Biblio

This tutorial will introduce participants to Grounded Theory, which is a qualitative framework to discover new theory from an empirical analysis of data. This form of analysis is particularly useful when analyzing text, audio or video artifacts that lack structure, but contain rich descriptions. We will frame Grounded Theory in the context of qualitative methods and case studies, which complement quantitative methods, such as controlled experiments and simulations. We will contrast the approaches developed by Glaser and Strauss, and introduce coding theory - the most prominent qualitative method for performing analysis to discover Grounded Theory. Topics include coding frames, first- and second-cycle coding, and saturation. We will use examples from security interview scripts to teach participants: developing a coding frame, coding a source document to discover relationships in the data, developing heuristics to resolve ambiguities between codes, and performing second-cycle coding to discover relationships within categories. Then, participants will learn how to discover theory from coded data. Participants will further learn about inter-rater reliability statistics, including Cohen's and Fleiss' Kappa, Krippendorf's Alpha, and Vanbelle's Index. Finally, we will review how to present Grounded Theory results in publications, including how to describe the methodology, report observations, and describe threats to validity.

Polynomial masking is a glitch-resistant and higher-order masking scheme based upon Shamir's secret sharing scheme and multi-party computation protocols. Polynomial masking was first introduced at CHES 2011, while a 1st-order implementation of the AES S-box on FPGA was presented at CHES 2013. In this latter work, the authors showed a 2nd-order univariate leakage by side-channel collision analysis on a tuned measurement setup. This negative result motivates the need to evaluate the performance, area-costs, and security margins of combined \shuffled\ and higher-order polynomially masking schemes to counteract trivial univariate leakages. In this work, we provide the following contributions: first, we introduce additional principles for the selection of efficient addition chains, which allow for more compact and faster implementations of cryptographic S-boxes. Our 1st-order AES S-box implementation requires approximately 27% less registers, 20% less clock cycles, and 5% less random bits than the CHES 2013 implementation. Then, we propose a lightweight shuffling countermeasure, which inherently applies to polynomial masking schemes and effectively enhances their univariate security at negligible area expenses. Finally, we present the design of a \combined\ \shuffled\ \and\ higher-order polynomially masked AES S-box in hardware, while providing ASIC synthesis and side-channel analysis results in the Electro-Magnetic (EM) domain.

Eddy is a privacy requirements specification language that privacy analysts can use to express requirements over data practices; to collect, use, transfer and retain personal and technical information. The language uses a simple SQL-like syntax to express whether an action is permitted or prohibited, and to restrict those statements to particular data subjects and purposes. Eddy also supports the ability to express modifications on data, including perturbation, data append, and redaction. The Eddy specifications are compiled into Description Logic to automatically detect conflicting requirements and to trace data flows within and across specifications. Conflicts are highlighted, showing which rules are in conflict (expressing prohibitions and rights to perform the same action on equivalent interpretations of the same data, data subjects, or purposes), and what definitions caused the rules to conflict. Each specification can describe an organization's data practices, or the data practices of specific components in a software architecture.

Masking requires splitting sensitive variables into at least d+1 shares to provide security against DPA attacks at order d. To this date, this minimal number has only been deployed in software implementations of cryptographic algorithms and in the linear parts of their hardware counterparts. So far there is no hardware construction that achieves this lower bound if the function is nonlinear and the underlying logic gates can glitch. In this paper, we give practical implementations of the AES using d+1 shares aiming at first- and second-order security even in the presence of glitches. To achieve this, we follow the conditions presented by Reparaz et al. at CRYPTO 2015 to allow hardware masking schemes, like Threshold Implementations, to provide theoretical higher-order security with d+1 shares. The decrease in number of shares has a direct impact in the area requirements: our second-order DPA resistant core is the smallest in area so far, and its S-box is 50% smaller than the current smallest Threshold Implementation of the AES S-box with similar security and attacker model. We assess the security of our masked cores by practical side-channel evaluations. The security guarantees are met with 100 million traces.

