Computer Engineering Commons^™

Learning Global Context For Sparse Activity Recognition In Lengthy Recordings With Limited Dataset Size, Zeyu Tang

All Dissertations

This dissertation describes methods to analyze lengthy recordings of data in order to detect sparsely occurring activities. The narrative below describes the progression of research that led to the development of these methods and their generalization into a unified framework. My research started with designing models for dietary monitoring, including detecting meals from day-long recordings and detecting intake gestures from meal-length recordings. Both tasks share some common characteristics: (a) the target event takes only a small portion of data recordings, and (b) there is global context within full-length data recordings that can help a model make better decisions. After finishing …

Design Of An Extensible And Scalable Data Acquisition System For Pulse Shape Discrimination, Prince John

McKelvey School of Engineering Graduate Student Theses & Dissertations

This thesis presents the design and development of a highly scalable, end-to-end data acquisition (DAQ) system for nuclear physics experiments that can be deployed in configurations ranging from a few to thousands of detector channels. The system is built as an extensible platform composed of modular 16-channel chipboards that support a wide range of scintillator and detector types and perform real-time, on-board data sparsification and pulse-shape processing. Three versions of the chipboard have been fabricated to date.

The DAQ architecture is based on a family of analog pulse-shape-processing application- specific integrated circuits (ASICs) developed by the IC Design Laboratory at …

Conditioning Hierarchical Diffusion Transformers On Rf Circuit Parameters, Ethan Morton

McKelvey School of Engineering Graduate Student Theses & Dissertations

As modern ML techniques have become increasingly advanced, they have begun to be integrated into wireless RF systems for classification, identification, and spectrum management. Deep Neural Networks (DNNs) enable RF system operators and designers to design more flexible systems with greater robustness to errors and attacks. However, neural networks require significant amounts of properly annotated data to train. Current data labeling methods lack the ability to obtain reliable true labels for circuit properties such as carrier frequency offset (CFO), power amplifier (PA) non-linearity, and in-phase/quadrature (IQ) imbalance. This thesis investigates the efficacy of a novel architecture, RF-Diffusion, for generating high-quality …

Temporal Logic Planning In Semantic Maps Of Unknown Environments Using Tl-Rrt, Dongrui Yang

McKelvey School of Engineering Graduate Student Theses & Dissertations

Autonomous mobile robots are increasingly expected to perform complex missions in unstructured environments. Traditional path planning approaches handle simple point-to-point navigation, but struggle with complex tasks that involve temporal and logical orderings of objectives. Linear Temporal Logic (LTL) provides a method for complex missions (e.g., sequential visits to multiple targets or surveillance tasks) in a strict way. This thesis presents an integrated planning framework that enables a robot to satisfy LTL-based task specifications in an unknown environment by combining a Temporal Logic RRT* (TL-RRT*) planner with semantic mapping. The robot builds a semantic map of its environment online using simultaneous …

Database State Reconstruction And Audit Log Reconciliation For Data Systems Security, Mahfuzul I. Nissan

LSU New Orleans Theses and Dissertations

Modern organizations rely on diverse database management systems to store critical data across heterogeneous storage architectures. These platforms complicate forensic investigations because database engines vary profoundly in their layouts, encodings, and update policies. Malicious insiders can exploit this architectural opacity to alter data and disable audit logging undetected, leaving investigators with an untrusted view of database activity. Furthermore, building custom forensic solutions for every database version is fundamentally unscalable.

This dissertation advances data-system security through a tamper-aware, cross-layer framework that reconstructs database states and detects unauthorized operations even when audit logs are manipulated. By unifying the analysis of physical storage …

Robust Hardware Trojan Detection Leveraging Dual‑Domain Features And Stacked Ensemble Learning, Sefatun-Noor Puspa

All Theses

In Cyber-physical systems rely on sensors, communication, and computing, all powered by integrated circuits (ICs). These ICs are vulnerable to malicious hardware attacks, with hardware Trojans being one of the stealthiest threats. Trojans are malicious implants in the circuitry, which are often inserted during design or fabrication stages. This stealthy addition remains dormant until triggered and might cause functional disruptions or sensitive information leakage once triggered. Traditional IC validation methods, such as functional testing and logic analysis, usually fail to capture these subtle anomalies because hardware Trojans are intentionally designed to mimic normal circuit behavior. They often remain dormant under …

Multimodal Aerial Image-Based Ground Object Detection And Classification Using Yolo, Erik Nielsen

