Computer Engineering Commons^™

Enhancing Stem Education With Modeling, Simulation, And Ai Technologies: From Virtual Laboratories To Intelligent Teaching Assistants, Yiyang Li

Electrical & Computer Engineering Theses & Dissertations

Rapid advancements in modeling and simulation (M&S) and artificial intelligence (AI) present new opportunities to enhance various aspects of STEM education, from virtual laboratories that simulate physical lab environments in software to intelligent teaching assistants that provide on-demand, curriculum-aligned instructional support. Virtual laboratories offer a potential solution to the access and scalability challenges of laboratory courses by allowing students to conduct experiments without physical equipment or geographical constraints. AI-powered teaching assistants, particularly those grounded in course-specific materials, can help mitigate the instructional support gap that arises when students work independently in digital learning environments. This dissertation presents three-phase research into …

Physics-Guided Deep Learning For Predictive Modeling Of Spatiotemporal Dynamical Systems, Niharika Deshpande

Engineering Management & Systems Engineering Theses & Dissertations

Many physical and networked systems evolve under continuously changing spatial and temporal conditions. Transportation networks respond to fluctuating demand, atmospheric fields reorganize as storms intensify, and coastal response depends on localized forcing pathways. Modeling such systems requires learning formulations that adapt to evolving structure, operate on irregular geometries, and provide interpretable measures of predictive uncertainty. This dissertation develops a physics-guided spatiotemporal learning framework designed for structured dynamical systems whose governing interactions are neither static nor Euclidean. The central premise is that spatial relationships in these systems are dynamic and geometry-dependent. To represent this behavior, system states are modeled on time-varying …

Deep Multimodal Sensor Fusion For Autonomous Driving: A Real-World Deployment On Little Blue, Raphael T. Pamie-George

Electrical & Computer Engineering Theses & Dissertations

Accurate perception and control systems are essential for autonomous vehicles to safely interpret and navigate complex environments. Individual sensing modalities such as cameras, LiDAR, and inertial measurement units (IMUs) provide valuable environmental and vehicle state information; however, each sensor exhibits inherent limitations when used independently. Camera performance can degrade under varying lighting conditions, LiDAR measurements may be sparse or affected by environmental noise, and IMU measurements are susceptible to drift over time. Little Blue, a research autonomous vehicle platform from ODU (Old Dominion University), previously relied primarily on camera-based data for neural network training, limiting the system’s ability to exploit …

Hardware-In-The-Loop Evaluation Of Sensor-Source Selection For Prosthetic Locomotion Intent Recognition, Victoria Asencio-Clemens

Master's Theses

Active lower-limb prostheses use intent-recognition systems to identify a user’s locomotion mode and select an appropriate control strategy, but sensor configurations that perform well offline may be unsuitable for resource-constrained embedded hardware. Existing sensor-selection methods generally prioritize classification accuracy without directly accounting for processing latency, memory usage, or other hardware-dependent requirements. To address this limitation, this thesis develops a hardware-in-the-loop source-selection framework for embedded classification of level walking, ramp ascent, ramp descent, stair ascent, and stair descent using multimodal biomechanical data from transtibial amputee participants. Subject-specific linear support vector machine classifiers were evaluated using trial-held-out validation, and candidate configurations from …

Developing A Rapid Urban Forest Assessment System For Sustainable City Greenification, Daniel Gonzalez

Master's Theses

This thesis presents the Rapid Urban Forest Assessment (RUFA) system, a web-based platform that integrates urban tree inventories and aerial tree detection to assess forest health across California’s census-designated places. RUFA combines inventoried tree records with coordinates detected from high-resolution multispectral imagery using convolutional neural networks, then computes a composite RUFA Score from four metrics: canopy cover percentage, trees per capita, tree diversity (TD-50), and tree evenness. The thesis addresses two engineering challenges in building the dashboard: querying and aggregating over seven million tree records in real time, and rendering spatial summaries at multiple zoom levels without recomputing cluster assignments …

Can Kinematics Specify Tool Use Affordances?, Tyler Duffrin

All Dissertations

Object properties such as length, shape, and heaviness, as well as affordance-based properties such as strike-with-ability and poke-with-ability, can be perceived via dynamic touch as functions of inertia tensor invariants. The primary question addressed here was whether people can perceive these same power and precision properties for purely virtual objects, which lack dynamic physical forces. The Kinematic Specification of Dynamics (KSD) principle asserts that missing dynamics are lawfully specified by kinematic (visual) motion patterns, such that kinematic information alone may support successful interactions with virtual objects. In three experiments, we investigated (1) whether virtual reality (VR) users could calibrate to …

