Search

The ever-increasing ubiquity of cyber-physical systems is creating an attractive target for developing malware and cyber attacks. From consumer-grade IoT devices like drones to specialized industrial equipment in critical infrastructure, many new and varied attacks and malware emerge to afflict these systems, impacting the physical environment and processes handled by the devices.A fundamental part of protecting a system against malware and cyber attacks involves understanding the nature of the attack to devise an effective countermeasure. However, due to the nature of some of these systems, a comprehensive analysis might be unfeasible or outright dangerous. To overcome this limitation, we propose using a virtualized scenario, emulating the system's behavior, more specifically the physical and networking behavior, to dynamically analyze cyber-attacks and malware without compromising the physical integrity of the systems and their environment.We first test the feasibility of this idea by simulating cyber-attacks against a virtual quadcopter drone before testing the attacks on its physical counterpart once the attacks are deemed "safe" for the user and the drone. Then, we emulate an industrial process using software-defined networks to evaluate the feasibility of implementing defense mechanisms against attacks with some features provided by software-defined networks. Finally, we combine the virtualization of cyber-physical systems and the use of software-defined networks to simulate a power grid system to dynamically analyze the behavior of an actual malware known as "Win32.Industroyer" by infecting a virtual machine in an isolated virtual environment. Our ultimate goal is to refine a framework to allow researchers to safely simulate cyber-physical systems to test attacks, defenses, and malware against a virtual avatar of the existing system.

In our rapidly evolving technological landscape, the proliferation of enabling technologies for autonomous systems has given rise to a burgeoning realm of societal-scale smart systems. One noteworthy category within this domain is Human-Cyber-Physical Systems (H-CPS), which encompass physical systems controlled by a blend of computer-based algorithms and human inputs. Examples of H-CPS include the smart grid and autonomous transportation systems. These systems harness the potential of distributed computing units, fast communication channels, and real-time data collection, offering efficient mechanisms for their management. This requires a synthesis of tools from distributed optimization, machine learning, game theory, and stochastic control.However, the advent of H-CPS also presents novel challenges. Human decisions, often stochastic and beyond direct control, must be factored into the developed mechanisms. Moreover, the dependable operation of H-CPS hinges on secure communication between physical systems and computing units, raising concerns regarding user data privacy and system security. The growing number of humans and devices generates copious amounts of data from sensing units, necessitating computationally efficient data processing to ensure seamless H-CPS operation.This thesis aims to design network control, optimization, and learning frameworks that enhance safety, robustness, and efficiency in H-CPS, with practical applications in smart infrastructure systems like the power grid and transportation networks. Additionally, its relevance extends to diverse Internet of Things applications, emphasizing user data privacy, such as the development of language models from text data. The thesis unfolds in three interconnected chapters. In the first chapter, we introduce provably efficient and adversarially robust multi-agent optimization algorithms tailored for distributed resource allocation and distributed learning scenarios in the presence of malicious agents. Moving forward to the second chapter, we aim to design prices for shared resources that do not violate hard (mainly physical) constraints of the system, without any two-way communications with the users as common in distributed optimization based methods. The third chapter focuses on crafting and analyzing joint ride pricing and fleet management policies for the control of autonomous urban mobility fleets. Throughout these chapters, we not only analyze the theoretical performance of our proposed mechanisms but also substantiate their effectiveness through extensive simulations on real-world problems.

