Digital Commons Network^™

Softmax Expressiveness With Linear Complexity By Investigating The Linear-Softmax Attention Accuracy Gap, Gabriel Mongaras, Eric Larson

Computer Science and Engineering Theses and Dissertations

Softmax attention Vaswani et al. [2017] Bahdanau et al. [2015] has gained traction as one of the most important parts of most modern machine learning models. Despite being the most adopted modern machine learning algorithm, exploration of softmax attention itself has been neglected. While softmax attention is a very powerful architectural component, it has a major drawback of being quadratic in sequence length. More specifically, softmax attention has quadratic complexity during training and linear complexity (when using a KV cache) during inference, which limits the sequence length it can model due to system memory constraints. Linear attention was proposed, (proposed …

Insimul: A Hybrid Ai System For Social Immersion, Daniel Dekerlegand

LSU New Orleans Theses and Dissertations

Creating believable, socially intelligent virtual characters that are culturally and historically grounded requires more than any single artificial intelligence paradigm can provide. Symbolic social physics systems such as Ensemble offer interpretable, human-authored representations of social norms and relationships but produce linguistically stilted interaction. Statistical approaches---particularly large language models---generate fluent dialogue but lack behavioral coherence and cultural grounding. Virtual reality provides embodied presence that transforms abstract social computation into experiential engagement but contributes no social intelligence of its own. This dissertation argues that the integration of all three technologies---symbolic AI, statistical AI, and virtual reality---is necessary for producing socially coherent, linguistically …

Data Driven Monitoring And Control Of Laser Powder Bed Fusion Process, Jose De Jesus Galarza

Theses and Dissertations

The Laser Powder Bed Fusion Process (LPBF) has been one of the main processes of additive manufacturing, enabling the manufacturing of complex geometries, customization, and lightweight parts. Modern LPBF processes have integrated monitoring systems that capture the light emissions per layer for quality assurance. However, standard defect detection algorithms have not yet achieved the high precision required due to the inherently variable nature of the signal, insufficient data for model training, and the confounding effects of the print.

The processes still have some challenges, such as characterizing the roughness from the build parameters alone, improving the pore detection using the …

A Machine Learning Based Analysis Of Traffic Accident Severity Using Environmental, Temporal And Road Condition Factors, Khalid Alaleeli

Theses

Traffic accidents on the road have been one of the most significant global issues as far as public safety is concerned, with the severity of crashes, as well as their fatalities, representing an undue burden on human and economic damages. The research constructs a machine learning-based model to examine and forecast the level of traffic accidents based on environmental and time, road and human variables. There were 798 valid observations and 14 variables discussing a dataset using the exploratory data analysis tool, supervised classification models (Decision Tree and Random Forest), and unsupervised K-Means clustering. The technique of feature engineering was …

Computer Vision And Machine Learning Approaches For Defect Detection In 3d-Printed Cementitious Materials: A Systematic Review, Muhammad Ali Musarat, Ruben Paul Borg, Jingjie Wei, Carl James Debono, Kamal Khayat

Civil, Architectural and Environmental Engineering Faculty Research & Creative Works

3D printing is evolving at a fast pace in both the manufacturing and construction sectors. These advancements can greatly benefit these industries. However, the 3D printing of concrete structures presents some challenges due to defects in the 3D concrete printed elements. Hence, this study systematically reviews Artificial Intelligence (AI)-driven techniques, such as Computer Vision and Machine Learning, to identify surface defects that can occur in 3D-printed cementitious material structures. The adopted methodology was the PRISMA statement with the aim of reporting the systematic review and meta-analysis. Two well-known databases, Web of Science and Scopus, were utilised for data extraction of …

Unified Deep Learning Techniques For Spatial Detection And Temporal Forecasting Across Visual Domains, John Olawale Olamofe

All Dissertations

This dissertation proposed a unified deep learning framework for spatial detection and temporal forecasting across visual domains, designed to address limited supervision, class imbalance, and scale variability. The framework was structured around four complementary principles: Domain-Aware Input Rebalancing, Representation-Centric Learning, Diversity-Driven Robustness, and Transfer Across Scale and Modality, which together enabled robust visual representation learning across heterogeneous data sources.

