I2S Masters/ Doctoral Theses

Chapter 1 introduces TinyML, emphasizing the shift from large-scale machine learning to embedded, resource-constrained devices. It explores DeepPiCar, a project demonstrating the application of TinyML principles to autonomous driving on low-cost platforms. The chapter details the motivation, setup, and fine-tuning process used to improve DeepPiCar's adaptability in changing environments using a Raspberry Pi Zero 2 W. Key findings highlight the feasibility and efficiency of on-device fine-tuning.

Chapter 2 delves into prior research that has demonstrated the effectiveness of end-to-end deep learning for robotic navigation, where the control signals are directly derived from raw sensory data. However, the majority of existing end-to-end navigation solutions are predominantly camera-based. In this chapter, we introduce TinyLidarNet, a lightweight 2D LiDAR-based end-to-end deep learning model for autonomous racing. We systematically analyze its performance on untrained tracks and computing requirements for real-time processing. We find that TinyLidarNet's 1D Convolutional Neural Network (CNN) based architecture significantly outperforms widely used Multi-Layer Perceptron (MLP) based architecture. In addition, we show that it can be processed in real-time on low-end micro-controller units (MCUs).

Chapter 3 comprehensively analyzes the variation of TinyLidarNet, an enhanced deep learning model optimized for F1TENTH autonomous racing using the ESP32S3 microcontroller and RPLiDAR A1 sensor. It is aimed at addressing challenges like size, weight, power efficiency, and environmental adaptability in the F1TENTH autonomous racing competition. Through a combination of expert driving data and the Dataset Aggregation (DAGger) technique, the model was trained to improve navigation accuracy and processing efficiency. Testing demonstrated that the vehicle could complete multiple laps on both familiar and new tracks without collisions while meeting real-time processing requirements.

In modern computer systems, sensors play a critical role in enabling a wide range of functionalities, from navigation in autonomous vehicles to environmental monitoring in smart homes. Acting as an interface between physical and digital worlds, sensors collect data to drive automated functionalities and decision-making. However, this reliance on sensor data introduces significant potential vulnerabilities, leading to various physical, sensor-enabled attacks such as spoofing, tampering, and signal injection. Sensor spoofing attacks, where adversaries manipulate sensor input or inject false data into target systems, pose serious risks to system security and privacy.

In this work, we have developed two novel sensor spoofing attack methods that significantly enhance both efficacy and practicality. The first method employs physical signals that are imperceptible to humans but detectable by sensors. Specifically, we target deep learning based facial recognition systems using infrared lasers. By leveraging advanced laser modeling, simulation-guided targeting, and real-time physical adjustments, our infrared laser-based physical adversarial attack achieves high success rates with practical real-time guarantees, surpassing the limitations of prior physical perturbation attacks. The second method embeds physical signals, which are inherently present in the system, into legitimate patterns. In particular, we integrate trigger signals into standard operational patterns of actuators on mobile devices to construct remote logic bombs, which are shown to be able to evade all existing detection mechanisms. Achieving a zero false-trigger rate with high success rates, this novel sensor bomb is highly effective and stealthy.

Our study on emerging sensor-based threats highlights the urgent need for comprehensive defenses against sensor spoofing. Along this direction, we design and investigate two defense strategies to mitigate these threats. The first strategy involves filtering out physical signals identified as potential attack vectors. The second strategy is to leverage beneficial physical signals to obfuscate malicious patterns and reinforce data integrity. For example, side channels targeting the same sensor can be used to introduce cover signals that prevent information leakage, while environment-based physical signals serve as signatures to authenticate data. Together, these strategies form a comprehensive defense framework that filters harmful sensor signals and utilizes beneficial ones, significantly enhancing the overall security of cyber systems.

