Computer Sciences Commons^™

Information-Theoretic Methods For Efficient Training And Robust Evaluations In Self-Supervised Learning, Oscar Skean

Theses and Dissertations--Computer Science

Self-supervised learning (SSL) has become a cornerstone of modern machine learning, offering a scalable alternative to costly human annotation by constructing pretext tasks directly from raw data. While SSL has delivered strong results across vision, language, and multimodal domains, two major limitations persist: (1) SSL methods are often significantly slower to train than supervised counterparts, and (2) evaluation protocols remain narrow, with most studies relying on linear probing accuracy on the pretraining dataset. . These challenges are particularly acute for large language models (LLMs), where training costs and interpretability of intermediate representations are critical concerns.

In this work, we propose …

Predicting Mental Health Disparities Using Machine Learning For African Americans In Southeastern Virginia, Ismail El Moudden, Michael C. Bittner, Matvey V. Karpov, Isaac O. Osunmakinde, Akosua Acheamponmaa, Breshell J. Nevels, Mamadou T. Mbaye, Tonya L. Fields, Karthiga Jordan, Messaoud Bahoura

Department of Obstetrics & Gynecology Faculty Publications

This study examined mental health disparities among African Americans using AI and machine learning for outcome prediction. Analyzing data from African American adults (18–85) in Southeastern Virginia (2016–2020), we found Mood Affective Disorders were most prevalent (41.66%), followed by Schizophrenia Spectrum and Other Psychotic Disorders. Females predominantly experienced mood disorders, with patient ages typically ranging from late thirties to mid-forties. Medicare coverage was notably high among schizophrenia patients, while emergency admissions and comorbidities significantly impacted total healthcare charges. Machine learning models, including gradient boosting, random forest, neural networks, logistic regression, and Naive Bayes, were validated through 100 repeated 5-fold cross-validations. …

Fares On Fairness: Using A Total Error Framework To Examine The Role Of Measurement And Representation In Training Data On Model Fairness And Bias, Patrick Oliver Schenk, Christoph Kern, Trent D. Buskirk

Data Science Faculty Publications

Data-driven decisions, often based on predictions from machine learning (ML) models are becoming ubiquitous. For these decisions to be just, the underlying ML models must be fair, i.e., work equally well for all parts of the population such as groups defined by gender or age. What are the logical next steps if, however, a trained model is accurate but not fair? How can we guide the whole data pipeline such that we avoid training unfair models based on inadequate data, recognizing possible sources of unfairness early on? How can the concepts of data-based sources of unfairness that exist in the …

Harnessing The Power Of Gradient-Based Simulations For Multi-Objective Optimization In Particle Accelerators, Kishansingh Rajput, Malachi Schram, Auralee Edelen, Jonathan Colen, Armen Kasparian, Ryan Roussel, Adam Carpenter, He Zhang, Jay Benesch

Data Science Faculty Publications

Particle accelerator operation requires simultaneous optimization of multiple objectives. Multi-objective optimization (MOO) is particularly challenging due to trade-offs between the objectives. Evolutionary algorithms, such as genetic algorithms (GAs), have been leveraged for many optimization problems, however, they do not apply to complex control problems by design. This paper demonstrates the power of differentiability for solving MOO problems in particle accelerators using a deep differentiable reinforcement learning (DDRL) algorithm. We compare the DDRL algorithm with model-free reinforcement learning (MFRL), GA, and Bayesian optimization (BO) for simultaneous optimization of heat load and trip rates in the continuous electron beam accelerator facility. The …

Training Set Augmentation And Harmonization Enables Radiomic Models To Detect Early Onset Of Lung Cancer, Claire Huchthausen, Menglin Shi, Gabriel L.A. Sousa, James Larner, Einsley Janowski, Jonathan Colen, Krishni Wijesooriya

Data Science Faculty Publications

Radiomics-based machine learning models have the potential to detect lung cancer at inception from CT scans and transform patient outcomes. Low malignancy rates in early-development pulmonary nodules (PNs) and variable image acquisition hinder development of clinically applicable radiomics-based early detection models. To address these challenges, we augmented training using later-development PNs and harmonized for acquisition effects. We first trained machine learning models to predict PN malignancy using radiomic features from scans of early-development benign and malignant PNs (n = 187) harmonized using ComBat. Observing near-chance performance, we augmented training with later-development benign and malignant PNs (n = 225). We evaluated …

