<p dir="ltr">Differential privacy (DP) uses a probabilistic framework to measure the level of privacy protection of a mechanism that releases data analysis results to the public. Although DP is widely used by both government and industry, there is still a lack of research on statistical inference under DP guarantees. On the one hand, existing DP mechanisms mainly aim to extract dataset-level information instead of population-level information. On the other hand, DP mechanisms introduce calibrated noises into the released statistics, which often results in sampling distributions more complex and intractable than the non-private ones. This dissertation aims to provide general-purpose methods for statistical inference, such as confidence intervals (CIs) and hypothesis tests (HTs), that satisfy the DP guarantees. </p><p dir="ltr">In the first part of the dissertation, we examine a DP bootstrap procedure that releases multiple private bootstrap estimates to construct DP CIs. We present new DP guarantees for this procedure and propose to use deconvolution with DP bootstrap estimates to derive CIs for inference tasks such as population mean, logistic regression, and quantile regression. Our method achieves the nominal coverage level in both simulations and real-world experiments and offers the first approach to private inference for quantile regression.</p><p dir="ltr">In the second part of the dissertation, we propose to use the simulation-based ``repro sample'' approach to produce CIs and HTs based on DP statistics. Our methodology has finite-sample guarantees and can be applied to a wide variety of private inference problems. It appropriately accounts for biases introduced by DP mechanisms (such as by clamping) and improves over other state-of-the-art inference methods in terms of the coverage and type I error of the private inference. </p><p dir="ltr">In the third part of the dissertation, we design a debiased parametric bootstrap framework for DP statistical inference. We propose the adaptive indirect estimator, a novel simulation-based estimator that is consistent and corrects the clamping bias in the DP mechanisms. We also prove that our estimator has the optimal asymptotic variance among all well-behaved consistent estimators, and the parametric bootstrap results based on our estimator are consistent. Simulation studies show that our framework produces valid DP CIs and HTs in finite sample settings, and it is more efficient than other state-of-the-art methods.</p>

<p> The growing trend of encrypted network traffic is changing the cybersecurity threat scene. Most critical infrastructures and organizations enhance service delivery by embracing digital platforms and applications that use encryption to ensure that data and Information are moved across networks in an encrypted form to improve security. While this protects data confidentiality, hackers are also taking advantage of encrypted network traffic to hide malicious software known as malware that will easily bypass the conventional detection mechanisms on the system because the traffic is not transparent for the monitoring mechanism on the system to analyze. Cybercriminals leverage encryption using cryptographic protocols such as SSL/TLS to launch malicious attacks. This hidden threat exists because of the SSL encryption of benign traffic. Hence, there is a need for visibility in encrypted traffic. This research was conducted to detect malware in encrypted network traffic without decryption. The existing solution involves bulk decryption, analysis, and re-encryption. However, this method is prone to privacy issues, is not cost-efficient, and is time-consuming, creating huge overhead on the network. In addition, limited research exists on detecting malware in encrypted traffic without decryption. There is a need to strike a balance between security and privacy by building an intelligent framework that can detect malicious activity in encrypted network traffic without decrypting the traffic prior to inspection. With the payload still encrypted, the study focuses on extracting metadata from flow features to train the machine-learning model. It further deployed this set of features as input to an autoencoder, leveraging the construction error of the autoencoder for anomaly detection. </p>

<p dir="ltr">Protecting personal information is growing increasingly important to the general public, to the point that major tech companies now advertise the privacy features of their products. Despite this, it remains challenging to implement applications that do not leak private information either directly or indirectly, through timing behavior, memory access patterns, or control flow side channels. Existing security and cryptographic techniques such as secure multiparty computation (MPC) provide solutions to privacy-preserving computation, but they can be difficult to use for non-experts and even experts.</p><p dir="ltr">This dissertation develops the design, theory and implementation of various language-based techniques that help programmers write privacy-critical applications under a strong threat model. The proposed languages support private structured data, such as trees, that may hide their structural information and complex policies that go beyond whether a particular field of a record is private. More crucially, the approaches described in this dissertation decouple privacy and programmatic concerns, allowing programmers to implement privacy-preserving applications modularly, i.e., to independently develop application logic and independently update and audit privacy policies. Secure-by-construction applications are derived automatically by combining a standard program with a separately specified security policy.</p><p><br></p>