Domain-Aware Input Rebalancing mitigates non-uniform and sparse data distributions by actively reshaping inputs before learning via class-aware augmentation, sampling, resolution manipulation, and super-resolution. This principle underpins object detection in overhead satellite imagery (xView), dense urban aerial scenes (CADOT), temporally sparse NDVI …

Prediabetes Prediction Before Disease Onset Using Multimodal Health Data: A Machine Learning Approach, Luisa Veronica Gracia Mazuca

Open Access Theses & Dissertations

Prediabetes is a critical health condition that increases the risk of developing type 2 diabetes. The hemoglobin A1c (HbA1c) test diagnoses patients with prediabetes, but the disease has already caused metabolic alterations. Early detection is essential for timely interventions, and machine learning models offer a promising approach to identify prediabetic individuals through the analysis of biomarkers such as cytokines. We compared four classifiers-logistic regression, decision tree, random forest, and k-nearest neighbors - using cytokines (TNF-α, MCP-1, IL-1β, IL-6, IFN-γ), age, BMI, and waist-to-hip ratio (WHR). Models were evaluated using stratified 5-fold cross-validation and ROC-AUC. K-Nearest Neighbors (k-NN) achieved the highest …

Understanding Machine Learning Model Behavior Under Fairness And Privacy Constraints, David Anthony Sanchez

Open Access Theses & Dissertations

Machine learning systems deployed in high-stakes domains are increasingly expected to satisfy demands beyond predictive accuracy-including fairness across demographic groups, protection of sensitive information, and explanations that human stakeholders can inspect and trust. This thesis investigates how those demands can be met through learning frameworks that explicitly govern the relationship between data and models, arguing that trustworthiness is a design problem rather than a post hoc correction. The thesis is organized around three studies, each targeting a distinct point of data-facing control. The first develops CondFairGen, a fairness-aware conditional generator for tabular data that improves subgroup equity by dynamically reweighting …

Predicting Flight Fares With Machine Learning: Enhancing Aviation Industry Pricing Forecasting, Reem Almulla

Theses

This research investigates the increasing challenge of accurate flight fare prediction for travel agencies that are functioning and working in the post-pandemic aviation market. As the prices fluctuate while demand is unstable and competition pressure increases, therefore it influences decision-making and profitability. In most cases, traditional ticket pricing methods often can be ineffective when capturing complex and non-linear relationships within factors that influence the ticket fare dynamics. As a result, highlighting the urge of more adaptive pricing and data-driven predictive machine learning models. In response to this challenge, the research examines the effectiveness of machine learning techniques for enhancing flight …

Hybrid Machine Learning For Zero-Day Malware Detection: An Adaptive Static-Dynamic Analysis Approach, Garrett Farmer

Theses and Dissertations

In an era of swiftly evolving cyber threats, zero-day malware continues to be one of the most challenging classes of attacks to detect and mitigate. Traditional signature-based methods often fail to detect novel malicious code, leaving institutions vulnerable to unknown exploits. This thesis proposes a machine learning (ML)-based framework that is designed to detect unknown malware variants. By combining both static and dynamic techniques, such as file structure exam ination and sandbox-based runtime analysis, this approach aims to successfully capture malicious characteristics. The proposed custom pipeline addresses the computational overhead that is associ ated with deep inspections, outlining a staged …

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 …

Machine-Learning Landslide Susceptibility And Runout Modeling In The Nolichucky River Gorge After Hurricane Helene, Grace Braver