There are two broad approaches for differentially private data analysis. The interactive approach aims at developing customized differentially private algorithms for various data mining tasks. The non-interactive approach aims at developing differentially private algorithms that can output a synopsis of the input dataset, which can then be used to support various data mining tasks. In this paper we study the effectiveness of the two approaches on differentially private k-means clustering. We develop techniques to analyze the empirical error behaviors of the existing interactive and non-interactive approaches. Based on the analysis, we propose an improvement of DPLloyd which is a differentially private version of the Lloyd algorithm. We also propose a non-interactive approach EUGkM which publishes a differentially private synopsis for k-means clustering. Results from extensive and systematic experiments support our analysis and demonstrate the effectiveness of our improvement on DPLloyd and the proposed EUGkM algorithm.

I will explain the linkage between threshold implementation masking schemes and multi-party computation. The basic principles that need to be taken from multi-party computation will be presented, as well as some basic protocols. The different natures of the resources and threat models between the two different applications of secret sharing will also be covered.

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.

Event discovery from single pictures is a challenging problem that has raised significant interest in the last decade. During this time, a number of interesting solutions have been proposed to tackle event discovery in still images. However, a large scale benchmarking image dataset for the evaluation and comparison of event discovery algorithms from single images is still lagging behind. To this aim, in this paper we provide a large-scale properly annotated and balanced dataset of 490,000 images, covering every aspect of 14 different types of social events, selected among the most shared ones in the social network. Such a large scale collection of event-related images is intended to become a powerful support tool for the research community in multimedia analysis by providing a common benchmark for training, testing, validation and comparison of existing and novel algorithms. In this paper, we provide a detailed description of how the dataset is collected, organized and how it can be beneficial for the researchers in the multimedia analysis domain. Moreover, a deep learning based approach is introduced into event discovery from single images as one of the possible applications of this dataset with a belief that deep learning can prove to be a breakthrough also in this research area. By providing this dataset, we hope to gather research community in the multimedia and signal processing domains to advance this application.

In most adaptive video streaming systems adaptation decisions rely solely on the available network resources. As the content of a video has a large influence on the perception of quality our belief is that this is not sufficient. Thus, we have proposed a support service for content-aware video adaptation on mobile devices: Video Adaptation Service (VAS). Based on the content of a streamed video, the adaptation process is improved by setting a target quality level for a session based on an objective video quality metric. In this work, we demonstrate VAS and its advantages of a reduced data traffic by only streaming the lowest video representation which is necessary to reach a desired quality. By leveraging the content properties of a video stream, the system is able to keep a stable video quality and at the same time reduce the network load.

TCP congestion control has been known for its crucial role in stabilizing the Internet and preventing congestion collapses. However, with the rapid advancement in networking technologies, resulting in the emergence of challenging network environments such as data center networks (DCNs), the traditional TCP algorithm leads to several impairments. The shortcomings of TCP when deployed in DCNs have motivated the development of multiple new variants, including DCTCP, ICTCP, IA-TCP, and D2TCP, but all of these algorithms exhibit their advantages at the cost of a number of drawbacks in the Global Internet. Motivated by the belief that new innovations need to be established on top of a solid foundation with a thorough understanding of the existing, well-established algorithms, we have been working towards a comprehensive analysis of various conventional TCP algorithms in DCNs and other modern networks. This paper presents our first milestone towards the completion of our comparative study in which we present the results obtained by simulating multiple TCP variants: NewReno, Vegas, HighSpeed, Scalable, Westwood+, BIC, CUBIC, and YeAH using a fat tree architecture. Each protocol is evaluated in terms of queue length, number of dropped packets, average packet delay, and aggregate bandwidth as a percentage of the channel bandwidth.