<p dir="ltr">The Internet of Things (IoT) systems include sensors that measure the physical world, actuators that influence it, and IoT apps that automate these sensors and actuators. Although IoT environments have revolutionized our lives by integrating digital connectivity into physical processes, they also introduce unique security and privacy concerns. Particularly, these systems include multiple components that are unified through the cyber and physical domains. For instance, smart homes include various devices and multiple IoT apps that control these devices. Thus, attacks against any single component can have rippling effects, amplifying due to the composite behavior of sensors, actuators, apps, and the physical environment.</p><p dir="ltr">In this dissertation, I explore the emerging security and privacy issues that arise from the complex physical interactions in IoT environments. To discover and mitigate these emerging issues, there is a need for composite reasoning techniques that consider the interplay between digital and physical domains. This dissertation addresses these challenges to build secure IoT environments and enhance user privacy with new formal techniques and systems.</p><p dir="ltr">To this end, I first describe my efforts in ensuring the safety and security of IoT en- vironments. Particularly, I introduced IoTSeer, a security service that discovers physical interaction vulnerabilities among IoT apps. I then proposed attacks that evade prior event verification systems by exploiting the complex physical interactions between IoT sensors and actuators. To address them, I developed two defenses, software patching and sensor placement, to make event verification systems robust against evasion attacks. These works provide a suite of tools to achieve compositional safety and security in IoT environments. </p><p dir="ltr">Second, I discuss my work that identifies the privacy risks of emerging IoT devices. I designed DMC-Xplorer to find vulnerabilities in voice assistant platforms and showed that an adversary can eavesdrop on privacy-sensitive device states and prevent users from controlling devices. I then developed a remote side-channel attack against intermittent devices to infer privacy-sensitive information about the environment in which they are deployed. These works highlight new privacy issues in emerging commodity devices used in IoT environments.</p>

<p dir="ltr">Distributed cryptographic protocols typically assume a bounded number of malicious parties (who behave arbitrarily) in the system---and in turn, a lower bound on the number of <i>honest</i> parties (who follow and only follow a protocol faithfully/honestly without performing unspecified computations)---for their respective security guarantees to hold. However, when deploying these protocols in practice, the nature of computing parties does not necessarily align nicely with the protocols' assumptions. Specifically, there may be only a few honest/compliant parties, or none exists. Instead, non-malicious parties may be <i>semi-honest</i> (who follow the protocol specifications but are curious to learn as much information as possible from semi-honest parties' transcripts) or <i>rational</i> (who take actions that maximize their utilities instead of actions benefiting the protocol the most, e.g., performing extra computations or not following protocols). In such cases, the security guarantees of such protocols may deviate greatly in real life from what is theoretically promised, leaving a huge gap between theory and practice. </p><p dir="ltr">In this thesis, I bridge such a gap by enhancing the fault tolerance of various distributed cryptographic primitives by <i>relaxing the assumption on the existence of honest parties</i>.</p><p dir="ltr">First, in the context of <b>secure multi-party computations</b>, without honest parties, my goal is to induce honest (i.e., not compromising correctness) and non-curious (i.e., not harming privacy) behaviors from rational participants via game theoretic and cryptographic techniques. In particular, I first demonstrate how to ensure protocol correctness and deter collusion among parties to recover secrets---which also breaks privacy---in multiserver private information retrieval with a singleton access structure. Then for primitives with more general (non-singleton) access structures, I introduce a distinct treatment through the lens of verifiable secret sharing. The two solutions are designed with a public bulletin board, commitment schemes, digital signature schemes, zkSNARKs (zero-knowledge succinct non-interactive arguments of knowledge), and distinct incentive structures tailored for varying access structures underlying the schemes.</p><p dir="ltr">Second, in <b>permissionless blockchain systems</b>, for protocols without privacy guarantees like computation outsourcing and consensus, my goal is to incentivize rational parties to behave correctly. This means to act according to the protocol specifications or as implied by the security requirements of the primitive, e.g., fairly distribute rewards to participants based on contributions in proof-of-work (PoW) blockchains. Specifically, I present a defense against an undercutting attack in PoW blockchains from a game theory perspective and propose a decentralized computation outsourcing protocol built on permissionless blockchain systems based on multi-unit auctions.</p>