In modern computer systems, sensors play a critical role in enabling a wide range of functionalities, from navigation in autonomous vehicles to environmental monitoring in smart homes. Acting as an interface between physical and digital worlds, sensors collect data to drive automated functionalities and decision-making. However, this reliance on sensor data introduces significant potential vulnerabilities, leading to various physical, sensor-enabled attacks such as spoofing, tampering, and signal injection. Sensor spoofing attacks, where adversaries manipulate sensor input or inject false data into target systems, pose serious risks to system security and privacy.

In this work, we have developed two novel sensor spoofing attack methods that significantly enhance both efficacy and practicality. The first method employs physical signals that are imperceptible to humans but detectable by sensors. Specifically, we target deep learning based facial recognition systems using infrared lasers. By leveraging advanced laser modeling, simulation-guided targeting, and real-time physical adjustments, our infrared laser-based physical adversarial attack achieves high success rates with practical real-time guarantees, surpassing the limitations of prior physical perturbation attacks. The second method embeds physical signals, which are inherently present in the system, into legitimate patterns. In particular, we integrate trigger signals into standard operational patterns of actuators on mobile devices to construct remote logic bombs, which are shown to be able to evade all existing detection mechanisms. Achieving a zero false-trigger rate with high success rates, this novel sensor bomb is highly effective and stealthy.

Our study on emerging sensor-based threats highlights the urgent need for comprehensive defenses against sensor spoofing. Along this direction, we design and investigate two defense strategies to mitigate these threats. The first strategy involves filtering out physical signals identified as potential attack vectors. The second strategy is to leverage beneficial physical signals to obfuscate malicious patterns and reinforce data integrity. For example, side channels targeting the same sensor can be used to introduce cover signals that prevent information leakage, while environment-based physical signals serve as signatures to authenticate data. Together, these strategies form a comprehensive defense framework that filters harmful sensor signals and utilizes beneficial ones, significantly enhancing the overall security of cyber systems.

Quantum computing emerges as a promising direction for high-performance computing in the post-Moore era. Leveraging quantum mechanical properties, quantum devices can theoretically provide significant speedup over classical computers in certain problem domains. Quantum algorithms are typically expressed as quantum circuits composed of quantum gates, or as unitary matrices. Execution of quantum algorithms on physical devices requires translation to machine-compatible circuits -- a process referred to as quantum compilation or synthesis. 

 Quantum synthesis is a challenging problem. Physical quantum devices support a limited number of native basis gates, requiring synthesized circuits to be composed of only these gates. Moreover, quantum devices typically have specific qubit topologies, which constrain how and where gates can be applied. Consequently, logical qubits in input circuits and unitaries may need to be mapped to and routed between physical qubits on the device.

 Current Noisy Intermediate-Scale Quantum (NISQ) devices present additional constraints, through their gate errors and high susceptibility to noise. NISQ devices are vulnerable to errors during gate application and their short decoherence times leads to qubits rapidly succumbing to accumulated noise and possibly corrupting computations. Therefore, circuits synthesized for NISQ devices need to have a low number of gates to reduce gate errors, and short execution times to avoid qubit decoherence. 

 The problem of synthesizing device-compatible quantum circuits, while optimizing for low gate count and short execution times, can be shown to be computationally intractable using analytical methods. Therefore, interest has grown towards heuristics-based compilation techniques, which are able to produce approximations of the desired algorithm to a required degree of precision. In this work, we investigate using Genetic Algorithms (GAs) -- a proven gradient-free optimization technique based on natural selection -- for circuit synthesis. In particular, we formulate the quantum synthesis problem as a multi-objective optimization (MOO) problem, with the objectives of minimizing the approximation error, number of multi-qubit gates, and circuit depth. We also employ fuzzy logic for runtime parameter adaptation of GA to enhance search efficiency and solution quality of our proposed quantum synthesis method.

There is an increasing trend in the number of wireless devices that is now already over 14 billion and is expected to grow to 40 billion devices by 2030. In addition, we are witnessing an unprecedented proliferation of applications and technologies with wireless connectivity requirements such as unmanned aerial vehicles, connected health, and radars for autonomous vehicles. The advent of new wireless technologies and devices will only worsen the current spectrum crunch that service providers and wireless operators are already experiencing. In this PhD study, we address these challenges through the following research thrusts, in which we consider two emerging applications aimed at advancing spectrum efficiency and high-frequency connectivity solutions.