Electronic Theses and Dissertations

Extreme rainfall from Hurricane Helene (September 2024) triggered widespread landslides across the southern Appalachian region, highlighting the need for rapid landslide susceptibility assessments that capture both landslide initiation and downstream runout. Traditional susceptibility models often focus solely on initiation zones, limiting their ability to identify which slopes will generate destructive landslides or where material will travel. This study addresses that gap by (1) integrating Geographic Information System (GIS)-based machine learning susceptibility modeling using ArcGIS Pro: Maximum Entropy (MaxEnt) and Random Forest-Based and Boosted Classification and Regression (FBBC) and (2) the U.S. Geological Survey (USGS) Grfin (Growth, Flow, and Inundation) runout …

Artificial Intelligence In Medicine: Barriers, Solutions, And Strategies, Anil Harrison, Melissa Stradley Moreno, Caroline E. Williams, Munevver Mine Subasi, Ersoy Subasi

HCA Healthcare Journal of Medicine

The integration of artificial intelligence (AI) and machine learning (ML) into health care holds the potential to revolutionize patient care by enhancing clinical decision-making, improving diagnostic accuracy, and reducing costs. Despite this promise, adoption remains limited due to a range of technical, regulatory, educational, and cultural barriers. This paper examines these challenges and proposes strategies to support safe and effective implementation of AI in clinical practice.

Key barriers include the lack of model interpretability, often referred to as the "black box" problem, which undermines clinician trust and accountability in clinical settings, evolving regulatory frameworks and unresolved questions surrounding liability, and …

Dementia Detection In Low-Resource Languages: Evaluating Translation-Assisted Transfer Learning For Multilingual Clinical Assessment, Kylar A. Deloach

Honors Theses

Alzheimer's disease (AD) is a growing global health concern, with millions of people affected worldwide and cases expected to rise significantly in the coming decades. Early detection is critical for patient treatment and care, and recent advances in natural language processing (NLP) have shown promise in identifying linguistic markers associated with AD. However, most existing work has focused on English, leaving speakers of other languages with limited access to such tools. This study investigates how effective AD detection models trained on English data are at transferring to Greek, a low-resource language with limited dementia-related speech data available. We propose a …

Increasing The Prediction Accuracy Of Plant Oil Production Processes Through Adjusting The Parameters Of Machine Learning Models, Umidjon Ruziev, M.K. Shodiev, A.T. Rajabov

Chemical Technology, Control and Management

Vegetable oil production is characterized by high variability in output indicators due to nonlinear interactions between raw material parameters, equipment modes, and heat and mass transfer conditions. Existing approaches to applying machine learning in this field, as a rule, do not account for the impact of hyperparameter adjustments on forecasting quality across specific technological stages. The article presents a systematic methodology for adjusting model parameters (Ridge regression, SVR, GBM, LSTM) applied to three key tasks: predicting residual oil content in oil cake, color index during bleaching, and free fatty acid content during deodorization. In a set of 1000 observations, including …

Automatic Modulation Classification On Satellite Communications Across Orbital Domains, Lorraine (Raina) Freeman

Theses

In recent years, the space domain has exploded with both commercial and government activity, making situational awareness critically important. In this work, we highlight the need for signal intelligence capabilities that can transfer gracefully from terrestrial systems to space networks. This already difficult task is additionally complicated by the fact that satellites operate at multiple different orbits. Geostationary (GEO) satellites face significantly different channel conditions than their Low Earth Orbit (LEO) counterparts, meaning that any cross-orbital signal analysis tool must be able to handle widely varying conditions. Such tools may soon be critical for identifying satellites and detecting malicious cyber …

A Machine Learning Framework For Ddos Attack Detection In Sdn-Enabled Mobile Wireless Networks, Ishita Sharma, Satyam Agarwai, Shashi Shekhar Jha, Sumit Chakravarty

Faculty Articles

Attacks on network components and devices pose a significant threat to service continuity, necessitating robust detection mechanisms. This paper presents a Distributed Denial of Service (DDoS) attack detection framework tailored for heterogeneous mobile wireless networks within a Software-Defined Networking architecture. A two-tier model is proposed: localized attack detection at access points (APs) using a Multi-Layer Perceptron (MLP) classifier, and centralized detection under mobility at the controller using a Long Short-Term Memory (LSTM) model. The system incorporates novel traffic features such as flow count, speed of source IP, source and destination IP address entropy, proportion of bidirectional flows, and handover frequency, …