<p>Over the past decade, distributed representation learning has emerged as a popular alternative to conventional centralized machine learning training. The increasing interest in distributed representation learning, specifically federated learning, can be attributed to its fundamental property that promotes data privacy and communication savings. While conventional ML encourages aggregating data at a central location (e.g., data centers), distributed representation learning advocates keeping data at the source and instead transmitting model parameters across the network. However, since the advent of deep learning, model sizes have become increasingly large often comprising million-billions of parameters, which leads to the problem of communication latency in the learning process. In this thesis, we propose to tackle the problem of communication latency in two different ways: (i) learning private representation of data to enable its sharing, and (ii) reducing the communication latency by minimizing the corresponding long-range communication requirements.</p> <p><br></p> <p>To tackle the former goal, we first start by studying the problem of learning representations that are private yet informative, i.e., providing information about intended ''ally'' targets while hiding sensitive ''adversary'' attributes. We propose Exclusion-Inclusion Generative Adversarial Network (EIGAN), a generalized private representation learning (PRL) architecture that accounts for multiple ally and adversary attributes, unlike existing PRL solutions. We then address the practical constraints of the distributed datasets by developing Distributed EIGAN (D-EIGAN), the first distributed PRL method that learns a private representation at each node without transmitting the source data. We theoretically analyze the behavior of adversaries under the optimal EIGAN and D-EIGAN encoders and the impact of dependencies among ally and adversary tasks on the optimization objective. Our experiments on various datasets demonstrate the advantages of EIGAN in terms of performance, robustness, and scalability. In particular, EIGAN outperforms the previous state-of-the-art by a significant accuracy margin (47% improvement), and D-EIGAN's performance is consistently on par with EIGAN under different network settings.</p> <p><br></p> <p>We next tackle the latter objective - reducing the communication latency - and propose two timescale hybrid federated learning (TT-HF), a semi-decentralized learning architecture that combines the conventional device-to-server communication paradigm for federated learning with device-to-device (D2D) communications for model training. In TT-HF, during each global aggregation interval, devices (i) perform multiple stochastic gradient descent iterations on their individual datasets, and (ii) aperiodically engage in consensus procedure of their model parameters through cooperative, distributed D2D communications within local clusters. With a new general definition of gradient diversity, we formally study the convergence behavior of TT-HF, resulting in new convergence bounds for distributed ML. We leverage our convergence bounds to develop an adaptive control algorithm that tunes the step size, D2D communication rounds, and global aggregation period of TT-HF over time to target a sublinear convergence rate of O(1/t) while minimizing network resource utilization. Our subsequent experiments demonstrate that TT-HF significantly outperforms the current art in federated learning in terms of model accuracy and/or network energy consumption in different scenarios where local device datasets exhibit statistical heterogeneity. Finally, our numerical evaluations demonstrate robustness against outages caused by fading channels, as well favorable performance with non-convex loss functions.</p>

