This paper introduces DISPERSE, a distributed scalable architecture for delivery of content and services that provides resilience against node failure through location-independent storage and replication of content. Current content delivery networks (CDNs) have, at least to some degree, a centralized structure thus susceptible to a single point of failure. DISPERSE addresses this limitation by implementing a fully de-centralized structure. DISPERSE is a two-layer architecture: the first layer (front-end layer) exposes services (e.g., Web, SFTP) to clients; the second layer (back-end layer) provides reliable distributed storage of content and application state. Content in DISPERSE's back-end layer is stored and exchanged as Named Data Network (NDN) content objects. This allows DISPERSE to implement fine-grained, location-independent, fully decentralized content replication mechanisms. We validate the performance of DISPERSE under two node failure scenarios. In the first scenario, content can be stored in any DISPERSE node, and all nodes are equally likely to fail. In this scenario, we use non-linear optimization techniques to determine the optimal number of content copies under availability and latency constraints. In the second scenario, different nodes fail with different probabilities, and content is stored in nodes according to its value, node failure probability, and resource availability. This scenario is addressed as an instance of the minimum cost flow problem. Our results show that DISPERSE reduces the failure of content retrieval by five orders of magnitude compared to common CDN implementations, without significantly increasing content retrieval delay. Further, numerical results show that DISPERSE improves content availability by a factor of 1.3× - 2.3× when deploying the minimum cost flow algorithm.

This paper studies leakage of user passwords and PINs based on observations of typing feedback on screens or from projectors in the form of masked characters (∗ or ∙) that indicate keystrokes. To this end, we developed an attack called Password and Pin Information Leakage from Obfuscated Typing Videos ( PILOT ). Our attack extracts inter-keystroke timing information from videos of password masking characters displayed when users type their password on a computer, or their PIN at an ATM. We conducted several experiments in various attack scenarios. Results indicate that, while in some cases leakage is minor, it is quite substantial in others. By leveraging inter-keystroke timings, PILOT recovers 8-character alphanumeric passwords in as little as 19 attempts. When guessing PINs, PILOT significantly improved on both random guessing and the attack strategy adopted in our prior work (In European Symposium on Research in Computer Security ( 2018 ) 263–280 Springer). In particular, we were able to guess about 3% of the PINs within 10 attempts. This corresponds to a 26-fold improvement compared to random guessing. Our results strongly indicate that secure password masking GUIs must consider the information leakage identified in this paper.

Garbled circuits are general tools that allow two parties to compute any function without disclosing their respective inputs. Applications of this technique vary from distributed privacy-preserving machine learning tasks to secure outsourced authentication. Unfortunately, the energy cost of garbled circuit evaluation protocols is substantial. This limits the applicability of garbled circuits in scenarios that involve battery-operated devices, such as Internet-of-Things (IoT) devices and smartphones. In this paper, we propose MEG, a Memory- and Energy-efficient Garbled circuit evaluation mechanism. MEG utilizes batch data transmission and multi-threading to reduce memory and energy consumption. We implement MEG on an Android smartphone and compare its performance and energy consumption with state-of-the-art techniques using two garbled circuits of widely different sizes (AES-128 and 256-bit edit distance). Our results show that, compared with “plain” garbled circuit evaluation, MEG decreases memory consumption by more than 90%. When compared with current pipelined garbled circuit evaluation techniques, MEG's energy usage was 42% lower for AES-128 and 23% lower for EDT-256. Furthermore, our multi-thread implementation of MEG decreased circuit evaluation time by up to 56.7% for AES-128, and by up to 13.5% for EDT-256, compared with state-of-the-art pipelining techniques.

In this paper, we show that keystroke latencies used in continuous user authentication systems disclose application context, i.e., in which application user is entering text. Using keystroke data collected from 62 subjects, we show that an adversary can infer application context from keystroke latencies with 95.15% accuracy. To prevent leakage from keystroke latencies, and prevent exposure of application context, we develop privacy-preserving authentication protocols in the outsourced authentication model. Our protocols implement two popular matching algorithms designed for keystroke authentication, called Absolute (“A”) and Relative (“R”). With our protocols, the client reveals no information to the server during authentication, besides the authentication result. Our experiments show that these protocols are fast in practice: with 100 keystroke features, authentication was completed in about one second with the “A” protocol, and in 595 ms with the “R” protocol. Further, because the asymptotic cost of our protocols is linear, they can scale to a large number of features. On the other hand, by leveraging application context we were able to reduce HTER from 14.7% with application-agnostic templates, to as low as 5.8% with application-specific templates.