Machine Learning-Based Prediction And Experimental Validation Of The Pyrolysis Product Distribution In Tar-Rich Coals, Dong Shuai, Li Hongqiang, Wu Zhiqiang, Yu Zunyi, Lyu Zitao, Guo Wei, Yang Panxi, Fu Keming, Hao Xuanzhi, Liu Gen, Yang Bolun

Coal Geology & Exploration

Background Tar-rich coals serve as an important coal-based oil and gas resource in China, while their pyrolysis product distribution is governed by the coupling effects of coal properties and reaction conditions. Therefore, rapidly identifying the pyrolysis product distribution patterns holds great significance for the resource evaluation and experimental design of tar-rich coals.Methods Existing studies on tar-rich coals suffer from the insufficient integration of exclusive data and limited synergistic prediction capacities for multiple products. To address these issues, this study constructed a dedicated dataset involving proximate analysis, ultimate analysis, elemental molar ratios, maceral composition, and pyrolysis conditions by systematically collecting …

Physiobridge: Physiology-Constrained Self-Supervised Foundation Model For Cross-Device Ecg–Ppg Learning With Conformal Risk Control, Abbas Alzubaidi, Ali Al-Shuwaili, Ali Al-Bayaty

Electrical and Computer Engineering Faculty Publications and Presentations

Wearable and bedside sensors continuously generate electrocardiograms (ECG), photoplethysmograms (PPG), and related physiological waveforms that could enable earlier detection of deterioration and more personalized care. However, current deep learning pipelines in biomedical signal processing often remain taskand device-specific, degrade under domain shift (new hospitals, sensors, skin tones, motion), and provide limited uncertainty information for safety-critical decisions. We propose PhysioBridge, a foundation-model approach that learns a shared representation space for ECG and PPG via self-supervised pretraining and explicit physiology constraints, then supports downstream adaptation with distribution-free risk control. PhysioBridge introduces (i) multi-rate patch tokenization that preserves clinically meaningful morphology across heterogeneous …

Robust Non-Invasive Cardiac Index Prediction Via Feature Integration And Data-Augmented Neural Networks, Chih-Hao Chang, Mei-Ling Chan, Yu-Hung Fang, Po-Lin Huang, Tsung-Yi Chen, Tsun-Kuang Chi, I Elizabeth Cha, Tzong-Rong Ger, Kuo-Chen Li, Shih-Lun Chen, Liang-Hung Wang, Jia-Ching Wang, Patricia Angela R. Abu

Department of Information Systems & Computer Science Faculty Publications

Concurrent with the rising consumption of ultra-processed, high-calorie diets and the decline in physical activity, obesity and related cardiovascular conditions among young adults have continued to increase, becoming an important global public health concern. This study integrates non-invasive Internet of Things (IoT) sensing devices, including the TERUMO ES-P2000 blood pressure monitor (Terumo Corp., Tokyo, Japan) and the PhysioFlow PF07 Enduro cardiac hemodynamic analyzer (Manatec Biomedical, Poissy, France), with an artificial neural network (ANN) for cardiac index (CI) prediction. Through appropriate data preprocessing and model training strategies, the generalization ability and stability of the proposed CI prediction model were significantly enhanced. …

A Computer Vision Approach To Analyzing Taxane Effects On Prostate Cancer Cells, Diana Elizabeth Dancea

Electronic Theses and Dissertations 2020 - Present

Actin is a family of proteins that help create the structure of the cytoskeleton, which gives shape to the cell. In many chemotherapy treatments, researchers target actin because it controls the cell division process. Therefore, if they are able to understand the actin fibers, that may help in formulating methods to stop or slow down cancer cells from reproducing. Another important protein is PAK6, which regulates actin. In our research, a collaborative effort with Prof. Michael Lu’s lab at Florida Atlantic University, we use machine learning techniques to analyze cells which had their PAK6 protein knocked out, and compare them …