First, we focus on effectively utilizing the existing spectrum resources for emerging applications such as networked UAVs operating within the Unmanned Traffic Management (UTM) system. In this thrust, we develop a coexistence framework for UAVs to share spectrum with traditional cellular networks by using machine learning (ML) techniques so that networked UAVs act as secondary users without interfering with primary users. We propose federated learning (FL) and reinforcement learning (RL) solutions to establish a collaborative spectrum sensing and dynamic spectrum allocation framework for networked UAVs. In the second part, we explore the potential of millimeter-wave (mmWave) and terahertz (THz) frequency bands for high-speed data transmission in urban settings. Specifically, we investigate THz-based midhaul links for 5G networks, where a network's central units (CUs) connect to distributed units (DUs). Through numerical analysis, we assess the feasibility of using 140 GHz links and demonstrate the merits of high-frequency bands to support high data rates in midhaul networks for future urban communications infrastructure. Overall, this research is aimed at establishing frameworks and methodologies that contribute toward the sustainable growth and evolution of wireless connectivity.

Person tracking and association is a complex task in computer vision applications. Even with a powerful detector, a highly accurate association algorithm is necessary to match and track the correct person across all frames. This method has numerous applications in surveillance, and its complexity increases with the number of detected objects and their movements across frames. A significant challenge in person tracking is occlusion, which occurs when an individual being tracked is partially or fully blocked by another object or person. This can make it difficult for the tracking system to maintain the identity of the individual and track them effectively.

In this research, we propose a solution to the occlusion problem by utilizing an occlusion-aware spatial attention transformer. We have divided the entire tracking association process into two scenarios: occlusion and no-occlusion. When a detected person with a specific ID suddenly disappears from a frame for a certain period, we employ advanced methods such as Detector Integration and Pose Estimation to ensure the correct association. Additionally, we implement a spatial attention transformer to differentiate these occluded detections, transform them, and then match them with the correct individual using the Cosine Similarity Metric.

The features extracted from the attention transformer provide a robust baseline for detecting people, enhancing the algorithms adaptability and addressing key challenges associated with existing approaches. This improved method reduces the number of misidentifications and instances of ID switching while also enhancing tracking accuracy and precision.

The global textile and apparel (T&A) industry is under growing scrutiny for its substantial environmental and social impact, producing 92 million tons of waste annually and contributing to 20% of global water pollution. In Bangladesh, one of the world's largest apparel exporters, the integration of Environmental, Social, and Governance (ESG) practices is critical to meet international sustainability standards and maintain global competitiveness. This master's study leverages Artificial Intelligence (AI) and Machine Learning (ML) methodologies to comprehensively analyze unstructured corporate data related to ESG practices among LEED-certified Bangladeshi T&A factories.

Our study employs advanced techniques, including Web Scraping, Natural Language Processing (NLP), and Topic Modeling, to extract and analyze sustainability-related information from factory websites. We develop a robust ML framework that utilizes Non-Negative Matrix Factorization (NMF) for topic extraction and a Random Forest classifier for ESG category prediction, achieving an 86% classification accuracy. The study uncovers four key ESG themes: Environmental Sustainability, Social : Workplace Safety and Compliance, Social: Education and Community Programs, and Governance. The analysis reveals that 46% of factories prioritize environmental initiatives, such as energy conservation and waste management, while 44% emphasize social aspects, including workplace safety and education. Governance practices are significantly underrepresented, with only 10% of companies addressing ethical governance, healthcare provisions and employee welfare.

