Data Science Commons^™

Exploring The Synergy Between Very Large Transformer And Lstm Models For Effective Medical Captioning From Videos To Text: The Impact Of Captioning In Healthcare, R. V. Aswiga, Moin Ahmed Zahir

Data Science Faculty Publications

In today’s rapidly evolving digital landscape, the demand for accurate and contextually relevant subtitles for image and video content, particularly in the medical domain, is increasingly critical. Despite the proliferation of visual data across various platforms, existing captioning systems often struggle due to variations in visual settings, complex temporal relationships, and nuanced semantics. Additionally, challenges such as limited datasets, privacy issues, and specialized annotation requirements make medical image captioning particularly difficult. To tackle these challenges, we investigate cutting-edge deep learning methodologies, specifically Transfer Learning and Transformer models, through a comparative analysis. Specifically, we focus on Transfer Learning through the MedVisionCapturer …

Explainable Physics-Based Constraints On Reinforcement Learning For Accelerator Optimization, Jonathan Colen, Malachi Schram, Kishansingh Rajput, Armen Kasparian

Data Science Faculty Publications

We present a reinforcement learning (RL) framework for optimizing particle accelerator experiments that builds explainable physics-based constraints on agent behavior. The goal is to increase transparency and trust by letting users verify that the agent’s decision-making process incorporates suitable physics. Our algorithm uses a learnable surrogate function for physical observables, such as energy, and uses them to fine-tune how actions are chosen. This surrogate can be represented by a neural network or by an interpretable sparse dictionary model. We test our algorithm on a range of particle accelerator optimization environments designed to emulate the Continuous Electron Beam Accelerator Facility at …

Applying Transfer Learning For Street-Scale Nuisance Flood Forecasting In Coastal-Urban Environments, Binata Roy, Jonathan L. Goodall, Diana Mcspadden, Chetan Kumar, Steven Goldenberg, Yidi Wang, Malachi Schram

Data Science Faculty Publications

An important challenge with Machine Learning (ML) is its transferability; that is, whether an ML model trained on one set of data can be applied to a second set of data without requiring full retraining of the model. Transfer Learning (TL) addresses this challenge by transferring knowledge learned in the source domain (the data it was trained on) to the target domain (a second set of data that is statistically different but related, which the model was not trained on). This study investigates the use of TL for street-scale nuisance flood forecasting by exploring whether an ML model trained for …

Predictive Analytics For Optimising Delivery Performance In Supply Chain Management, Vishwanath Sai Krishna Teja

ICT

Supply chain management has always been a challenging area where delivery delays stand as one of its main ongoing issues. Many companies continue to use outdated data collection systems despite the advancements in data collection tools. This study investigates how predictive analytics can help companies to prevent their operational problems from reaching their customers.

In this study, two datasets were used. The first is the Data Co Smart Supply Chain dataset which covers order level operations from 2015 to 2018, and the second is the World Bank Logistics Performance Index, which covers logistics quality across more than 160 countries. After …

Balancing Accuracy And Discovery: A Bias-Aware Evaluation Framework For Music Recommendation, Jose Luis Maidana

ICT

Music recommender systems play an important role in helping users navigate large catalogues, but their reliance on historical interaction data can reinforce popularity bias and limit the visibility of less popular, long-tail items. This research develops and evaluates a bias-aware framework for assessing and mitigating popularity bias in music recommendation using implicit user-item interaction data. The study first quantifies popularity bias through inequality and long-tail metrics, before evaluating two baseline recommender approaches: a non-personalised Most-Popular model and a personalised Matrix Factorisation model. Their performance is assessed using both predictive accuracy measures, including Precision, Recall and NDCG, and beyond-accuracy measures relating …

Healthcare Digital Twins: A Methodological Literature Review On Integrating Iot And Ai For Personalized Medicine And Predictive Care, Sara Shahnazinia, Mahsa Tavasoli, Abdolhossein Sarrafzadeh, Ali Karimoddini