This Ph.D. dissertation presents a comprehensive investigation into addressing security and reliability challenges in embedded and Cyber-Physical Systems (CPS). Our research leverages advanced machine learning techniques such as Graph Neural Networks (GNN) to develop novel methodologies for cross-layer security analysis.This dissertation addresses the growing risk posed by the globalization of the Integrated Circuit (IC) supply chain, whereby the majority of the design, fabrication, and testing processes have been outsourced to untrusted third-party entities across the globe. This development has significantly increased the threat of malicious modifications, known as Hardware Trojans (HTs), being inserted into Third-Party Intellectual Property (3PIP). HTs pose a substantial risk to IC integrity, functionality, and performance.Despite numerous HT detection methods proposed in existing literature, most limitations include reliance on a golden reference circuit, lack of generalizability, limited detection scope, low localization resolution, and manual feature extraction and property definition. Furthermore, the equally important task of HT localization has been neglected.This research proposes an innovative, golden reference-free method for HT detection and localization at the pre-silicon stage of IC development, employing models based on GNN. The circuit design is converted into a graph that is an intrinsic data structure for hardware design and captures the computational structure and data dependencies. We develop a graph classification model to distinguish between HT-free and circuits infected with known or even unknown HTs.To push the boundaries further, we extract node attributes from the HDL code and devise a Graph Convolutional Network (GCN) that facilitates automatic feature extraction, enabling the classification of nodes as either Trojan or benign. This methodology offers an automated approach to HT detection and localization, relieving designers of the need for time-consuming manual code review. The developed method achieves exceptional performance in detecting HT-infected circuits and locating the HT. The approach outlined in this dissertation sets a new benchmark for HT detection and localization, offering a scalable, efficient, and highly accurate tool for securing the pre-silicon IC supply chain.This dissertation expands to encompass the challenges facing IP piracy. The productivity gap, coupled with time-to-market pressure, has led to increased interest in hardware Intellectual Property (IP) core design within the semiconductor industry, dramatically reducing design and verification costs. Recognizing these challenges, this dissertation proposes a novel IP piracy detection methodology, modeling circuits and assessing similarity between IP designs. Contrary to traditional methods that embed a signature within the circuit design, our method does not introduce additional hardware overhead, nor is it vulnerable to removal, masking, or forging attacks. This approach effectively exposes IP infringements, even when the original IP is complicated by the adversary to deceive the IP owner. To represent the circuit accurately for modeling, we translate the hardware design into a data-flow graph due to similar data types and properties and subsequently model it using state-of-the-art graph learning methods. This approach effectively complements the GNN-based techniques proposed earlier in this dissertation, presenting a robust and comprehensive suite of solutions for security and reliability challenges in the semiconductor industry. Moving to the CPS domain, the dissertation addresses security challenges in IoT systems through the development of adaptive anomaly detection methods. The first proposed approach utilizes IoT sensor data and fog computing to ensure data integrity and detect anomalous incidents. The proposed methodology incorporates our sensor association algorithm, LSTM neural networks, and Gaussian estimation for real-time anomaly detection. The dissertation further extends the research to multi-modal data fusion, where the integration of sensor and communication data using GNN enables improved anomaly detection, source identification, and recovery in IoT systems.Overall, this dissertation showcases the application of advanced techniques such as GNN and machine learning in enhancing security and reliability in hardware design and IoT systems. The proposed methodologies for anomaly detection, hardware Trojan detection, IP piracy detection, and cross-layer security analysis contribute to advancing the state-of-the-art in ensuring the integrity and security of critical systems in the digital era.

