader with a broad overview of the concept. In doing so, they introduce the topic and place it in context of the social and systematic factors which help to define the term. They proceed to discuss in depth the mechanisms and domains across numerous disciplines before describing the available tools and proofs for implementing accountability in systems. Finally, they summarize the ideas, key published papers and ideas for future work. This monograph provides the reader with a thorough overview of the concept of accountability in modern day computing systems. In enables the reader to quickly understand the concept and the progress that has been to date in implementing the tools for the next generation of online security systems.

In this paper, we study differential privacy in noisy search. This problem is connected to noisy group testing: the goal is to find a defective or anomalous item within a group using only aggregate group queries, not individual queries. Differentially private noisy group testing has the potential to be used for anomaly detection in a way that provides differential privacy to the non-anomalous individuals while still helping to allow the anomalous individuals to be located. To do this, we introduce the notion of anomaly-restricted differential privacy. We then show that noisy group testing can be used to satisfy anomaly-restricted differential privacy while still narrowing down the location of the anomalous samples, and evaluate our approach experimentally.

We discuss the widely increasing range of applications of a cryptographic technique called multi-party computation. For many decades, this was perceived to be of purely theoretical interest, but now it has started to find application in a number of use cases. We highlight in this paper a number of these, ranging from securing small high-value items such as cryptographic keys, through to securing an entire database.

Abstract Machine learning algorithms based on deep Neural Networks (NN) have achieved remarkable results and are being extensively used in different domains. On the other hand, with increasing growth of cloud services, several Machine Learning as a Service (MLaaS) are offered where training and deploying machine learning models are performed on cloud providers’ infrastructure. However, machine learning algorithms require access to the raw data which is often privacy sensitive and can create potential security and privacy risks. To address this issue, we present CryptoDL, a framework that develops new techniques to provide solutions for applying deep neural network algorithms to encrypted data. In this paper, we provide the theoretical foundation for implementing deep neural network algorithms in encrypted domain and develop techniques to adopt neural networks within practical limitations of current homomorphic encryption schemes. We show that it is feasible and practical to train neural networks using encrypted data and to make encrypted predictions, and also return the predictions in an encrypted form. We demonstrate applicability of the proposed CryptoDL using a large number of datasets and evaluate its performance. The empirical results show that it provides accurate privacy-preserving training and classification.

We use ideas from distributed computing and game theory to study dynamic and decentralized environments in which computational nodes, or decision makers, interact strategically and with limited information. In such environments, which arise in many real-world settings, the participants act as both economic and computational entities. We exhibit a general non-convergence result for a broad class of dynamics in asynchronous settings. We consider implications of our result across a wide variety of interesting and timely applications: game dynamics, circuit design, social networks, Internet routing, and congestion control. We also study the computational and communication complexity of testing the convergence of asynchronous dynamics. Our work opens a new avenue for research at the intersection of distributed computing and game theory.