In this paper, we show that public USB charging stations pose a significant privacy risk to smartphone users even when no data communication is possible between the station and the user's mobile device. We present a side-channel attack that allows a charging station to identify which Webpages are loaded while the smartphone is charging. To evaluate this side-channel, we collected power traces of Alexa top 50 Websites on multiple smartphones under several conditions, including battery charging level, browser cache enabled/disabled, taps on the screen, Wi-Fi/LTE, TLS encryption enabled/disabled, time elapsed between collection of training and testing data, and location of the Website. The results of our evaluation show that the attack is highly successful: in many settings, we were able to achieve over 90% Webpage identification accuracy. On the other hand, our experiments also show that this side-channel is sensitive to some of the aforementioned conditions. For instance, when training and testing traces were collected 70 days apart, accuracies were as low as 2.2%. Although there are studies that show that power-based side-channels can predict browsing activity on laptops, this paper is unique, because it is the first to study this side-channel on smartphones, under smartphone specific constraints. Further, we demonstrate that Websites can be correctly identified within a short time span of 2 × 6 seconds, which is in contrast with prior work, which uses 15-s traces. This is important, because users typically spend less than 15 s on a Webpage.

Common smartphone authentication mechanisms (e.g., PINs, graphical passwords, and fingerprint scans) are not designed to offer security post-login. Multi-modal continuous authentication addresses this issue by frequently and unobtrusively authenticating the user via behavioral biometric signals, such as touchscreen interaction and hand movements. Because smartphones can easily fall into the hands of the adversary, it is critical that the behavioral biometric information collected and processed on these devices is secured. This can be done by offloading encrypted template information to a remote server, and then performing authentication via privacy-preserving protocols. In this paper, we demonstrate that the energy overhead of current privacy-preserving protocols for continuous authentication is unsustainable on smartphones. To reduce energy consumption, we design a technique that leverages characteristics unique to the authentication setting in order to securely outsource computation to an untrusted Cloud. Our approach is secure against a colluding smartphone and Cloud, thus making it well suited for authentication. We performed extensive experimental evaluation. With our technique, the energy requirement for running an authentication instance that computes Manhattan distance is 0.2 mWh, which corresponds to a negligible fraction of the smartphone's battery capacity. In addition, for Manhattan distance, our protocol runs in 0.72 and 2 s for 8 and 28 biometric features, respectively. We were also able to compute Hamming distance in 3.29 s, compared with 95.57 s achieved with the previous fastest outsourced computation protocol (Whitewash). These results demonstrate that ours is presently the only technique suitable for low-latency continuous authentication (e.g., with authentication scan windows of 60 s or shorter).

We introduce hand movement, orientation, and grasp (HMOG), a set of behavioral features to continuously authenticate smartphone users. HMOG features unobtrusively capture subtle micro-movement and orientation dynamics resulting from how a user grasps, holds, and taps on the smartphone. We evaluated authentication and biometric key generation (BKG) performance of HMOG features on data collected from 100 subjects typing on a virtual keyboard. Data were collected under two conditions: 1) sitting and 2) walking. We achieved authentication equal error rates (EERs) as low as 7.16% (walking) and 10.05% (sitting) when we combined HMOG, tap, and keystroke features. We performed experiments to investigate why HMOG features perform well during walking. Our results suggest that this is due to the ability of HMOG features to capture distinctive body movements caused by walking, in addition to the hand-movement dynamics from taps. With BKG, we achieved the EERs of 15.1% using HMOG combined with taps. In comparison, BKG using tap, key hold, and swipe features had EERs between 25.7% and 34.2%. We also analyzed the energy consumption of HMOG feature extraction and computation. Our analysis shows that HMOG features extracted at a 16-Hz sensor sampling rate incurred a minor overhead of 7.9% without sacrificing authentication accuracy. Two points distinguish our work from current literature: 1) we present the results of a comprehensive evaluation of three types of features (HMOG, keystroke, and tap) and their combinations under the same experimental conditions and 2) we analyze the features from three perspectives (authentication, BKG, and energy consumption on smartphones).

We design privacy-preserving protocols for scaled Manhattan and scaled Euclidean verifiers, secure against malicious clients and honest-but-curious server. We then augment our protocols with principal component analysis (PCA), which can help to improve authentication accuracy. We evaluate the performance of our protocols on an emerging application-namely, continuous authentication of smartphone users. We compare the performance of protocols secure under the malicious client model, with three protocols secure in the honest-but-curious model. We report tradeoffs between computation overhead, communication cost, and authentication accuracy. Our key observations are: 1) scaled Manhattan without PCA gives the best tradeoff between security, accuracy, and overhead and 2) with PCA, memory availability on current smartphones limits the number of features that can be used with scaled Manhattan, and prevents the scaled Euclidean protocol from running. Our extended evaluation on a laptop client shows that PCA with both scaled Manhattan and scaled Euclidean verifiers is feasible given sufficient memory.

Most continuous user authentication techniques based on typing behavior rely on the keystroke dynamics or on the linguistic style of the user. However, there is a rich spectrum of cognition-centric behavioral traits that a typist exhibits during different stages of text production (e.g., composition, translation, and revision), which to our knowledge, have not been considered for continuous authentication. We study the continuous authentication performance of 123 behavioral traits extracted from discrete cognitive units called bursts. We performed experiments on typing data collected from 486 volunteer subjects. Our findings include: (1) features from bursts delimited by pause events have significantly higher availability and authentication performance compared to bursts delimited by revision events; (2) bursts with pause durations of at least one second provide the best authentication accuracy and availability; and (3) fusing our features with traditional keystroke dynamics features reduced authentication error rates. We achieved an equal error rate between 13.37 and 4.55 percent for authentication windows as low as 30 seconds to 3.5 minutes.