Cyber-physical systems (CPS) are increasingly adopting learning-enabled components having deep neural networks in their decision-making pipelines. Deep neural networks show the promise to simplify the CPS pipelines for high-dimensional sensors as they require little pre-processing of data and are shown to be more accurate than their traditional counterparts. However, integrating neural networks into the sense-infer-actuate pipeline of CPS faces several challenges. In this dissertation, we study the following challenges in the context of learning-enabled CPS and propose new algorithms and system design strategies to address them. First, we study the challenge of characterizing uncertainty in sensor data timestamps and its impact on multimodal fusion applications. Motivated by smartphones' integration in several CPS applications, we quantify the data timestamp uncertainty across modern smartphone devices. To our surprise, we find drastic timestamping errors ranging up to multiple seconds in Android devices. Then, we explore if these timing errors are significant enough to impact the neural network's performance. Our evaluation shows that the observed timing errors can cripple the deep neural networks doing multimodal fusion due to data misalignments. Our finding signifies the need to rethink the shared notion of time on smartphones. To mitigate timestamp errors, we introduce approaches to improve time across smartphones having up to 200 microseconds of timing accuracy. We also propose a novel time-shift data augmentation technique to train time-resilient neural networks robust to the inevitability of timing errors and, as such, degrade gracefully in the face of timing errors.As a second challenge, we explore the impact of variable delays on the emerging deep reinforcement learning (RL) controllers, which are preferred due to their capability to handle high-dimensional data. Conventional controllers can model and account for delay variations in their design. However, handling variable delays in deep-RL is challenging as a black-box neural network represents the controller policy. Researchers currently use domain randomization and worst-case delay modeling to train deep-RL policies on a spread of expected delay variations. We demonstrate a significant performance degradation in applications even when using the state-of-the-art domain randomization approach. To address this, we propose Time-in-State RL, a delay-aware deep RL approach that augments the agent's state with temporal properties (sampling interval and execution latency). Time-in-State RL trains policies that show superior performance by adapting to the variable timing characteristics at runtime. We further show the superior performance of Time-in-State to the worst-case delay controllers when worst-case delays are significant. We demonstrate the efficacy of Time-in-State RL on HalfCheetah, Ant, and car in simulation and on a real scaled car robot.Thirdly, we study the challenge of modeling the CPS environment to train end-to-end controllers using deep-RL for closed-loop systems.We specifically consider the example of autonomous pan-tilt-zoom (PTZ) controllers. Existing autonomous PTZ controllers have multiple stages: detecting objects of interest, short-term tracking, and control of pan, tilt and zoom parameters to keep objects in the field of view. The multiple stages suffer from performance bottlenecks as it is difficult to optimize each step. Further, these multiple stages are computationally intensive to be realized in real-time on embedded camera platforms. Despite these shortcomings, developers adopt existing multi-stage solutions due to the lack of simulators needed to develop end-to-end controller policies.We propose Eagle, an end-to-end deep-RL approach using raw images to control a PTZ camera. To enable successful training of Eagle, we also introduce EagleSim, a simulation framework to study PTZ cameras in photo-realistic virtual worlds. Our evaluation across a suite of PTZ tracking scenarios shows that Eagle outperforms current multi-stage approaches by providing superior tracking performance. Further, we also show that Eagle policies are transferable to real-scene videos and are lightweight to enable real-time deployment on Raspberry PI and Jetson Nano class devices.Finally, we study the challenge of developing machine learning classifiers having optimal accuracy within the desired resource budget of CPS applications. Selecting an optimal classifier is becoming increasingly complex, with many choices for classifiers and their rich hyperparameter parameter spaces. Although several hyperparameter tuning frameworks exist, their practical adoption is hindered due to inferior search algorithms, inflexible architecture, software dependencies, or closed source nature. As a solution, we propose designing a lightweight library with a flexible architecture and state-of-the-art parallel optimization algorithms. We present Mango, a parallel hyperparameter tuning library, to realize the proposed design. Mango is currently used in production at Arm for more than 30 months and is available open-source. We evaluate Mango on several benchmarks to highlight its superior performance.We discuss production use cases of Mango in an AutoML framework and commercial CPU design pipeline. We also showcase another advantage of Mango in enabling hardware-aware neural architecture search to transfer deep neural networks to TinyML platforms (microcontroller class devices) used by CPS/IoT applications.

This thesis considers the analysis and design of algorithms for the management and control of uncertain intelligent systems which are observable through (limited) online-accessible data. Examples include online equity trading systems under extreme price fluctuations, robotic systems moving in unknown environments, and transportation systems subject to uncertain drivers’ actions and other (accident) events.To ensure safe, reliable, and resilient system behaviors, this thesis studies various theoretical problem scenarios, which focus on reducing uncertainty with performance guarantees via the assimilation of streaming data, the data-driven design of control, and online learning of system models, resilient operations in uncertain environments, and anomaly detection.These formulations are largely rooted in two mechanisms: online optimization and distributionally robust optimization, where the first enables online-tractable formulations of the problem, and the latter accounts for systemic uncertainty with high confidence. Both approaches are applicable beyond the particular systems of study, to virtually any type of dynamic system where sensitive data is progressively available and may be exploited to the advantage of management and control. This work is unique in that it brings together current tools in optimization, control of dynamical systems, data-based modeling and probability theory, significantly advancing the state of the art.