Electrical & Computer Engineering Faculty Publications

Digital Twin (DT) technology has the potential to revolutionize healthcare delivery and enhance patient outcomes through personalized and precision medicine, simulation models for operations and interventions, and drug discovery. However, successful implementation of DTs in Internet of Things (IoT) and artificial intelligence (AI) healthcare is contingent upon addressing key challenges such as privacy, ethics, and robust data security. This paper presents a methodological literature review of DT applications in healthcare, systematically analyzing the current state of research, key enabling technologies, and implementation challenges. The review summarizes DT categorization approaches (application-based, technology-based, and real-time function-based); delineates core DT components such as …

Integer-Valued Time Series Model Via Copula-Based Bivariate Skellam Distribution, Mohammed Alqawba, Norou Diawara, Mame Mor Sene

Mathematics & Statistics Faculty Publications

Time series analysis is crucial for modeling and forecasting diverse real-world phenomena. Traditional models typically assume continuous-valued data; however, many applications involve integer-valued series, often including negative integers. This paper introduces an approach that combines copula theory with the bivariate Skellam distribution to handle such integer-valued data effectively. Copulas are widely recognized for capturing complex dependencies among variables. By integrating copulas, our proposed method respects integer constraints while modeling positive, negative, and temporal dependencies accurately. Through simulation and an empirical study on a real-life example, we demonstrate that our class of models performs well. This approach has broad applicability in …

A Bayesian Late-Fusion Supportability Framework For Rare-Disease Severity Prediction In Glut1 Deficiency Syndrome, Jordan M. Rodriguez

Mathematics Dissertations

Glucose transporter type 1 deficiency syndrome (GLUT1-DS) is a rare neurometabolic disorder with heterogeneous neurological and developmental severity. Because patient-level severity is not observed as a single validated outcome, this dissertation develops a Bayesian late-fusion supportability framework for constructing and predicting an ordered latent severity phenotype from clinical, genetic, and EEG-derived evidence. The primary target was constructed in a larger clinical cohort using age-5 symptom burden and learning cognition, then assigned to an aligned multimodal prediction cohort. Target-defining variables were excluded from supervised predictors, and models were evaluated using patient-exclusive cross-validation with training-fold preprocessing and fold-wise EEG PCA.

The primary …

Hybrid Patchtst And Physics-Based Framework For Predicting Lithium-Ion Battery State Of Health, Pavan Ravuri

Honors Undergraduate Theses

Accurately forecasting the state of health of lithium-ion batteries is critical for improving performance, reliability and lifetime in energy storage applications. Battery capacity degrades into a nonlinear pattern over cycling due to electrochemical processes where neither purely data driven nor physics-based models can capture alone. This study looks at a hybrid framework combining that PatchTST patch-based transformer architecture with physics-based features derived from the solid electrolyte interphase and pseudo two-dimensional models. Physics inspired proxy features were computed from cycling data and concatenated with electrochemical measurements as added input channels before patch segmentation. There are five model configurations that were evaluated …

From Lap To Map: How Musical Scale, Place, And Play Drive The Interconnected Mario Kart World, Cameron Cummins

Honors Undergraduate Theses

With their deserts, castles, and ghost houses, the environments of Super Mario games are colorful, whimsical, and charming, but why are they so compelling, and what happens when our analysis of these environments extends beyond individual levels to expansive game worlds? Drawing on Cresswell’s theory of place (2014) and recent work on musical place-building in Mario Kart 8 (Heazlewood-Dale, 2024), I propose a spectrum between localized and globalized scale in games. As game environments become increasingly globalized, the music may be similarly altered to account for this shift in scale. Consequently, players may then encounter a broader, less musically congruent …

Comparing Unet And Hrnet For Geomorphic Feature Extraction From Digital Terrain Model-Derived Land Surface Parameters, Behnam Solouki

Graduate Theses, Dissertations, and Problem Reports (ETD)