The Texture Of A Threat: Adversarial Training, Cnns, And Obfuscated Malware Detection, Kaelyn Haynie

Senior Honors Theses

Accurately detecting malicious programs is an expanding field of research for machine learning (ML), with a novel approach incorporating a bytecode-to-image pipeline that produces images representative of software. These images are provided to convolutional neural networks (CNNs) to be examined for malicious pattern indicators. However, CNNs struggle to generalize these patterns effectively while still being robust against adversarial data, an issue which this research addresses with adversarial training. In this paper, three unique CNN architectures (a DBFS-MC-inspired baseline, MIRACLE, and PSP-CNN) are trained for binary classification with 15,000 benign and malicious software samples encoded into images for Android, Windows, and …

Efficient Intrusion Detection For Iomt: Integrating Machine Learning, Feature Selection, And Fuzzy Logic, Ghaida Mansour Balhareth

Electronic Theses and Dissertations 2020 - Present

The internet of medical things (IoMT) has transformed healthcare by enabling real-time patient monitoring, remote diagnoses, and effective data exchange among connected medical devices and clinical systems. The increasing reliance on interconnected medical equipment has also intensified cybersecurity risks, as resource-constrained devices and wireless communication channels are vulnerable to attacks such as man-in-the-middle, spoofing, data injection, and ransomware. Intrusion Detection Systems (IDSs) play a critical role in mitigating these threats; however, traditional IDS approaches often struggle with high-dimensional IoMT data, class imbalance, and uncertainty in traffic patterns, which can increase false alarms and reduce reliability in safety-critical environments. This dissertation …

Real-Time Data Collection System For Low-Cost Diesel Emissions Monitoring, Zachary Driskill

Theses and Dissertations

Diesel trucks are a major source of pollutants, including particulate matter (PM) and NO$_x$. Roadside emissions monitoring can identify high-emitting diesel vehicles, but conventional sensing systems rely on expensive, complex sensing instruments that limit widespread deployment. Low-cost sensors offer increased scalability due to their small size, low cost, and low power consumption, but they have lower accuracy and slower response time. This thesis presents the design and evaluation of a low-cost roadside emissions monitoring system, establishing the foundation for a scalable and widely deployable alternative. We apply machine learning calibration to roadside emission monitoring to improve the usability of low-cost …

Advanced Machine Learning Approaches For Bankruptcy Prediction In The Shipping Industry: A Comparative Analysis Across Time Horizons, Minchul Sohn, Saeyeon Roh, Minsu Kwon

Plymouth Business School

This study evaluates machine learning methods for predicting bankruptcies in the Korean shipping industry over 1-, 3-, and 5-year horizons. Using an imbalanced dataset, preprocessing techniques such as outlier capping, skewness correction, and SMOTE-based oversampling are applied. Two feature sets—financial variables and those enriched with macroeconomic and shipping indicators—are analyzed. Results demonstrate that ensemble methods like LightGBM, CatBoost, XGBoost outperform linear models and deep neural networks across all horizons. Short-term predictions are driven by liquidity factors, while long-term forecasts benefit from shipping-specific indices like freight rates. SHAP‐based analysis reveals that internal cash‐flow as critical for one‐year predictions, whereas prolonged freight …

Development Of A Novel Cfd Approach For Predicting Multiphase Flow Enhanced By Machine Learning, Abdulrahman Elazazi Mohamed

Thesis/ Dissertation Defenses

Accurate estimation of interface orientation is crucial for ensuring both the accuracy and robustness of Volume of Fluid (VOF) schemes in multiphase flow simulations, especially when using non-uniform Cartesian meshes. Conventional gradient reconstruction approaches, such as least-squares (LSQ) methods, often suffer from significant errors and strong oscillations on highly stretched grids. This thesis proposes a grid-transferable, learning-based methodology that predicts interface unit normal vectors directly from the local volume-fraction field on non-uniform structured Cartesian grids by means of a compact feedforward neural network.