<p dir="ltr">This thesis presents an in depth study and evaluation of a class of secure multiparty protocols that enable execution of a confidential software program $\mathcal{P}$ owned by Alice, on confidential data $\mathcal{D}$ owned by Bob, without revealing anything about $\mathcal{P}$ or $\mathcal{D}$ in the process. Our initial adverserial model is an honest-but-curious adversary, which we later extend to a malicious adverarial setting. Depending on the requirements, our protocols can be set up such that the output $\mathcal{P(D)}$ may only be learned by Alice, Bob, both, or neither (in which case an agreed upon third party would learn it). Most of our protocols are run by only two online parties which can be Alice and Bob, or alternatively they could be two commodity cloud servers (in which case neither Alice nor Bob participate in the protocols' execution - they merely initialize the two cloud servers, then go offline). We implemented and evaluated some of these protocols as prototypes that we made available to the open source community via Github. We report our experimental findings that compare and contrast the viability of our various approaches and those that already exist. All our protocols achieve the said goals without revealing anything other than upper bounds on the sizes of program and data.</p><p><br></p>

<p dir="ltr">With the ability to connect to networks and send and receive data, Internet of Things (IoT) devices involve associated security risks and threats, for a given environment. These threats are even more of a concern in a Smart Home network, where there is a lack of a dedicated security IT team, unlike a corporate environment. While efficient user interface(UI) and ease of use is at the front and center of IoT devices within Smart Home which enables its wider adoption, often security and privacy have been an afterthought and haven’t kept pace when needed. Therefore, a unsafe possibility exists where malicious actors could exploit vulnerable devices in a domestic home environment.</p><p dir="ltr">This thesis involves a detailed study of the cybersecurity for a Smart Home and also examines the various types of cyberthreats encountered, such as DDoS, Man-In-Middle, Ransomware, etc. that IoT devices face. Given, IoT devices are commonplace in most home automation scenarios, its crucially important to detect intrusions and unauthorized access. Privacy issues are also involved making this an even more pertinent topic. Towards this, various state of the art industry standard tools, such as Nmap, Nessus, Metasploit, etc. were used to gather data on a Smart Home environment to analyze their impacts to detect security vulnerabilities and risks to a Smart Home. Results from the research indicated various vulnerabilities, such as open ports, password vulnerabilities, SSL certificate anomalies and others that exist in many cases, and how precautions when taken in timely manner can help alleviate and bring down those risks.</p><p dir="ltr">Also, an IoT monitoring dashboard was developed based on open-source tools, which helps visualize threats and emphasize the importance of monitoring. The IoT dashboard showed how to raise alerts and alarms based on specific threat conditions or events. In addition, currently available cybersecurity regulations, standards, and guidelines were also examined that can help safeguard against threats to commonly used IoT devices in a Smart Home. It is hoped that the research carried out in this dissertation can help maintain safe and secure Smart Homes and provide direction for future work in the area of Smart Home Cybersecurity.</p>

<p dir="ltr">The research work presented in this thesis focuses on two highly important topics in the modern age. The first topic of research is the development of various image-based Network Intrusion Detection Systems (NIDSs) and performing a comprehensive analysis of their performance. Wi-Fi networks have become ubiquitous in enterprise and home networks which creates opportunities for attackers to target the networks. These attackers exploit various vulnerabilities in Wi-Fi networks to gain unauthorized access to a network or extract data from end users' devices. The deployment of an NIDS helps detect these attacks before they can cause any significant damages to the network's functionalities or security. Within the scope of our research, we provide a comparative analysis of various deep learning (DL)-based NIDSs that utilize various imaging techniques to detect anomalous traffic in a Wi-Fi network. The second topic in this thesis is the development of learning modules for security and privacy in Machine Learning (ML). The increasing integration of ML in various domains raises concerns about its security and privacy. In order to effectively address such concerns, students learning about the basics of ML need to be made aware of the steps that are taken to develop robust and secure ML-based systems. As part of this, we introduce a set of hands-on learning modules designed to educate students on the importance of security and privacy in ML. The modules provide a theoretical learning experience through presentations and practical experience using Python Notebooks. The modules are developed in a manner that allows students to easily absorb the concepts regarding privacy and security of ML models and implement it in real-life scenarios. The efficacy of this process will be obtained from the results of the surveys conducted before and after providing the learning modules. Positive results from the survey will demonstrate the learning modules were effective in imparting knowledge to the students and the need to incorporate security and privacy concepts in introductory ML courses.</p>