We address the problem of stabilizing control for complex queueing systems with known parameters but unobservable Markovian random environment. In such systems, the controller needs to assign servers to queues without having full information about the servers' states. A control challenge is to devise a policy that matches servers to queues in a way that takes state estimates into account. Maximally attainable stability regions are non-trivial. To handle these situations, we model the system under given decision rules. The model is using Quasi-Birth-and-Death (QBD) structure to find a matrix analytic expression for the stability bound. We use this formulation to illustrate how the stability region grows as the number of controller belief states increases.

IP tracking and cloaking are practices for identifying users which are used legitimately by websites to provide services and content tailored to particular users. However, it is believed that these practices are also used by malicious websites to avoid detection by anti-virus companies crawling the web to find malware. In addition, malicious websites are also believed to use IP tracking in order to deliver targeted malware based upon a history of previous visits by users. In this paper we empirically investigate these beliefs and collect a large dataset of suspicious URLs in order to identify at what level IP tracking takes place that is at the level of an individual address or at the level of their network provider or organisation (Network tracking). Our results illustrate that IP tracking is used in a small subset of domains within our dataset while no strong indication of network tracking was observed.

Large scale sensor networks are ubiquitous nowadays. An important objective of deploying sensors is to detect anomalies in the monitored system or infrastructure, which allows remedial measures to be taken to prevent failures, inefficiencies, and security breaches. Most existing sensor anomaly detection methods are local, i.e., they do not capture the global dependency structure of the sensors, nor do they perform well in the presence of missing or erroneous data. In this paper, we propose an anomaly detection technique for large scale sensor data that leverages relationships between sensors to improve robustness even when data is missing or erroneous. We develop a probabilistic graphical model-based global outlier detection technique that represents a sensor network as a pairwise Markov Random Field and uses graphical model inference to detect anomalies. We show our model is more robust than local models, and detects anomalies with 90% accuracy even when 50% of sensors are erroneous. We also build a synthetic graphical model generator that preserves statistical properties of a real data set to test our outlier detection technique at scale.

The speech emotion recognition accuracy of prosody feature and voice quality feature declines with the decrease of SNR (Signal to Noise Ratio) of speech signals. In this paper, we propose novel sub-band spectral centroid weighted wavelet packet cepstral coefficients (W-WPCC) for robust speech emotion recognition. The W-WPCC feature is computed by combining the sub-band energies with sub-band spectral centroids via a weighting scheme to generate noise-robust acoustic features. And Deep Belief Networks (DBNs) are artificial neural networks having more than one hidden layer, which are first pre-trained layer by layer and then fine-tuned using back propagation algorithm. The well-trained deep neural networks are capable of modeling complex and non-linear features of input training data and can better predict the probability distribution over classification labels. We extracted prosody feature, voice quality features and wavelet packet cepstral coefficients (WPCC) from the speech signals to combine with W-WPCC and fused them by Deep Belief Networks (DBNs). Experimental results on Berlin emotional speech database show that the proposed fused feature with W-WPCC is more suitable in speech emotion recognition under noisy conditions than other acoustics features and proposed DBNs feature learning structure combined with W-WPCC improve emotion recognition performance over the conventional emotion recognition method.

Heart rate monitoring has become increasingly popular in the industry through mobile phones and wearable devices. However, current determination of heart rate through mobile applications suffers from high corruption of signals during intensive physical exercise. In this paper, we present a novel technique for accurately determining heart rate during intensive motion by classifying PPG signals obtained from smartphones or wearable devices combined with motion data obtained from accelerometer sensors. Our approach utilizes the Internet of Things (IoT) cloud connectivity of smartphones for selection of PPG signals using deep learning. The technique is validated using the TROIKA dataset and is accurately able to predict heart rate with a 10-fold cross validation error margin of 4.88%.

This paper proposes a method for segmentation of nuclei of single/isolated and overlapping/touching immature white blood cells from microscopic images of B-Lineage acute lymphoblastic leukemia (ALL) prepared from peripheral blood and bone marrow aspirate. We propose deep belief network approach for the segmentation of these nuclei. Simulation results and comparison with some of the existing methods demonstrate the efficacy of the proposed method.