Secure And Compassionate Dementia Care: Non-Intrusive Remote Monitoring Of Falls, Wandering, And Agitation, Awan-Ur- Rahman

Master's Theses

Alzheimer’s disease and related dementias (ADRD) present significant safety challenges, as affected individuals may experience falls, wandering, agitation, and a progressive decline in independence. Although continuous monitoring can enable timely intervention, many existing systems rely on cameras or wearable devices, which may introduce concerns related to privacy, comfort, and sustained use. This thesis explores privacy-preserving and non-intrusive remote monitoring through two complementary studies. The first study presents an ambient Wi-Fi channel state information framework for recognizing agitation and eight daily activities. The proposed approach translates behavioral and physiological markers commonly captured by wearable sensors into the Wi-Fi sensing domain. The …

Optimizing Few-Shot Learning In Pruned Large Language Models With Task-Specific Prompts, Danyal Aftab

Dissertations

Few-shot learning enables large language models to efficiently perform tasks given only a limited number of labeled examples. However, training these models entirely from scratch requires substantial computational resources, making it challenging for many organizations to fully leverage their potential. This thesis explores how structured pruning, task-specific prompting, and parameter-efficient fine-tuning can be combined to preserve few-shot learning capabilities in compressed LLMs, while also extending their utility to real-world recommendation systems.

In this research, we propose the Tailored LLM framework, which first reduces model size through structured pruning and then enhances few-shot learning performance using carefully designed prompts. We experiment …

Towards More Realistic And Practical Graph Backdoor Attacks, Jiawei Chen

Computer Science Theses & Dissertations

Graph Neural Networks (GNNs) have demonstrated remarkable performance on graph-based learning tasks and are increasingly deployed in security-critical applications. However, recent studies have shown that they are highly vulnerable to graph backdoor attacks (GBAs), where adversaries implant malicious triggers to induce targeted misclassification during inference. Despite their effectiveness, existing GBAs are often developed under unrealistic assumptions, such as focusing exclusively on simple homogeneous graphs or assuming the adversary possesses privileged access to target nodes during inference. This dissertation aims to systematically investigate and design graph backdoor attacks under significantly more realistic graph settings and adversarial constraints.

First, we investigate the …

Genwriter 2.0: A Hybrid Case-Based And Llm Rewriting Approach For Mitigating Implicit Gender Cues In Text, Shweta Soundararajan, Sarah Jane Delany

Conference papers

Gendered language is the use of words or phrases that indicate an individual's gender. Although useful in some contexts, gendered language can reinforce gender stereotypes and introduce bias, particularly in machine learning models for tasks involving people such as recruitment or occupation classification. When textual content about individuals, such as biographies, contain gender cues, models can learn spurious associations between gender and other characteristics of individuals, such as profession, potentially resulting in unfair outcomes such as reduced hiring opportunities for women.

To address this challenge, we propose GenWriter 2.0, a hybrid approach that integrates Case-Based Reasoning (CBR) with Large …

Practical On-Device Intelligence For Resource-Constrained Embedded And Wearable Devices, Xiao Ma

Dissertations and Theses Collection (Open Access)

Embedded and wearable devices increasingly rely on deep neural networks to support intelligent sensing applications directly on the device. However, building practical on-device intelligence remains challenging throughout the deployment lifecycle. Before deployment, models must remain accurate despite severe memory, computation, and energy budgets, and a single device often needs to support multiple sensing applications without deploying a separate model for each application. After deployment, the device continues to operate in changing real-world environments, where shifts in data distribution can degrade the model’s performance over time, so the model must be able to adapt to such shifts efficiently.

The first challenge …

Understanding The Correlation Between Prediction Performance And Trust In Ai Through Scenario-Based Tasks, Likhitha Kammara

Theses and Dissertations

Trust in artificial intelligence is commonly assessed through self-reported scales or behavioral reliance, yet behavioral reliance is retrospective and can only be observed after a decision has already been made. This thesis examines whether prediction accuracy — a user's ability to predict what an AI system will recommend before its output is revealed — can serve as a prospective correlate of trust in the same empirical sense as behavioral reliance. The study was conducted in two phases using scenario-based AI decision tasks across disaster response, healthcare, and infrastructure restoration contexts, employing a between-group design in which participants either predicted AI …

Unsupervised Learning For Minimum Error Adaptive Sampling Of Atmospheric Vertical Temperature Profiles, Alejandro Medina