All Theses

Object detection in unmanned aerial vehicles (UAVs) present a unique challenge due to small object sizes, varying viewpoints, and changing environmental conditions. These challenges are exacerbated when operating during daytime and nighttime scenarios where illumination differences can heavily impact detection performance. This work is motivated by military object detection applications where the ability to reliably identify small objects such as landmines or unexploded ordnance from aerial imagery presents a critical safety need and a significant technical challenge. In this paper, we investigate object detection using both visible (RGB) and infrared (IR) imagery to improve robustness and reliability across diverse operating …

Exploration Of Real-Time Power Electronic Simulation On Amd Npu, Shouyu Du

All Theses

The rapid electrification of the automotive and data center sectors has created a critical demand for high-fidelity, real-time simulation of complex power electronic systems. While Hardware- in-the-Loop (HIL) simulation on Field-Programmable Gate Arrays (FPGAs) is the current industry standard, it presents significant challenges regarding development complexity and memory limita- tions. This dissertation investigates the feasibility of utilizing Neural Processing Units (NPUs)— specifically the AMD AI Engine (AIE) XDNA2 architecture—as a novel platform for real-time, deterministic circuit simulation. This study establishes a comprehensive automated modeling framework based on graph theory and Massarini’s method to derive linear state-space representations for arbitrary circuit …

Moodify: A Mood-Based Music Recommendation System, Meghana Kagitha

Theses and Dissertations

Music has long been recognised as a powerful tool for emotional regulation, yet existing music streaming platforms often fail to align song recommendations with a user's current emotional state. Moodify is a mood-based music recommendation system designed to bridge this gap by delivering personalised playlists that reflect how a user feels in real time.

This project presents the design, development, and evaluation of Moodify, a mobile application that leverages the Circumplex Model of Emotion to capture user mood through an intuitive two-dimensional valence-arousal interface. Rather than relying on text input or manual search, users plot their emotional state directly onto …

Hdra-Fusion: Hybrid Detection With Routed Architecture For Manipulation-Aware Ai Face Forgery Detection, Omar Ebeid

Theses and Dissertations

With the rapid advancements in artificial intelligence-based image generation and manipulation tools, it is extremely difficult to detect if an image is genuine or artificially crafted. Despite extensive research in this area, existing image detection systems suffer from three major problems: suboptimal cross-dataset generalization due to shortcut learning of dataset-specific patterns, unreliable probability estimates due to domain shift, particularly in cross-manipulation evaluation settings, and an inability to detect images manipulated by multiple types of manipulations within a single detection framework. To address these limitations, we propose HDRA-Fusion (Hybrid Detection with Routed Architecture), a framework built on the conclusion that different …

A Webcam Eye Tracking Infrastructure For Software Engineering Tasks, Zachary M. Kozak

School of Computing: Dissertations, Theses, and Student Research

Performing eye tracking utilizing commodity webcams has been explored for over a decade, but limited camera quality and sensitivity to head movements have hindered its adoption in research settings. Recent advances in consumer-grade webcams and machine learning methods present an opportunity to improve the accuracy of webcam eye tracking and extend the feasibility of studies beyond controlled laboratory environments.

Current popular webcam eye tracking methods restrict implementations to the browser and rely on continuous user interactions for calibration, limiting the kinds of studies that can be conducted. This thesis presents a feature-based gaze prediction system that incorporates eye geometry and …

Reward Representation Learning, Gregory M. Hyde

Dartmouth College Ph.D Dissertations

The \emph{Markov decision process} (MDP) has long served as the canonical model for sequential decision-making. However, it assumes that the reward function is Markov with respect to a given state representation---an assumption that often does not hold in practice. Instead, agents typically only perceive streams of observations and actions and must infer the latent structure according to which reward unfolds over time. From this perspective, reward prediction is initially non-Markov, reflecting a mismatch between the agent's current representation and the underlying structure of the environment.

In this thesis, we advance the view that reward is not simply a signal to …

An Investigation Of Data Granularity In Rag Pipelines For Personalized Medicine, Paritosh Pandey

Master's Theses

Generative AI, exemplified by large language models like the OpenAI GPT and Meta LLaMA families, can produce diverse content in response to prompts. This capability offers a promising solution to challenges in precision medicine, which seeks to tailor treatments to individual clinical profiles but often struggles with data collection, cost, and privacy concerns. By generating realistic, privacy-preserving patient data, generative AI has the potential to transform patient-centric healthcare. With such motivation, this research develops a comprehensive Generative AI pipeline emphasizing data granularity for accurate prediction of personalized treatments. The pipeline features a central Large Language Model interacting with a Machine …

Design And Evaluation Of A Single-Gate Multi-Threshold Null Convention Logic Architecture For Area Efficiency, John Edward Swaim