Abnormality detection is useful in reducing the amount of data to be processed manually by directing attention to the specific portion of data. However, selections of suitable features are important for the success of an abnormality detection system. Designing and selecting appropriate features are time-consuming, requires expensive domain knowledge and human labor. Further, it is very challenging to represent high-level concepts of abnormality in terms of raw input. Most of the existing abnormality detection system use handcrafted feature detector and are based on shallow architecture. In this work, we explore Deep Belief Network for abnormality detection and simultaneously, compared the performance of classic neural network in terms of features learned and accuracy of detecting the abnormality. Further, we explore the set of features learn by each layer of the deep architecture. We also provide a simple and fast mechanism to visualize the feature at the higher layer. Further, the effect of different activation function on abnormality detection is also compared. We observed that deep learning based approach can be used for detecting an abnormality. It has better performance compare to classical neural network in separating distinct as well as almost similar data.

With the increased popularity of ubiquitous computing and connectivity, the Internet of Things (IoT) also introduces new vulnerabilities and attack vectors. While secure data collection (i.e. the upward link) has been well studied in the literature, secure data dissemination (i.e. the downward link) remains an open problem. Attribute-based encryption (ABE) and outsourced-ABE has been used for secure message distribution in IoT, however, existing mechanisms suffer from extensive computation and/or privacy issues. In this paper, we explore the problem of privacy-preserving targeted broadcast in IoT. We propose two multi-cloud-based outsourced-ABE schemes, namely the parallel-cloud ABE and the chain-cloud ABE, which enable the receivers to partially outsource the computationally expensive decryption operations to the clouds, while preventing user attributes from being disclosed. In particular, the proposed solution protects three types of privacy (i.e., data, attribute and access policy privacy) by enforcing collaborations among multiple clouds. Our schemes also provide delegation verifiability that allows the receivers to verify whether the clouds have faithfully performed the outsourced operations. We extensively analyze the security guarantees of the proposed mechanisms and demonstrate the effectiveness and efficiency of our schemes with simulated resource-constrained IoT devices, which outsource operations to Amazon EC2 and Microsoft Azure.

For single-owner multi-user wireless sensor networks, there is the demand to implement the user privacy-preserving access control protocol in WSNs. Firstly, we propose a new access control protocol based on an efficient attribute-based signature. In the protocol, users need to pay for query, and the protocol achieves fine-grained access control and privacy protection. Then, the protocol is analyzed in detail. Finally, the comparison of protocols indicates that our scheme is more efficient. Our scheme not only protects the privacy of users and achieves fine-grained access control, but also provides the query command validation with low overhead. The scheme can better satisfy the access control requirements of wireless sensor networks.

Outsourcing a huge amount of local data to remote cloud servers that has been become a significant trend for industries. Leveraging the considerable cloud storage space, industries can also put forward the outsourced data to cloud computing. How to collect the data for computing without loss of privacy and confidentiality is one of the crucial security problems. Searchable encryption technique has been proposed to protect the confidentiality of the outsourced data and the privacy of the corresponding data query. This technique, however, only supporting search functionality, may not be fully applicable to real-world cloud computing scenario whereby secure data search, share as well as computation are needed. This work presents a novel encrypted cloud-based data share and search system without loss of user privacy and data confidentiality. The new system enables users to make conjunctive keyword query over encrypted data, but also allows encrypted data to be efficiently and multiply shared among different users without the need of the "download-decrypt-then-encrypt" mode. As of independent interest, our system provides secure keyword update, so that users can freely and securely update data's keyword field. It is worth mentioning that all the above functionalities do not incur any expansion of ciphertext size, namely, the size of ciphertext remains constant during being searched, shared and keyword-updated. The system is proven secure and meanwhile, the efficiency analysis shows its great potential in being used in large-scale database.