Open Access Theses & Dissertations

Uncrewed aerial systems (UAS) collect atmospheric data on fixed schedules, and endurance limits make blind searching costly. This thesis develops an unsupervised representation that summarizes a multi-decade radiosonde archive into a compact library of atmospheric states, giving a UAS an expectation of the column before it flies. The method standardizes both axes of a profile against the sounding's own surface conditions, which makes the representation independent of season and of station elevation. Applied to 27,270 soundings from Norman, Oklahoma, over the lowest 1.5 km of the atmosphere, 12 representative profiles reconstruct the record to within 1.0 °C of mean absolute …

Improving Collective Aggregation Strategies For Hpc And Ai Workloads, Mikaila J. Gossman

All Dissertations

Modern high-performance computing (HPC) and artificial intelligence (AI) workloads increasingly generate data at rates that outpace the storage systems responsible for persisting it. While computational throughput has scaled dramatically through parallel processors, GPUs, and specialized accelerators, storage systems have improved far more slowly. As such, application performance is increasingly limited by the ability to move data efficiently through deep, heterogeneous storage hierarchies rather than by raw computational capability. This dissertation investigates how aggregation and data movement strategies must be redesigned to close this gap, addressing four central challenges: resource contention between foreground computation and background I/O, the fine-grained and irregular …

Evaluating Machine Learning Models On Classification Of Novel Cyber Attacks In The Healthcare Domain, Promise Ehimen

Dissertations, Theses, and Projects

The increasing adoption of the Internet of Medical Things (IoMT) has improved healthcare delivery through connected medical devices while simultaneously expanding the cybersecurity risks facing healthcare organizations. Although machine learning based intrusion detection systems have demonstrated high detection accuracy, their ability to respond reliably to previously unseen cyberattacks remains uncertain. This study investigated how a Neural Network model and a Logistic Regression model classified novel cyberattacks within the IoMT environment. The Neural Network and Logistic Regression models were both trained and tested using a subset of the CICIoMT2024 benchmark dataset. The Neural Network achieved 99.82% test accuracy and a 0.94 …

Novel Dynamic Batch-Sensitive Adam Optimiser For Vehicular Accident Injury Severity Prediction, Daniel Asare Kyei, Alimatu Saadia-Yussiff, Maame G. Asante-Mensah, Abdul Lateef-Yussiff, Charles Roland Haruna, Derry Emmanuel

Iraqi Journal for Computer Science and Mathematics

The choice of optimiser is important in deep learning, as it strongly influences model efficiency and speed of convergence. However, many commonly used optimisers encounter difficulties when applied to imbalanced and sequential datasets, limiting their ability to capture patterns of minority classes. In this study, we propose Dynamic Batch-Sensitive Adam (DBS-Adam), an optimiser that dynamically scales the learning rate using a batch difficulty score derived from exponential moving averages of gradient norms and batch loss. DBS-Adam improves training stability and accelerates convergence by increasing updates for difficult batches and reducing them for easier ones. We evaluate DBS-Adam by integrating it …

Genpix: A Diverse Dataset For Fake Image Detection, Guessoum Dalila, Benblidia Nadjia, Boumahdi Fatima, Remmide Mohamed Abdelkarim, Nouri Tarek-Amine, Bataouche Azeddine-Lotfi

Iraqi Journal for Computer Science and Mathematics

The rapid advancement of sophisticated generative models has intensified the need for robust fake image detection systems. However, many existing benchmark datasets suffer from limited diversity in content types and generation techniques, constraining the generalization ability of detection models. To address these limitations, we introduce GenPix (Generalized Pixels), a comprehensive dataset encompassing over 80,000 images spanning diverse categories, including faces, objects, and scenes, generated by multiple state-of-the-art models such as Generative Adversarial Networks (GANs) and diffusion-based architectures. The dataset includes samples from different generation methods to ensure broad coverage of fake image characteristics.