To deepen our understanding of the ESG themes, we conducted a Centrality Analysis to identify the most influential keywords within each category, using measures such as degree, closeness, and eigenvector centrality. Furthermore, our analysis reveals that higher certification levels, like Platinum, are associated with a more balanced emphasis on environmental, social, and governance practices, while lower levels focus primarily on environmental efforts. These insights highlight key areas where the industry can improve and inform targeted strategies for enhancing ESG practices. Overall, this ML framework provides a data-driven, scalable approach for analyzing unstructured corporate data and promoting sustainability in Bangladesh’s T&A sector, offering actionable recommendations for industry stakeholders, policymakers, and global brands committed to responsible sourcing.

Face recognition systems are widely used in various applications, from mobile banking apps to personal smartphones. However, these systems often store biometric templates in raw form, posing significant security and privacy risks. Pro-ID addresses this vulnerability by incorporating SimHash, an algorithm of Locality Sensitive Hashing (LSH), to create secure and irreversible hash codes of facial feature vectors. Unlike traditional methods that leave raw data exposed to potential breaches, SimHash transforms the feature space into high-dimensional hash codes, safeguarding user identity while preserving system functionality. 

The proposed system creates a balance between two aspects: security and the system’s performance. Additionally, the system is designed to resist common attacks, including brute force and template inversion, ensuring that even if the hashed templates are exposed, the original biometric data cannot be reconstructed.  

A key challenge addressed in this project is minimizing the trade-off between security and performance. Extensive evaluations demonstrate that the proposed method maintains competitive accuracy rates comparable to traditional face recognition systems while significantly enhancing security metrics such as irreversibility, unlinkability, and revocability. This innovative approach contributes to advancing the reliability and trustworthiness of biometric systems, providing a secure framework for applications in face recognition systems. 

Wavelength Division Multiplexing (WDM) is essential for managing rapid network traffic growth in fiber optic systems. Each WDM channel demands a narrow-linewidth, frequency-stabilized laser diode, leading to complexity and increased energy consumption. Multi-wavelength laser sources, generating optical frequency combs (OFC), offer an attractive solution, enabling a single laser diode to provide numerous equally spaced spectral lines for enhanced bandwidth efficiency.

Quantum-dot and quantum-dash OFCs provide phase-synchronized lines with low relative intensity noise (RIN), while Quantum Well (QW) OFCs offer higher power efficiency, but they have higher RIN in the low frequency region of up to 2 GHz. However, both quantum-dot/dash and QW based OFCs, individual spectral lines exhibit high phase noise, limiting coherent detection. Output power levels of these OFCs range between 1-20 mW where the power of each spectral line is typically less than -5 dBm. Due to this requirement, these OFCs require excessive optical amplification, also they possess relatively broad spectral linewidths of each spectral line, due to the inverse relationship between optical power and linewidth as per the Schawlow-Townes formula. This constraint hampers their applicability in coherent detection systems, highlighting a challenge for achieving high-performance optical communication.

In this work, coherent system application of a single-section Quantum-Well Fabry-Perot (FP) laser diode is demonstrated. This laser delivers over 120 mW optical power at the fiber pigtail with a mode spacing of 36.14 GHz. In an experimental setup, 20 spectral lines from a single laser transmitter carry 30 GBaud 16-QAM signals over 78.3 km single-mode fiber, achieving significant data transmission rates. With the potential to support a transmission capacity of 2.15 Tb/s (4.3 Tb/s for dual polarization) per transmitter, including Forward Error Correction (FEC) and maintenance overhead, it offers a promising solution for meeting the escalating demands of modern network traffic efficiently.

Users of theorem provers often desire to to extract their verified code to a  more efficient, compiled language. Coq's current extraction mechanism provides this facility but does not provide a formal guarantee that the extracted code has the same semantics as the logic it is extracted from. Providing such a guarantee requires a formal semantics for the target code. The CakeML project, plemented in HOL4, provides a formally defined syntax and semantics for a subset of SML and includes a proof-producing translator from higher-order logic to CakeML. We use the CakeML definition to develop certifying extractor to CakeML from Gallina using the translation and proof techniques of the HOL4 CakeML translator. We also address how differences between HOL4 (higher-order logic) and Coq calculus of constructions) effect the implementation details of the Coq translator.