Security and usability are often paradoxically portrayed as zero-sum trade-offs; this is especially true of cyber-physical systems. Historically, security has taken a back seat in open source robotic software initiatives, such as Robot Operating System (ROS), which have focused on development productivity instead of deployment hardening, and this has contributed, perhaps inadvertently, to the belated adoption of ROS by industry, impeding technical transfer between industry and academia. This dissertation studies methods for automating and verifying middleware configurations to improve the security for distributed robotic systems.We introduce procedural tooling for automating the setup of authenticated encryption and access control across the distributed data bus network used in robotic middleware frameworks, in a concerted effort to remove sources of errors such as human factors during configuration development. This includes an immutable logging framework with which to record event data used by any tooling in order to continuously profile and/or monitor system security. We then present methods for conducting passive reachability and controllability analysis over the data bus, and for combining that with static information flow control to guarantee isolation between sensitive or mission critical computations over larger integrations. Finally, we propose principles for the design, verification, and deployment of robotic architectures composed of partially distrusting components. We seek to establish a set of automated tools, methods, and design principles to compose and audit multiple robotic subsystems into hardened stacks for service and industry, and we call it "Usable Security and Verification for Distributed Robotic Systems."

In the age of digitization, a layer of cyber software sits on a hardware circuit and controls the physical systems around us. The tight integration of cyber and physical components is referred to as Cyber-Physical Systems (CPS). The interactions between cyber and physical components brings unique challenges which traditional modeling tools struggle to resolve. Remarkably, they often fail to model the unintentional physical manifestation of cyber-domain information flows (side-channel signals), resulting in trust issues in the system.This thesis takes a data-driven approach to model CPS behavior when exposed to various information flows. First, we demonstrate how to extract valuable cyber-domain information by recording the acoustic noise generated by a DNA synthesizer. Then, we consider an integrated circuit as a CPS by itself and monitor the chip through electromagnetic and power side-channels to detect hardware Trojans (HT) in the chip.HT is a malicious modification of the hardware implementation of a circuit design which may lead to various security issues over the life-cycle of a chip. One of the major challenges for HT detection is its reliance on a trusted reference chip (a.k.a golden chip). However, in practice, manufacturing a golden chip is costly and often considered infeasible. This thesis investigates a creative neural network design and training methodology which eliminates the need for a golden chip. Furthermore, it proposes using hierarchical temporal memory (HTM) as a data-driven approach which can be updated over the chip's life-cycle and uses that for run-time HT detection.

This dissertation focuses on the challenges arising from real-time autonomous cyber-physical systems. Since many cyber-physical applications increasingly require high performance to run complex functionalities, e.g., self-driving software stacks, it is essential to use limited resources efficiently on resource-constrained embedded platforms. Besides, as real-time autonomy is expected to become more pervasive in various safety-critical application domains, e.g., aerospace and defense, timeliness end-to-end latency of critical computation chains is particularly crucial because the late response or the violation of timing constraints may cause catastrophic consequences. On top of that, any design and analysis methods to fulfill these requirements should be predictable in order to establish a reliable execution foundation.To address the aforementioned challenges, we develop analyzable yet practical scheduling techniques for practical real-time cyber-physical systems, with examples of autonomous vehicles. First, we propose a novel job-class-level scheduler (JCLS) equipped with a low-complexity analysis tool. The key observation behind this work is that many cyber-physical applications can often tolerate a certain degree of timing violations as long as the number of the violations is predictably bounded. By capturing this effect, JCLS exploits application-specific bounding constraints and efficiently manages limited resources, especially, enables overloaded workloads schedulable on embedded platforms. Secondly, we develop a chain-based scheduling method (CBS) to improve the data freshness of real-time tasks with data dependency. This technique provides better quality-of-service outputs by exploiting the effective job-level information flows in the read-execute-write model which is prevalent in automotive systems. Lastly, we propose a new scheduling architecture design Robot Operating System (ROS2), which is the most popular open-source robotic framework. Unlike the default fairness-based resource management methods in ROS2, our priority-driven chain-aware scheduling (PiCAS) enables prioritization of critical computation chains across system layers to minimize end-to-end latency, and its effectiveness has been verified under real-world scenarios. The contributions of this dissertation pave the road towards designing practical autonomous real-time systems with efficient and predictable scheduling and resource management schemes.