GenPix provides a realistic evaluation environment that …

Eyolov8-Msaff: An Enhanced Object Detection Algorithm With Multi-Scale Attention And Feature Fusion For Small And Dense Object Detection In Uav Applications, Mohammed Hasan Mutar, Morteza Valizadeh, Alaa Hussein Abdulaal, Mehdi Chehel Amirani

Iraqi Journal for Computer Science and Mathematics

The proliferation of unmanned aerial vehicles (UAVs) has necessitated the development of sophisticated object detection algorithms capable of handling the unique challenges posed by aerial imagery. Traditional detection methods often struggle with small object sizes, dense distributions, and complex backgrounds characteristic of UAV-captured scenes. This research presents EYOLOv8-MSAFF (Enhanced YOLOv8 with Multi-Scale Attention and Feature Fusion), a novel deep learning architecture specifically engineered for superior performance in UAV-based object detection tasks. The proposed methodology integrates four innovative components: a Hybrid Spatial-Channel Attention Mechanism (HSCAM) that processes attention information in parallel rather than sequentially, an Adaptive Multi-Scale Feature Fusion Module (AMSFFM) …

A Novel Hierarchical Multi-Scale Attention Mechanism For Enhanced Yolov8 Performance In Unmanned Aerial Vehicle-Based Small Object Detection, Mohammed Hasan Mutar, Morteza Valizadeh, Alaa Hussein Abdulaal, Mehdi Chehel Amirani

Iraqi Journal for Computer Science and Mathematics

Object detection in Unmanned Aerial Vehicle (UAV) images presents significant challenges due to the prevalence of small and densely packed objects, as well as variations in scale, orientation, and lighting conditions. This paper introduces a novel object detection algorithm, Hierarchical Multi-Scale Attention YOLO (HMSA-YOLO), which is an improved version of YOLOv8 designed to address these challenges. The proposed method incorporates a novel Hierarchical Multi-Scale Attention (HMSA) module, a Bidirectional Feature Pyramid Network (BiFPN) for enhanced feature fusion, a modified loss function, and an adaptive anchor optimization technique. The HMSA module effectively captures both channel and spatial dependencies at multiple scales, …

Improving The Fpga Radio Design Cycle: Implementing Dvb-S2 Modulation Using Verilator And Gnu Radio, Seth Pellegrino

Dissertations and Theses

In embedded digital radio systems, a central challenge is meeting the real-time throughput required to process the samples--especially for space-bound ultra-wideband SDRs which must handle more than 100 Gbps entirely onboard. An FPGA's programmable logic offers sufficient potential, but realizing a particular radio flow is a project usually fraught with defects and long turnaround times. We simulated Verilog modules in a custom harness that adapted a Verilated model to GNU Radio, which allowed for breaking down a complicated radio flow (here, DVB-S2 modulation) into a series of well-bounded problems each with clear criteria for success. The framework, built on open …

Rthermal: Gate Level Power And Thermal Simulation For 3d-Stacked Chips, Peter Xiong

Master's Theses

As the number of transistors in modern processors increases, heat dissipation has become a major bottleneck to scalability. The use of 3D stacking further intensifies this problem, as heat from multiple layers can accumulate vertically. These challenges create a growing need for tools that can accurately and efficiently simulate the thermal behavior of 3D chips during design and validation. Several existing tools model thermal behavior for 3D-stacked chips and can simulate average heat over large spatial regions or long time intervals. However, when heat is concentrated in a small area or over a short time window, such models can miss …

Tinyml-Based Embedded Vision System For Ic Detection In Microcontroller Manufacturing, Mark M. Pallones, King Harold A. Recto, Rynne Daven A. Barrios

Electronics, Computer, and Communications Engineering Faculty Publications

Mixing of microcontroller unit (MCU) integrated circuits (ICs) during the final testing stage of semiconductor manufacturing can lead to material waste, production delays, and customer dissatisfaction. This issue often occurs when standard JEDEC Matrix Trays (JMTs) are reused without confirming that all ICs have been removed after testing, a process typically performed through manual inspection and therefore susceptible to human error due to high test volumes, small IC package sizes, and visual similarity between IC packages and tray surfaces. This study develops an automated IC Detection Test System using embedded vision to determine whether JMT trays are empty prior to …

Moral: Multimodal Reasoning For Autonomous Language Models With Sensor-Grounded Spatial Bev Rendering, Ambarish Govindarajulu Kaliamurthi

Master's Theses

Autonomous-driving vision-language models describe scenes fluently but reason poorly about metric, safety-critical spatial relationships because they do not read sensor geometry in a grounded way. This thesis presents MoRAL (Multimodal Reasoning for Autonomous Language Models), a two-stage fine-tuning pipeline that teaches a compact 2-billion-parameter VLM to decode a physics-encoded Bird’s Eye View (BEV) representation – LiDAR distance as color, object class as cluster shape, radar Doppler velocity as directional wedges – and then trains it to reason over that representation for driving decisions. Stage 2 then fine-tunes on 57,696 teacher-generated chain-of-thought examples across eight question types, using Cosmos-Reason2-8B as teacher …