Graduate Theses and Dissertations

Asynchronous circuit design paradigms, such as NULL convention logic (NCL) and Multi-Threshold NCL (MTNCL), offer increased energy efficiency over synchronous equivalents and robust pipelines with minimal timing analysis. However, their dependency on dual-rail signal encoding causes significant overhead compared to single-rail equivalents. Previous single-gate and single-rail NCL paradigms have sought to reduce circuit area, but most compromise their quasi-delay-insensitivity and correct-by-construction nature with logic gate and system-level design choices. This thesis presents Single-Gate MTNCL (SG-MTNCL), a register controlled, single-gate, and dual-rail asynchronous architecture to improve the area efficiency of MTNCL without compromising the reliability of previous paradigms. The benefits of …

Faster Than The Speed Of Bram: In-Memory Computing For Next Generation Fpgas On The Edge, Nathaniel Joseph Fredricks

Graduate Theses and Dissertations

Traditional computer systems are hitting the Memory Wall as machine learning applications are bottlenecked by the bandwidth between separate memory and compute units. Current FPGA architectures are able to bypass this bottleneck with block RAMs (BRAMs) that provide on-chip, in-fabric storage. However, their potential as the foundation of computing components is often overlooked; machine learning accelerators implemented on FPGAs face additional delays when transferring data between memory and compute units. These penalties arise from BRAM bandwidth limitations and movement of data through the reconfigurable fabric. To support FPGA-based accelerators on the edge and break the Memory Wall, reconfigurable architectures must …

Automated Circuit Design Algorithm And Implementation For Asynchronous Reset (Ares) Physically Unclonable Functions, Andrew Michael Felder

Graduate Theses and Dissertations

Technology is increasingly interwoven in all aspects of society as it solves new problems, creates new possibilities, and enables new conveniences. With its unceasing evolution, environments handling protected information such as military, finance, and medicine demand ever-evolving threat mitigation. Combating attackers’ abilities to spoof devices, uncover cryptographic keys, and bypass security features is a never-ending task which drives technological advancement to improve existing capabilities and develop entirely new methodologies. In this dissertation work, the hybrid Asynchronous RESet Physically Unclonable Function (ARES PUF) is evaluated at the circuit to determine its merits as a PUF. As part of this evaluation, results …

A Low-Cost Motion Classification System For A Stuffed Animal Using An Imu And Machine Learning, Rachel N. Guynes

Honors Theses

One of the many fields that has seen the integration of robots is therapy. Zoomorphic robots (ZR) are designed to look and behave like animals to assist in Animal Assisted Therapy (AAT) practices. Studies show that ZRs can provide benefits similar to working with an actual animal; however, their high cost limits their accessibility. This thesis documents the process of building a real-time, low-cost motion classification system that can be attached to a stuffed animal to make it more interactive. Using a Random Forest (RF) classifier, the system identifies movements with approximately 81.67% accuracy.

Using Ali-4 Z-Implication In Controller Design, Shamil Azer Ahmadov

Chemical Technology, Control and Management

One of the widely used theories in the information processing is Professor Zadeh's fuzzy logic theory. Fuzzy implications form the basis of this theory. When processing information using fuzzy implication, the chosen judgment method and the type of implication affect the result. Referring to the review of the relevant literature on fuzzy implications, it can be noted that there are still unresolved problems and issues. For example, fuzzy implications cannot be used in processing imperfect information or information based on probabilistic and fuzzy uncertainty. Existing fuzzy implications face limitations in practical applications. Fuzzy implications only take into account inaccuracy, but …

Conceptual Model For Protecting Personal Data By Depersonalization In Information Systems: Principles, Components, And Life Cycle, Zarina Ildarovna Azizova

Chemical Technology, Control and Management

This article presents a systematic approach to personal data protection through depersonalization in the context of regulatory pressure and growing cyber threats. It proposes a comprehensive conceptual model that formalizes the de-identification process as a manageable sequence of steps, from attribute classification and method selection to mandatory verification of the result. The article also provides a comparative analysis of existing depersonalization methods in terms of their applicability within the proposed model. The model serves as a basis for the development of specific algorithms, as demonstrated by the example of a data shuffling approach.

Wavetable Modification In Frequency Domain: Fourier Analysis With Matlab Implementation, Caleb Kortens

Senior Honors Theses

One of the challenges in creating computer generated music is producing life-like sounds. Music produced through computer synthesis can often sound thin and synthetic rather than vibrant and energetic. Wavetable synthesis is a method of waveform generation that uses a wavetable, which is a set of single period waves, or frames, that have varying characteristics. This method allows a synthesizer to transition between waveforms to produce a sound with time varying tone and harmonic characteristics. This adds life and movement to computer generated sounds. Wavetable synthesis is often used to imitate actual instruments, but unique wavetables can be created that …