We consider the problem of privacy-sensitive anomaly detection - screening to detect individuals, behaviors, areas, or data samples of high interest. What defines an anomaly is context-specific; for example, a spoofed rather than genuine user attempting to log in to a web site, a fraudulent credit card transaction, or a suspicious traveler in an airport. The unifying assumption is that the number of anomalous points is quite small with respect to the population, so that deep screening of all individual data points would potentially be time-intensive, costly, and unnecessarily invasive of privacy. Such privacy violations can raise concerns due sensitive nature of data being used, raise fears about violations of data use agreements, and make people uncomfortable with anomaly detection methods. Anomaly detection is well studied, but methods to provide anomaly detection along with privacy are less well studied. Our overall goal in this research is to provide a framework for identifying anomalous data while guaranteeing quantifiable privacy in a rigorous sense. Once identified, such anomalies could warrant further data collection and investigation, depending on the context and relevant policies. In this research, we focus on privacy protection during the deployment of anomaly detection. Our main contribution is a differentially private access mechanism for finding anomalies using a search algorithm based on adaptive noisy group testing. To achieve this, we take as our starting point the notion of group testing [1], which was most famously used to screen US military draftees for syphilis during World War II. In group testing, individuals are tested in groups to limit the number of tests. Using multiple rounds of screenings, a small number of positive individuals can be detected very efficiently. Group testing has the added benefit of providing privacy to individuals through plausible deniability - since the group tests use aggregate data, individual contributions to the test are masked by the group. We follow on these concepts by demonstrating a search model utilizing adaptive queries on aggregated group data. Our work takes the first steps toward strengthening and formalizing these privacy concepts by achieving differential privacy [2]. Differential privacy is a statistical measure of disclosure risk that captures the intuition that an individual's privacy is protected if the results of a computation have at most a very small and quantifiable dependence on that individual's data. In the last decade, there hpractical adoption underway by high-profile companies such as Apple, Google, and Uber. In order to make differential privacy meaningful in the context of a task that seeks to specifically identify some (anomalous) individuals, we introduce the notion of anomaly-restricted differential privacy. Using ideas from information theory, we show that noise can be added to group query results in a way that provides differential privacy for non-anomalous individuals and still enables efficient and accurate detection of the anomalous individuals. Our method ensures that using differentially private aggregation of groups of points, providing privacy to individuals within the group while refining the group selection to the point that we can probabilistically narrow attention to a small numbers of individuals or samples for further attention. To summarize: We introduce a new notion of anomaly-restriction differential privacy, which may be of independent interest. We provide a noisy group-based search algorithm that satisfies the anomaly-restricted differential privacy definition. We provide both theoretical and empirical analysis of our noisy search algorithm, showing that it performs well in some cases, and exhibits the usual privacy/accuracy tradeoff of differentially private mechanisms. Potential anomaly detection applications for our work might include spatial search for outliers: this would rely on new sensing technologies that can perform queries in aggregate to reveal and isolate anomalous outliers. For example, this could lead to privacy-sensitive methods for searching for outlying cell phone activity patterns or Internet activity patterns in a geographic location.

In settings where data instances are generated sequentially or in streaming fashion, online learning algorithms can learn predictors using incremental training algorithms such as stochastic gradient descent. In some security applications such as training anomaly detectors, the data streams may consist of private information or transactions and the output of the learning algorithms may reveal information about the training data. Differential privacy is a framework for quantifying the privacy risk in such settings. This paper proposes two differentially private strategies to mitigate privacy risk when training a classifier for anomaly detection in an online setting. The first is to use a randomized active learning heuristic to screen out uninformative data points in the stream. The second is to use mini-batching to improve classifier performance. Experimental results show how these two strategies can trade off privacy, label complexity, and generalization performance.

We propose a framework for organizing and classifying research results in the active field of secure multiparty computation (MPC). Our systematization of secure computation consists of (1) a set of definitions circumscribing the MPC protocols to be considered; (2) a set of quantitative axes for classifying and comparing MPC protocols; and (3) a knowledge base of propositions specifying the known relations between axis values. We have classified a large number of MPC protocols on these axes and developed an interactive tool for exploring the problem space of secure computation. We also give examples of how this systematization can be put to use to foster new research and the adoption of MPC for real-world problems.

We consider a scenario for data outsourcing that supports performing database queries in the following three-party model: a client interested in making database queries, a data owner providing its database for client access, and a server (e.g., a cloud server) holding the (encrypted) outsourced data and helping both other parties. In this scenario, a natural problem is that of designing efficient and privacy-preserving protocols for checking compliance of a client’s queries to the data owner’s query compliance policy. We propose a cryptographic model for the study of such protocols, defined so that they can compose with an underlying database retrieval protocol (with no query compliance policy) in the same participant model. Our main result is a set of new protocols that satisfy a combination of natural correctness, privacy, and efficiency requirements. Technical contributions of independent interest include the use of equality-preserving encryption to produce highly practical symmetric-cryptography protocols (i.e., two orders of magnitude faster than “Yao-like” protocols), and the use of a query rewriting technique that maintains privacy of the compliance result.

Over the course of the last decade, Facebook has become an incredibly popular social networking site, reporting around a billion visitors monthly. Like any social networking site, Facebook’s design decisions have implications about what is shared, what is not shared, and how much control users have about such sharing. In this paper, we report a systematic analysis of side channels in Facebook—that is, channels that can reveal privacy-sensitive information to users through indirect mechanisms. While these side channels may not be particularly surprising to the security research community, they still represent potential threats to Facebook users, depending on user expectations and attitudes. We surveyed Facebook users to determine user expectations and attitudes, including whether users are aware of these channels, and whether there are privacy objections to the channels to those aware. We find that many users are unaware of the side channels and express surprise at finding out about the side channels. Among users who are aware of them, some users express concerns while others do not. Based on these results, we identify design implications for social network sites that wish to provide users with more control over such choices.