The Zeal Instruction Set Architecture, Joseph A. Gerani

Master's Theses

The Instruction Set Architecture of a CPU (Central Processing Unit) determines what type of instructions the CPU is able to understand, how those instructions are encoded, and what it should output upon receiving those instructions as input. There are currently three popular ISAs meant for the consumer market: x86, RISC-V, and ARM, as well as a fourth that mostly now exists in the server market by the name of Power. One of the most important parts of an ISA is for engineers to be able to understand it and make use of it. If an ISA is too complicated, nobody …

Data Augmentation For Vision-Language-Action Models: Bridging Vision And Language, Miaosen Zhou

Master's Theses

This thesis focuses on real-time task execution and object detection for autonomous robots through dataset augmentation. We propose a data augmentation approach to address dataset imbalance in Vision-Language-Action (VLA) models across both image and text modalities during the fine-tuning process. The proposed method takes an image as input and generates a structured textual description using a prompt engineering strategy to augment the textual input. The generated augmented text includes key elements such as the task goal, scene description, reasoning, and execution plan, along with other relevant contextual information. This enriched representation improves the quality of the training data and supports …

Towards Neural Network Optimization: Addressing Issues With Corrupted Weights Within Models, Nick Najafizadeh

Master's Theses

Neural networks are a recent popular technology inspired from human brains. Much of their popularity arises from how they excel in reasoning and logic, and are generally rather efficient in their tasks. With those strengths, they are frequently used in transportation and business among many other fields. However, neural networks have many factors that can deteriorate their performance, one of the most critical being weight corruption. Therefore, it is of utmost importance to detect and handle them as soon as possible so as to minimize the negative impact on a network’s performance. The optimization of neural networks would be especially …

Pipeline Optimization Of Agentic Llms For Text-To-Sql, Peter Conant

Master's Theses

Current database interfaces limit users’ interaction to those with technical skills, creating timely roadblocks for non-technical professionals. Text-to-SQL aims to simplify database interactions by translating natural language questions into database queries, but long-standing challenges like question understanding, question-schema linking, and SQL generation have held the field back. In the AI era, foundational LLMs prove to be very capable of question understanding SQL, perform well in Schema Linking, Generation, and Evaluation tasks. However the cost to run these model is a hurdle for many organization with low funds and resources. Text-to-SQL solutions often operate across large enterprise size databases with, and …

Early Failure Detection In Web Navigation Agents Via Closed Sequential Pattern Mining, Sergio Talavera

Master's Theses

LLM-based web navigation agents fail on the majority of tasks while consuming substantial computational resources before failure becomes apparent. This thesis investigates whether closed sequential pattern mining on the first K steps of agent execution traces can predict task failure early enough to enable meaningful computational savings with interpretable justification. We develop a two-phase system: an offline pipeline that symbolizes agent traces, extracts K-step prefixes, mines closed patterns via BIDE+, and ranks them by failure precision; and an online detector that matches live executions against the resulting pattern library. We evaluate on 1,544 MiniWoB++ traces across three open-weight language models …

Adaptive Task-Driven Lidar Point Cloud Compression For Autonomous Driving, Su Hyun Kim

Master's Theses

Autonomous-driving systems generate large LiDAR point clouds, but compression can damage the sparse object-support structure needed by 3D detectors even when reconstructions appear visually plausible. This thesis asks whether adaptive LiDAR compression can preserve downstream detection better than uniform compression by allocating more fidelity to detector-relevant regions. The main study builds a mask-aware range-image codec with an encoder-decoder bottleneck, an importance head, and an adaptive quantization variant. It compares this adaptive variant package with a confirmed masked uniform baseline under one fixed RangeDet evaluation surface and one fixed KITTI validation subset. Two supporting studies bound the result: a projection-reconstruction PointPillars …

Developing A Natural Language Interface For Knowledge Graphs, Ruth Assefa, Sarah Mendoza, Luke Voinov, Oyku Serap Ogut, Nurcan Yuruk

SMU Journal of Undergraduate Research

This paper proposes to solve the challenge of making databases more user-friendly by interfacing them with OpenAI's ChatGPT-3.5 model. We implemented this solution to assist researchers in easily finding others with similar research interests. Our study involves 184 researchers from 14 departments at Southern Methodist University (SMU). We collected researchers' areas of expertise and biographies and stored them in a Neo4j graph database. We used OpenAI's embedding models to create vector representations of the collected data, allowing for accurate similarity assessments via Neo4j's built-in algorithms. By integrating this system with LangChain, we enabled natural language queries. The results demonstrated high …