Cyber-Physical Systems (CPS) play an ubiquitous role in operation and control in many different domains: power systems, finance, robotics, and automation. The complex interplay between cyber components such as software, communication protocols, computer servers and physical components, such as sensors and pieces of dedicated hardware, requires advanced and sophisticated methods and algorithms that ensure safe and efficient operation. In this thesis we tackle both safety and efficiency: We develop novel detection algorithms that are able to identify malicious attacks, sensor corruption and faulty measurements. Our detection mechanisms have provable guarantees based on rigorous asymptotic and non-asymptotic statistical analysis and can be readily implemented in CPS, such as robotic systems and autonomous vehicles. In addition, we developed collusion detection mechanisms that can be used to identify whether two or more CPS are colluding or not. We also design a mechanism that is able to induce selfish systems/agents to behave cooperatively. We showcase the performance of our algorithms with several different case studies. In our analyses, we place emphasis on algorithms that can be implemented in real-time, that is can be used while the system is under operation in the real-world. On the efficiency side, we developed real-time non-linear Model Predictive Control (MPC) Methods that can provide optimal solutions to the Optimal Control problem faced by the CPS during operation. Our algorithm exploits the control structure and is tailored for implementation in embedded hardware and can operate both with memory and computation time constraints. We showcase the performance of our algorithm with a C/C++ implementation and we compare to several current state-of-the-art Optimal Control solvers. We also extend our methodology to be used together with Pseudo-spectral Methods and Hybrid Systems, developing an integrated Mixed-Integer MPC algorithm that can handle complex non-linear dynamics and both continuous and discrete variables. With this thesis, our goal is to provide real-time practical algorithms that have provable guarantees in performance both in the detection task and in the optimal control task. Our algorithms are based on rigorous theoretical analysis and display very good performance and can be readily implemented in practical Cyber-Physical Systems.

We are captivated by the promise of autonomous systems in our everyday life. However, ensuring that these systems act safely is an immense challenge: introducing complex systems into real-world uncertain environments while guaranteeing safety at all times is impossible in applications like self-driving vehicles, collaborative factory robots, and assistive robots. These systems will inevitably need to make real-time decisions with limited computational resources, and incomplete knowledge of the environment and other agents. This dissertation is an effort towards achieving trustworthy real-world autonomy by enabling autonomous systems to: (1) make theoretical safety guarantees efficiently based on known information, and(2) bridge the safety gap between this theory and the real world by reasoning about uncertainty in the environment and other agents.Towards this goal this dissertation covers various methods for scalable safety and real-time decision-making that draw from control theory, cognitive science, and learning, and are backed by both rigorous theory and physical testing on robotic platforms. We begin with an overview of reachability analysis and its applications for optimal control with safety guarantees. We then tackle the curse of dimensionality associated with reachability computation by decomposing systems or updating previous solutions in a warm-starting fashion. Next we explore planning in a simplified, low-dimensional space online with precomputed safety guarantees and tracking controls offline. This is extended to meta-planning, in which the online algorithm switches between faster/conservative modes and slower/accurate modes. Then we apply all of these tools towards navigating in uncertain environments among hard-to-predict agents such as human pedestrians. We use Bayesian machine learning methods to reason about human intention and navigate in a probabilistically safe manner. Finally, the dissertation ends with information about code bases and reachability tutorial examples.