Securing The Internet Of Wetland Things (Iowt) Using Machine And Deep Learning Methods: A Survey, Guma Ali, Wamusi Robert, Maad M. Mijwil, Malik Sallam, Jenan Ayad

Mesopotamian Journal of Computer Science

Wetlands are essential ecosystems that provide ecological, hydrological, and economic benefits. However, human activities and climate change are degrading their health and jeopardizing their long-term sustainability. To address these challenges, the Internet of Wetland Things (IoWT) has emerged as an innovative framework integrating advanced sensing, data collection, and communication technologies to monitor and manage wetland ecosystems. Despite its potential, the IoWT faces substantial security and privacy risks, compromising its effectiveness and hindering adoption. This survey explores integrating machine learning (ML) and deep learning (DL) techniques as solutions to address the security threats, vulnerabilities, and challenges inherent in IoWT ecosystems. The …

Nexus: Network Exploration For Exploiting Unsafe Sequences In Multi-Turn Llm Jailbreaks, Javad Rafiei Asl, Sidhant Narula, Mohammad Ghasemigol, Eduardo Blanco, Daniel Takabi

School of Cybersecurity Faculty Publications

Large Language Models (LLMs) have revolutionized natural language processing, yet remain vulnerable to jailbreak attacks—particularly multi-turn jailbreaks that distribute malicious intent across benign exchanges, thereby bypassing alignment mechanisms. Existing approaches often suffer from limited exploration of the adversarial space, rely on hand-crafted heuristics, or lack systematic query refinement. We propose NEXUS (Network Exploration for eXploiting Unsafe Sequences), a modular framework for constructing, refining, and executing optimized multi-turn attacks. NEXUS comprises: (1) ThoughtNet, which hierarchically expands a harmful intent into a structured semantic network of topics, entities, and query chains; (2) a feedback-driven Simulator that iteratively refines and prunes these chains …

Qos-Aware Link Adaptation For Beyond 5g Networks: A Deep Reinforcement Learning Approach, Ali Parsa, Neda Moghim, Sachin Shetty

Center for Secure and Intelligent Critical Systems (CSICS) Publications

Modern wireless communication systems face increasingly complex challenges due to rapidly changing channel conditions and the growing diversity of application-specific Quality of Service (QoS) requirements. Traditional link adaptation mechanisms primarily aim to maximize throughput and often lack the flexibility to support emerging applications, such as Extended Reality (XR) and Virtual Reality (VR), which demand simultaneous guarantees for high data rates, ultra low latency, and high reliability. These stringent and multidimensional QoS needs call for more intelligent and adaptive solutions. In this paper, we propose QDRLLA (QoS-aware Deep Reinforcement Learning-based Link Adaptation), a novel framework that employs deep reinforcement learning to …

An Integrated Machine Learning Approach For Identifying Emergency Rescue Messages On Social Media During Natural Disasters, Wael Khallouli, Jiang Li, Jingwei Huang, Ghaith Rabadi, Samuel Kovacic

School of Cybersecurity Faculty Publications

During large-scale disasters, emergency call centers are often overwhelmed by the large volume of rescue requests and calls for help. Consequently, people are turning to social media platforms to seek assistance. Rescue information posted on these platforms is extremely valuable for first responders to make informed rescue decisions. Therefore, the automatic identification of these requests from the vast amount of data posted on social media during crises is critical yet challenging. This work presents our ongoing research on applying deep learning techniques to extract actionable rescue information from social media during crises. We proposed a novel deep learning model that …

Strategic Identification Of Prognostic Biomarkers For Knee Osteoarthritis Via Optimized Regression Techniques, Varun Sri Sai Vemuri

Selected Full-Text Master Theses 2021-

Knee Osteoarthritis (KOA) is a progressive musculoskeletal disease involving cartilage matrix degradation, subchondral bone remodeling, and systemic inflammation, significantly impairing joint function and mobility. Existing KOA prediction models are not designed to account for nonlinear multimodal biomarker interactions or to integrate biochemical and imaging data, thus limiting their clinical utility. The current method for early detection and prediction of KOA disease progression is primarily based on machine learning-based approaches using radiographic imaging data, static feature selection, and deterministic outputs. These machine learning approaches often fail to capture the pathophysiology of KOA disease progression, which involves a complex cascade of processes, …