Models of human mobility have broad applicability in urban planning, ecology, epidemiology, and other fields. Starting with Call Detail Records (CDRs) from a cellular telephone network that have gone through a straightforward anonymization procedure, the prior WHERE modeling approach produces synthetic CDRs for a synthetic population. The accuracy of WHERE has been validated against billions of location samples for hundreds of thousands of cell phones in the New York and Los Angeles metropolitan areas. In this paper, we introduce DP-WHERE, which modifies WHERE by adding controlled noise to achieve differential privacy, a strict definition of privacy that makes no assumptions about the power or background knowledge of a potential adversary. We also present experiments showing that the accuracy of DP-WHERE remains close to that of WHERE and of real CDRs. With this work, we aim to enable the creation and possible release of synthetic models that capture the mobility patterns of real metropolitan populations while preserving privacy.

We consider the problem of modeling and designing publish/subscribe protocols that safeguard the privacy of clients’ subscriptions and of servers’ publications while guaranteeing efficient latency in challenging scenarios (i.e., real-time publication, high data arrival rate, etc.). As general solutions from the theory of secure function evaluation protocols would not achieve satisfactory performance in these scenarios, we enrich the model with a third party (e.g., a cloud server). Our main result is a three-party publish/subscribe protocol suitable for practical applications in such scenarios because the publication phase uses only symmetric cryptography operations (a result believed not possible without the third party). At the cost of only a very small amount of privacy loss to the third party, and with no privacy loss to the publishing server or the clients, our protocol has very small publication latency, which we measured for large parameter ranges to be just a small constant factor worse than a publish/subscribe protocol guaranteeing no privacy.

Facilitating use of sensitive data for research or commercial purposes, in a manner that preserves the privacy of participating entities, is an active area of study. Differential privacy is a popular, relatively recent, framework that formalizes data privacy. In this dissertation, I examine the often conflicting goals of privacy and utility within the framework of differential privacy. The contributions of this dissertation fall into two main categories: 1) We propose differentially private algorithms for several tasks that could potentially involve sensitive data, such as synthetic graph modeling, human mobility modeling using cellular phone data, regression, and computing statistics on online data. We study the tradeoff between privacy and utility for these analyses—theoretically in some cases, and experimentally in others. We show that for each of these tasks, both privacy and utility can be successfully achieved by considering a meaningful tradeoff between the two. 2) We also examine connections between information theory and differential privacy, demonstrating how differential privacy arises out of a tradeoff between information leakage and utility. We show that differentially private mechanisms arise out of minimizing the information leakage (measured using mutual information) under the constraint of achieving a given level of utility. Further, we establish a connection between differentially private learning and PAC-Bayesian bounds.

We consider the problem of modeling and designing efficient and privacy-preserving publish/subscribe protocols in a distributed model where parties can act as publishers or subscribers or both, and there are no brokers or other types of parties. The problem is particularly challenging as privacy demands on such protocols come with efficiency limitations; most notably, the publisher must send messages as long as the publications to all parties, and the cryptographic techniques to perform the publish/subscribe match need to be based on asymmetric cryptographic operation which are known to be less efficient than their symmetric counterpart. Our main result is a distributed publish/subscribe protocol which addresses and essentially nullifies the impact of both efficiency limitations, without sacrificing the required privacy properties. Our construction is based on very efficient design of a novel cryptographic tool, of independent interest, called 'hybrid conditional oblivious transfer protocol', as it resembles hybrid encryption, where asymmetric encryption is only used to transfer a short key, which enables (much more efficient) symmetric encryption of a long message.

We present a framework for the design and analysis of probing methods to monitor network performance, an important technique for collecting measurements in tasks such as fault detection. We use this framework to study the interaction among numerous, possibly conflicting, optimization goals in the design of a probing algorithm. We present a rigorous definition of a probing-algorithm design problem that can apply broadly to network-measurement scenarios. We also present several metrics relevant to the analysis of probing algorithms, including probing frequency and network coverage, communication and computational overhead, and the amount of algorithm state required. We show inherent tradeoffs among optimization goals and give hardness results for achieving some combinations of optimization goals. We also consider the possibility of developing approximation algorithms for achieving some of the goals and describe a randomized approach as an alternative, evaluating it using our framework. Our work aids future development of low-overhead probing techniques and introduces principles from IP-based networking to theoretically grounded approaches for concurrent path-selection problems.