Cyber-Physical Systems (CPS) are composed of computing devices interacting with physical systems. Model-based design is a powerful methodology in CPS design in the implementation of control systems. For instance, Model Predictive Control (MPC) is typically implemented in CPS applications, e.g., in path tracking of autonomous vehicles. MPC deploys a model to estimate the behavior of the physical system at future time instants for a specific time horizon. Ordinary Differential Equations (ODE) are the most commonly used models to emulate the behavior of continuous-time (non-)linear dynamical systems. A complex physical model may comprise thousands of ODEs that pose scalability, performance and power consumption challenges. One approach to address these model complexity challenges are frameworks that automate the development of model-to-model transformation. In this dissertation, a state-based model with tunable parameters is proposed to operate as a reconfigurable predictive model of the physical system. Moreover, we propose a run-time switching algorithm that selects the best model using machine learning. We employed a metric that formulates the trade-off between the error and computational savings due to model reduction. Building statistical models are constrained to having expert knowledge and an actual understanding of the modeled phenomenon or process. Also, statistical models may not produce solutions that are as robust in a real-world context as factors outside the model, like disruptions would not be taken into account. Machine learning models have emerged as a solution to account for the dynamic behavior of the environment and automate intelligence acquisition and refinement. Neural networks are machine learning models, well-known to have the ability to learn linear and nonlinear relations between input and output variables without prior knowledge. However, the ability to efficiently exploit resource-hungry neural networks in embedded resource-bound settings is a major challenge.Here, we proposed Priority Neuron Network (PNN), a resource-aware neural networks model that can be reconfigured into smaller sub-networks at runtime. This approach enables a trade-off between the model's computation time and accuracy based on available resources. The PNN model is memory efficient since it stores only one set of parameters to account for various sub-network sizes. We propose a training algorithm that applies regularization techniques to constrain the activation value of neurons and assigns a priority to each one. We consider the neuron's ordinal number as our priority criteria in that the priority of the neuron is inversely proportional to its ordinal number in the layer. This imposes a relatively sorted order on the activation values. We conduct experiments to employ our PNN as the predictive model in a CPS application. We can see that not only our technique will resolve the memory overhead of DNN architectures but it also reduces the computation overhead for the training process substantially. The training time is a critical matter especially in embedded systems where many NN models are trained on the fly.

Recent technological advances in networking, communication, and computation technologies have enabled the development of cyber-physical systems and cloud robotics, where computing, communication, and control are tightly coupled and integrated into a single distributed platform. These systems open the door to a myriad of new and exciting applications in transportation, health care, agriculture, energy, and many others. The need for the tight integration of different components, requirements, and time scales means that the modeling, analysis, and design of these systems present new challenges. We focus on two aspects of emerging systems architecture. Firstly, we investigate the presence of finite-rate, digital communication channels with delays in the feedback loop. In this context, we study event-triggering strategies that utilize timing information by transmitting in a state-dependent fashion. The proposed event-triggering strategies utilize the available communication resources more efficiently compared to existed time-triggering setups. Secondly, the distributed nature of cyber-physical systems and cloud robotics is their Achilles’ heel, as it is a source of vulnerability to cyber-attacks. In this regard, we introduce the problem of learning-based attacks in these systems, and we show how the controller can impede these attacks by superimposing a carefully crafted privacy-enhancing signal upon its control policy.