Automated extraction of geomorphic features from high spatial resolution terrain data is important for environmental monitoring, geohazard assessment, and landscape change analysis. This study compares two convolutional neural network architectures, UNet and High-Resolution Network (HRNet), for semantic segmentation of mine benches and sinkholes from lidar-derived digital terrain models and land surface parameters. Mine benches represent anthropogenic, elongated, stepped landforms associated with historic surface mining, whereas sinkholes are natural, compact depressions common in karst terrain. Three terrain-derived input layers were used: square-root-transformed slope, broad-scale topographic position index, and fine-scale annulus topographic position index. Models were trained using five training sample sizes, …

A Major Update And Improved Validation Functionality In The Mwtab Python Library And The Metabolomics Workbench File Status Website, P. Travis Thompson, Hunter N. B. Moseley

Markey Cancer Center Faculty Publications

Background: The Metabolomics Workbench (MW) is a public scientific data repository consisting of experimental data and metadata from metabolomics studies collected with mass spectroscopy (MS) and nuclear magnetic resonance (NMR) analyses. Although not as rapidly as in the past, MW has steadily evolved, updating its mwTab and JSON deposition text file formats and its web-based infrastructure. However, the growth of MW has been exponential since its inception in 2013 and continues to be exponential, with the number of datasets hosted on the repository increasing by 50% since April 2024. As part of regular maintenance to keep up with changes to …

An Integrated Data-Driven Framework For Arctic Shipping: Analyzing Vessel Speed, Environmental And Ecological Factors Through Innovative Statistical Spatio-Temporal Methods, Inverse Optimization And Machine Learning, Mauli Pant

Theses and Dissertations

This dissertation develops an integrated data-driven framework to analyze vessel navigation and ecological risk in the United States Arctic from 2010 to 2019. As environmental change and maritime activity increase in the region, understanding how vessels respond to dynamic conditions and how those responses interact with marine ecosystems has become increasingly important. A central theme of this dissertation is the treatment of vessel speed as both an observed outcome and a decision variable reflecting trade- offs among operational, environmental, and ecological factors. The first chapter develops a predictive framework for vessel speed over ground (SOG) using Gaussian Process Boosting (GPBoost), …

Machine Learning-Based Intrusion Detection System For Iot Networks Using The Rt-Iot 2022 Dataset, Bukunmi Ebenezer Afolabi

Theses, Dissertations and Capstones

The rapid expansion of the Internet of Things (IoT) has transformed modern computing by enabling seamless connectivity among heterogeneous devices across diverse application domains. However, this increased interconnectivity has significantly enlarged the attack surface of IoT networks, exposing them to a wide range of sophisticated cyber threats. Conventional security mechanisms often lack the capability to detect emerging attacks in real time, thereby necessitating the development of intelligent Intrusion Detection Systems (IDS) capable of accurately identifying malicious network activities. This study developed and evaluated a machine learning-based intrusion detection framework for multiclass IoT attack detection using the RT-IoT2022 dataset. The dataset …

Benchmarking And Advancing Heterogeneous Data Reasoning With Large Language Models, Yebowen Hu

Graduate Studies Theses and Dissertations 2026

Large language models (LLMs) increasingly face complex real-world tasks requiring complex reasoning over heterogeneous inputs, including text, numbers, and structured data. This dissertation first investigates the capability boundaries of LLMs under such scenarios. Through the MeetingBank and SportsMetrics benchmarks, we show that while LLMs exhibit linguistic fluency, they struggle with factual accuracy, information density, and quantitative cross-referencing in meeting summarization and sports analysis. Using DecipherPref, a pair-wise evaluation framework grounded in the Bradley-Terry-Luce model, we further probe LLMs' inherent preferences for information-rich and lengthy inputs, and we identify critical deficiencies in financial decision-making through DeFine, where models lack precise insight …

An Analysis Of Slms For Machine Translation Of Healthcare Documents, Wyatt Clausen