The methodology extends earlier work originally developed for uniform grids, which is first reproduced to assess how …

Ai Method For Classification Of Diagnosis Of Near-Infrared Breast Lesion Images, Kaiquan Chen, Fangyang Shen, Honggang Wang, Zhengchao Dong, Jizhong Xiao, Ming Ma, Afroza Aktar, Christopher Chow, Wenxiong Zhang

Publications and Research

In near-infrared optical breast lesion screening and diagnosis systems, high-speed four-dimensional scanners can dynamically acquire tens of thousands of lesion images within a five-minute period. Currently, manual computer annotation is required to generate standard samples from these scanned breast lesion images, a process that depends heavily on physicians with clinical expertise. On average, a single physician can annotate only approximately ten samples per working day. As a result, this process is time-consuming and labor-intensive, and the collected samples often suffer from low accuracy, large variability, and limited diagnostic reliability. Several AI-based annotation tools, such as QuPath, HALO AI™, and X-AnyLabeling, …

Federated Learning With Deep Learning: A Comprehensive Survey, Sarah H. Mnkash, Faiz A. Alawy, Israa T. Ali

Iraqi Journal of Computers, Communications, Control and Systems Engineering

Federated Learning (FL) introduces a decentralized machine learning paradigm where model training occurs directly on user devices, ensuring data privacy by keeping sensitive information local. This approach ensures data privacy and security by keeping all sensitive or personally identifiable information local to the device, thereby eliminating the need to transfer or centralize raw data. In this survey, we examine federated deep learning — its challenges, recent developments, application domains, privacy preservation strategies, and future trends. We will give some primitive reference papers of the heterogeneity of statistics and systems, communication cost, fairness, and trust. Applications in health, Internet of Things …

Using Machine Learning Algorithms To Clarify Relationships Between Soil Properties And Lead Stomach Bioaccessibility, Shehan Wijesinghe, Dibyendu Sarkar, Hadeer Saleh, Khalid Mustafa, Smitha Rao, Rupali Datta

Michigan Tech Publications

Featured Application: This paper has important applications in environmental health risk assessment and urban soil remediation planning. By demonstrating how machine learning can predict lead bioaccessibility in lead paint-contaminated soil, the study provides a scalable, cost-effective alternative to laboratory-based extraction methods. Such models could support rapid screening of contaminated sites, helping prioritize high-risk areas for intervention and allocate remediation resources more efficiently. This model is expected to advance data-driven decision-making for managing lead-contaminated soils and protect vulnerable urban populations from exposure. Lead contamination in urban soils, primarily from deteriorating lead-based paint, poses a significant health risk in the United States. …

A Hybrid Xfem-Ml Framework For High-Fidelity, Rapid Fracture Prediction In Cortical Bone: From Microstructure To Clinical Translation, Noor T. Al-Sharify, Ammar Fadhil Hussein Al-Maliki, Ahmed Ali Farhan Ogaili, Abdul-Rasool Kareem Jweri, Alaa Abdulhady Jaber, Luttfi A. Al-Haddad

Mesopotamian Journal of Computer Science

This paper proposes a hybrid computational model, which combines the eXtended Finite Element Method (XFEM) with machine learning (ML) to forecast the fracture behavior of cortical bone while maintaining microstructural fidelity. It has created a parametric dataset of about 450 three-dimensional XFEM models of single-edge notched bend (SENB) specimens, including important μ-structure features (e.g., osteon orientation, cement line properties, interfacial connectivity, etc.). Based on these simulations, 47 quantitative descriptors were obtained, and these were used to model supervised ML models, namely, Random Forest and Artificial Neural Networks, to estimate fracture load. The Random Forest model demonstrated exceptional predictive performance (R …