Unprecedented Cyber-Physical Systems (CPS) and Internet of Things (IoT) applications such as health care, connected vehicles, and augmented/virtual reality are revolutionizing smart spaces and change how we build and manage our systems. These applications span the cloud and the edge devices and give birth to new system designs with critical dependence on temporal use cases. As such, cloud services are expected to provide timely responses and schedulable demands, while edge devices are required to synchronize observations and choreograph actions across distributed entities. Both cloud and edge demand time awareness in general, and time-indexed queries, precise timestamping, and dynamic clock synchronization in particular. However, contemporary distributed system designs are inherently “clockless” and becoming increasingly complex. They fail to meet consistency, causality, and scheduling demands of underlying applications yet enabling time awareness for various applications running on commodity platforms and operating systems (OS) is a challenge in itself.In this dissertation, we devise a new way of acquiring time information by introducing the notion of Quality of Time (QoT) that collectively captures various time metrics such as resolution, accuracy, stability, and integrity. Analogous to Quality of Service (QoS) in networking, QoT treats time as a controllable OS primitive with observable performance. To provide QoT to applications, we proposed the first OS abstraction – timeline – that reacts to application timing demands and exposes QoT to applications in an easy-to-use, secure, and scalable way. This degree of richness of information had never been available to coordinated applications whose activities are choreographed across time and space. This flow of information was immediately relevant to the broader field of IoT addressing the emerging temporal use cases for applications at the cloud and the edge in a secure fashion. As such, QoT expanded distributed applications to a global scale with no significant overhead and no performance compromise.This dissertation focuses on covering various aspects of QoT. In the first part of this thesis, we design extensible abstractions to characterize timing uncertainty in the presence of timing variations. In an effort to reduce complexity and overhead of current distributed database designs, our abstractions and systems enable globally replicated lockless transactions with simplified design, low overhead and no loss in performance. The second part exposes timing vulnerabilities in trusted execution technologies and network security mechanisms and provides timing integrity by designing secure time architectures in the presence of vulnerabilities. Thus enabling trusted timestamping in commodity systems to preserve one’s digital rights and digital signals. The third part focuses on redesigning the hardware, OS and network interfaces that help time information flow between applications and systems, and enable timing precision. This precision boosts high-speed measurements at large-scale distributed entities. The final section addresses the inefficacy of testing mechanisms for time synchronization protocols deployed in safety-critical environments. We then propose a customized testbed for testing timing robustness under failures and adversarial attacks.Current designs in distributed systems rely on message-passing based protocols and come at a huge energy and bandwidth cost along with high system complexity. In contrast, our designed systems based on QoT support new temporal use cases of globally distributed applications with low computation and communication overhead. We released our system designs to support various time management and clock synchronization use cases in emerging distributed applications.

In this thesis, we consider secure communication in the presence of an eavesdropper. With the explosion in the growth of the data produced and communicated, sensitive information such as financial transactions, health records, and control signals for cyber-physical systems, has to be securely exchanged. Today, the ever-increasing computational power of adversaries is challenging the state-of-the-art cryptographic encryption mechanisms, as these mechanisms assume adversaries with limited computational power. Thus, with the advent of the quantum computing era, we require new mechanisms to guarantee a secure exchange of information. Moreover, the growing number of small and energy constrained connected devices involved in data exchange calls for lightweight encryption schemes, as low-complexity devices cannot implement complex schemes. We consider three different scenarios and exploit specific opportunities present in each of these scenarios; we develop lightweight encryption schemes that do not require sharing large keys in advance and are secure against eavesdroppers with unlimited computational capabilities.The first scenario we consider is multiple unicast traffic over wireline networks. In these networks, a single source is connected to m destinations interested in different messages. In designing encryption schemes, we exploit the fact that although the eavesdropper is computationally super-powerful, it might not have capabilities to eavesdrop the entire network. We use the multi-path diversity to securely communicate against the eavesdropper without requiring any pre-shared key.The second scenario we consider consists of millimeter wave (mmWave) networks. mmWave communication requires deploying networks of relays that communicate through directional beams to compensate for the high path-loss and the high blockage. Since we need to use beamforming and align beams to activate links, we cannot use all the underlying links of the network simultaneously. However, the degree of freedom in choosing the links to activate can be leveraged for secure communication against an eavesdropper. We show that we can achieve a secure capacity that in some cases, can be very close to the unsecure capacity. Here, capacity refers to the maximum flow of information over the network.For the third scenario, we consider cyber-physical systems and propose a distortion based security framework where the distortion measures the distance between the eavesdropper's estimates and the ground truth. The primary motivation for this framework is that the messages exchanged in these systems are embedded in a metric space having a notion of distance, and securing raw bits as in traditional encryption schemes might not be necessary. Instead, we show with an example of a linear dynamical system that a carefully designed encryption scheme can significantly distort the eavesdropper's view with just one bit of the pre-shared key.