All Graduate Theses, Dissertations, and Other Capstone Projects

The language barrier creates significant healthcare challenges for patients who struggle with English. While large language models (LLMs) have advanced machine translation, their high computational costs, environmental impact, and risk of data breaches have raised concerns. Small language models (SLMs) can address these concerns and be further adapted to specific tasks through parameter-efficient fine-tuning (PEFT) methods, including Quantized Low-Rank Adaptation (QLoRA). This study compares the Spanish-to-English translation quality of two 7-billion-parameter SLMs, Mistral and Llama 2 7B, in the medical domain, using BLEU, chrF, and COMET as evaluation metrics. It further investigates whether QLoRA fine-tuning and Retrieval-Augmented Generation (RAG) on …

Lost In The Language: Data Breaches And The Strategic Fog Of Risk Disclosures, Ling Tuo, Shipeng Han

Accounting Faculty Publications

This study examines whether firms strategically adjust the readability of Item 1A (“Risk Factors”) disclosures following data breaches. Using U.S. firm-year observations from 2006 to 2023, we find that data breaches are associated with a significant decline in Item 1A readability. This decline is not accompanied by a meaningful increase in informational content; instead, post-breach disclosures exhibit higher syntactic complexity, more positive tone, and lower textual similarity to prior and industry peers' filings, consistent with strategic obfuscation rather than transparent reporting. The readability decline is amplified among firms facing higher litigation risk but attenuated among firms with stronger reputations for …

Ai-Assisted Surface-Enhanced Raman Spectroscopy For Cardiovascular Diagnostics: From Plasmonic Materials To Clinical Translation, Anju Joshi, Gymama Slaughter

Center for Bioelectronics Publications

Raman spectroscopy (SERS) has emerged as a powerful analytical technique, offering molecular fingerprint specificity and ultrasensitive detection of cardiac biomarkers. Recent advances in plasmonic nanostructures, surface functionalization strategies, and flexible sensing platforms have significantly improved the analytical performance of SERS-based biosensors. In parallel, the integration of artificial intelligence (AI) and machine learning has enabled robust interpretation of complex spectral datasets, facilitating automated biomarker classification and improved diagnostic accuracy in heterogeneous biological environments. Despite these advances, the field remains fragmented, with limited integration between nanomaterial design, biomarker selection, and data-driven analysis, and persistent challenges related to reproducibility, standardization, and clinical validation. …

Graph-Based And Graph-Transformer Representation Learning For Healthcare Data, Rui Wang

Electronic Theses & Dissertations (2024 - present)

Healthcare data exhibit complex structures, including heterogeneous clinical entities, sparse observations, and longitudinal patient trajectories. Effectively modeling such data remains a fundamental challenge in computational healthcare research. Traditional machine learning approaches often rely on flat feature representations that fail to capture relationships among clinical events, limiting their ability to model complex healthcare processes. These challenges motivate structured learning frameworks that capture both relational structure and temporal dynamics in healthcare data. This dissertation develops a series of graph-based representation learning approaches, extended through graph-transformer architectures for modeling complex healthcare data. Such data can be represented as graphs, where nodes correspond to …

From Reviews To Value: Harnessing Crowdsourced Data To Capture Visitor Perceptions And Economic Benefits Of Recreation, Laura Costadone, Shan Zhang

ODU Articles

Quantifying the recreational value of protected natural areas is essential for sustainable management, conservation financing, and informed decision-making. Traditional approaches, such as on-site surveys, are often costly, spatially limited, and constrained by regulatory barriers. This study evaluates the potential of crowdsourced user-generated data to jointly assess visitor perceptions and the economic value of nature-based recreation, using Back Bay National Wildlife Refuge (Virginia, USA) as a case study. We integrated georeferenced photographs and textual content from four platforms (Flickr, TripAdvisor, Yelp, and AllTrails) with supplementary survey data to analyze visitation patterns, visitor sentiment, cultural ecosystem services, and recreational value using the …