In the area of the Internet of Things, cloud-based camera surveillance systems are ubiquitously available for industrial and private environments. However, the sensitive nature of the surveillance use case imposes high requirements on privacy/confidentiality, authenticity, and availability of such systems. In this work, we investigate how currently available mass-market camera systems comply with these requirements. Considering two attacker models, we test the cameras for weaknesses and analyze for their implications. We reverse-engineered the security implementation and discovered several vulnerabilities in every tested system. These weaknesses impair the users' privacy and, as a consequence, may also damage the camera system manufacturer's reputation. We demonstrate how an attacker can exploit these vulnerabilities to blackmail users and companies by denial-of-service attacks, injecting forged video streams, and by eavesdropping private video data - even without physical access to the device. Our analysis shows that current systems lack in practice the necessary care when implementing security for IoT devices.

Computing similarity, especially Jaccard Similarity, between two datasets is a fundamental building block in big data analytics, and extensive applications including genome matching, plagiarism detection, social networking, etc. The increasing user privacy concerns over the release of has sensitive data have made it desirable and necessary for two users to evaluate Jaccard Similarity over their datasets in a privacy-preserving manner. In this paper, we propose two efficient and secure protocols to compute the Jaccard Similarity of two users' private sets with the help of an unfully-trusted server. Specifically, in order to boost the efficiency, we leverage Minhashing algorithm on encrypted data, where the output of our protocols is guaranteed to be a close approximation of the exact value. In both protocols, only an approximate similarity result is leaked to the server and users. The first protocol is secure against a semi-honest server, while the second protocol, with a novel consistency-check mechanism, further achieves result verifiability against a malicious server who cheats in the executions. Experimental results show that our first protocol computes an approximate Jaccard Similarity of two billion-element sets within only 6 minutes (under 256-bit security in parallel mode). To the best of our knowledge, our consistency-check mechanism represents the very first work to realize an efficient verification particularly on approximate similarity computation.

Unease over data privacy will retard consumer acceptance of IoT deployments. The primary source of discomfort is a lack of user control over raw data that is streamed directly from sensors to the cloud. This is a direct consequence of the over-centralization of today's cloud-based IoT hub designs. We propose a solution that interposes a locally-controlled software component called a privacy mediator on every raw sensor stream. Each mediator is in the same administrative domain as the sensors whose data is being collected, and dynamically enforces the current privacy policies of the owners of the sensors or mobile users within the domain. This solution necessitates a logical point of presence for mediators within the administrative boundaries of each organization. Such points of presence are provided by cloudlets, which are small locally-administered data centers at the edge of the Internet that can support code mobility. The use of cloudlet-based mediators aligns well with natural personal and organizational boundaries of trust and responsibility.

With the increasing popularity of cloud storage services, many individuals and enterprises start to move their local data to the clouds. To ensure their privacy and data security, some cloud service users may want to encrypt their data before outsourcing them. However, this impedes efficient data utilities based on the plain text search. In this paper, we study how to construct a secure index that supports both efficient index updating and similarity search. Using the secure index, users are able to efficiently perform similarity searches tolerating input mistakes and update the index when new data are available. We formally prove the security of our proposal and also perform experiments on real world data to show its efficiency.

The wide presence of large graph data and the increasing popularity of storing data in the cloud drive the needs for graph query processing on a remote cloud. But a fundamental challenge is to process user queries without compromising sensitive information. This work focuses on privacy preserving subgraph matching in a cloud server. The goal is to minimize the overhead on both cloud and client sides for subgraph matching, without compromising users' sensitive information. To that end, we transform an original graph \$G\$ into a privacy preserving graph Gk, which meets the requirement of an existing privacy model known as k-automorphism. By making use of the symmetry in a k-automorphic graph, a subgraph matching query can be efficiently answered using a graph Go, a small subset of Gk. This approach saves both space and query cost in the cloud server. We also anonymize the query graphs to protect their label information using label generalization technique. To reduce the search space for a subgraph matching query, we propose a cost model to select the more effective label combinations. The effectiveness and efficiency of our method are demonstrated through extensive experimental results on real datasets.