Cyber-Physical System consists of the integration of computational components in the cyber-domain with the physical-domain processes. In cyber-domain, embedded computers and networks monitor and control the physical processes, and in the physical-domain the sensors and actuators aid in interacting with the physical world. This interaction between the cyber and physical domain brings unique modeling challenges one of which includes the integration of discrete and sequential models in cyber-domain with the continuous and parallel physical domain processes. However, the same cyber-physical interaction also opens new opportunities for modeling. For example, the information flow in the cyber-domain manifests physically in the form of energy flows in the physical domain. Some of these energy flows unintentionally provide information about the cyber-domain through the side-channels. In this thesis, the first part consists of an extensive analysis of the side-channels (such as acoustic, magnetic, thermal, power and vibration) of the cyber-physical system is performed. Based on this analysis data-driven models are estimated. These models are then used to perform security vulnerability analysis (for confidentiality and integrity), whereby, new attack models are explored. Furthermore, the data-driven models are also utilized to create a defensive mechanism to minimize the information leakage from the system and to detect attacks to the integrity of the system. The cyber-physical manufacturing systems are taken as use cases to demonstrate the applicability of the modeling approaches. In the second part, side-channel analysis is performed to aid in modeling digital twins of the cyber-physical systems. Specifically, a novel methodology to utilize low-end sensors to analyze the side-channels and build the digital twins is proposed. These digital twins are used to capture the interaction between the cyber-domain and the physical domain of the manufacturing systems, and aid in process quality inference and fault localization. Using side-channels these digital twins are kept up-to-date, which is one of the fundamental requirements for building digital twins. Finally, challenges relating to performing data-driven modeling using non-Euclidean data in the cyber-physical system are addressed in the third part of the thesis. Moreover, a novel structural graph convolutional neural network and a dynamic graph embedding algorithm are presented to handle non-Euclidean data.

We present VerifAI, a software toolkit for the formal design and analysis of systems that include artificial intelligence (AI) and machine learning (ML) components. VerifAI particularly addresses challenges with applying formal methods to ML components such as perception systems based on deep neural networks, as well as systems containing them, and to model and analyze system behavior in the presence of environment uncertainty. We describe the initial version of VerifAI, which centers on simulation-based verification and synthesis, guided by formal models and specifications. We give examples of several use cases, including temporal-logic falsification, model-based systematic fuzz testing, parameter synthesis, counterexample analysis, and data set augmentation.

The security of a cyber-physical system depends on the safety of the handled data, software, and underlying hardware. Securing the hardware is not a simple task because of the globalization of integrated circuits’ manufacturing flow. One hardware attack to be considered is the modification of the design to insert a “backdoor” which maliciously alters the behavior of the original system. Such a malicious and intentional insertion is called a Hardware Trojan Horse (HTH). In this thesis, an HTH detection technique was proposed and implemented. The detection technique made use of side-channel power analysis along with machine learning to detect the presence of an HTH. Power traces from a golden implementation (HTH-free) of the AES encryption algorithm on an FPGA were used to train a logistic regression model. The obtained model was then tested on new power traces collected from the golden implementation and was able to make correct predictions with 95% accuracy. Next, an HTH, of a few gates, was implemented in the AES circuit to carry out a denial-of-service attack along with a breach of plaintext secrecy. The power data from the HTH-infected circuit were collected and tested on the trained logistic regression model. An amount of 81% of the HTH-infected data was detected as flawed by the logistic regression model allowing the detection of the HTH even when it was not triggered. In fact, even when an HTH was dormant, the HTH would constantly be checking its triggering condition and hence consumed power.