Biometric key generation techniques are used to reliably generate cryptographic material from biometric signals. Existing constructions require users to perform a particular activity (e.g., type or say a password, or provide a handwritten signature), and are therefore not suitable for generating keys continuously. In this paper we present a new technique for biometric key generation from free-text keystroke dynamics. This is the first technique suitable for continuous key generation. Our approach is based on a scaled parity code for key generation (and subsequent key reconstruction), and can be augmented with the use of population data to improve security and reduce key reconstruction error. In particular, we rely on linear discriminant analysis (LDA) to obtain a better representation of discriminable biometric signals. To update the LDA matrix without disclosing user's biometric information, we design a provably secure privacy-preserving protocol (PP-LDA) based on homomorphic encryption. Our biometric key generation with PP-LDA was evaluated on a dataset of 486 users. We report equal error rate around 5% when using LDA, and below 7% without LDA.

Continuous authentication modalities allow a device to authenticate users transparently without interrupting them or requiring their attention. This is especially important on smartphones, which are more prone to be lost or stolen than regular computers, and carry plenty of sensitive information. There is a multitude of signals that can be harnessed for continuous authentication on mobile devices, such as touch input, accelerometer, and gyroscope, etc. However, existing public datasets include only a handful of them, limiting the ability to do experiments that involve multiple modalities. To fill this gap, we performed a large-scale user study to collect a wide spectrum of signals on smartphones. Our dataset combines more modalities than existing datasets, including movement, orientation, touch, gestures, and pausality. This dataset has been used to evaluate our new behavioral modality named Hand Movement, Orientation, and Grasp (H-MOG). This poster reports on the data collection process and outcomes, as well as preliminary authentication results.

We present a new attack called the snoop-forge-replay attack on keystroke-based continuous verification systems. The snoop-forge-replay is a sample-level forgery attack and is not specific to any particular keystroke-based continuous verification method or system. It can be launched with easily available keyloggers and APIs for keystroke synthesis. Our results from 2640 experiments show that: 1) the snoop-forge-replay attacks achieve alarmingly high error rates compared to zero-effort impostor attacks, which have been the de facto standard for evaluating keystroke-based continuous verification systems; 2) four state-of-the-art verification methods, three types of keystroke latencies, and 11 matching-pair settings (-a key parameter in continuous verification with keystrokes) that we examined in this paper were susceptible to the attack; 3) the attack is effective even when as low as 20 to 100 keystrokes were snooped to create forgeries. In light of our results, we question the security offered by current keystroke-based continuous verification systems. Additionally, in our experiments, we harnessed virtualization technology to generate thousands of keystroke forgeries within a short time span. We point out that virtualization setup such as the one used in our experiments can also be exploited by an attacker to scale and speedup the attack.

We derive the feature selection criterion presented in [CHECK END OF SENTENCE] and [CHECK END OF SENTENCE] from the multidimensional mutual information between features and the class. Our derivation: 1) specifies and validates the lower-order dependency assumptions of the criterion and 2) mathematically justifies the utility of the criterion by relating it to Bayes classification error.

Guo and Nixon proposed a feature selection method based on maximizingI(x;Y), the multidimensional mutual information between feature vectorxand class variableY. Because computingI(x;Y) can be difficult in practice, Guo and Nixon proposed an approximation ofI(x;Y) as the criterion for feature selection. We show that Guo and Nixon's criterion originates from approximating the joint probability distributions inI(x;Y) by second-order product distributions. We remark on the limitations of the approximation and discuss computationally attractive alternatives to computeI(x;Y) .

Wong and Poon [1] showed that Chow and Liu's tree dependence approximation can be derived by minimizing an upper bound of the Bayes error rate. Wong and Poon's result was obtained by expanding the conditional entropy H(w|X). We derive the correct expansion of H(w|X) and present its implication.

In this paper, we present "k-means+ID3", a method to cascade k-means clustering and the ID3 decision tree learning methods for classifying anomalous and normal activities in a computer network, an active electronic circuit, and a mechanical mass-beam system. The k-means clustering method first partitions the training instances into k clusters using Euclidean distance similarity. On each cluster, representing a density region of normal or anomaly instances, we build an ID3 decision tree. The decision tree on each cluster refines the decision boundaries by learning the subgroups within the cluster. To obtain a final decision on classification, the decisions of the k-means and ID3 methods are combined using two rules: 1) the nearest-neighbor rule and 2) the nearest-consensus rule. We perform experiments on three data sets: 1) network anomaly data (NAD), 2) Duffing equation data (DED), and 3) mechanical system data (MSD), which contain measurements from three distinct application domains of computer networks, an electronic circuit implementing a forced Duffing equation, and a mechanical system, respectively. Results show that the detection accuracy of the k-means+ID3 method is as high as 96.24 percent at a false-positive-rate of 0.03 percent on NAD; the total accuracy is as high as 80.01 percent on MSD and 79.9 percent on DED