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

VMASC Publications

An important challenge with Machine Learning (ML) is its transferability; i.e., whether a ML model trained on one set of data can be applied to a second set of data without requiring full re-training 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 a ML model trained for one …

Ai-Generated Messaging For Life Events Using Structured Prompts: A Comparative Study Of Gpt With Human Experts And Machine Learning, Christopher Lynch, Erik Jensen, Ross Gore, Virginia Zamponi, Kevin O'Brien, Brandon Feldhaus, Katherine Smith, Joseph Martínez, Madison H. Munro, Timur E. Ozkose, Tugce B. Gundogdu, Ann Marie Reinhold, Hamdi Kavak, Barry Ezell

VMASC Publications

Large Language Models (LLMs) play an increasingly integrated and pivotal role in generating diverse types of texts, such as social media messages, emails, narratives, and technical reports, among other textual communication forms. As AI-generated messaging filters into human communication, a systematic exploration of their effectiveness for mimicking human-like communication of life events is needed. In this study, we employ a zero-shot structured narrative prompt to generate 24,000 life event messages for birth, death, hiring, and firing events using OpenAI's GPT-4. From this dataset, we manually classify 2880 messages and evaluate their validity in conveying these life events through the form …

A Novel Intelligent Thermal Feedback Framework For Electric Motor Protection In Embedded Robotic Systems, Mohamed Shili, Salah Hammedi, Hicham Chaoui, Khaled Nouri

Electrical & Computer Engineering Faculty Publications

As robotic systems advance in autonomy and sophistication while being used in uncertain environments, the challenge of building reliable and robust electric motors that are embedded into robotic systems has never been a more important engineering problem. Thermal distress caused by extended operation or excessive loading can negatively affect a motor’s performance and efficiency and lead to catastrophic hardware failure. This paper proposes a novel intelligent control framework that includes real-time thermal feedback for hybrid electric motors that are embedded into robotic systems. The framework relies on adaptive control techniques and lightweight machine learning techniques to estimate internal motor temperatures …

A New Deepfake Detection Method With No-Reference Image Quality Assessment To Resist Image Degradation, Jiajun Jiang, Wen-Chao Yang, Chung-Hao Chen, Timothy Young

Electrical & Computer Engineering Faculty Publications

Deepfake technology, which utilizes advanced AI models such as Generative Adversarial Networks (GANs), has led to the proliferation of highly convincing manipulated media, posing significant challenges for detection. Existing detection methods often struggle with the low-quality or compressed press, which is prevalent on social media platforms. This paper proposes a novel Deepfake detection framework that leverages No-Reference Image Quality Assessment (NRIQA) techniques, specifically, BRISQUE, NIQE, and PIQUE, to extract quality-related features from facial images. These features are then classified using a Support Vector Machine (SVM) with various kernel functions. We evaluate our method under both intra-dataset and cross-dataset settings. For …

Predictive Maintenance In Naval Vessel Propulsion Systems For Enhanced Marine Operations Using A Bigmm-Hmm Framework With Divergence-Based Clustering, Farshid Javadnejad, Hyoshin John Park, Samuel Kovacic, Andres Sousa-Poza

Engineering Management & Systems Engineering Faculty Publications

This study introduces a BiGMM-HMM Integration Framework designed to improve predictive maintenance strategies for naval vessel propulsion systems, addressing the need for efficient and reliable operation in marine engineering applications. The framework effectively manages multimodal sensor data by leveraging a unique combination of Gaussian Mixture Models (GMMs) and Hidden Markov Models (HMMs) in a bidirectional architecture. It analyses the dynamic interactions between sensors and subsystems. Two preprocessing methods are evaluated: Method 1 focuses on subsystem interactions, employing divergence-based root cause analysis to identify key sensor variables by clustering of sensors and subsystems. In contrast, Method 2 processes the entire dataset …

Optimizing Port Logistics Through Generative Ai: Revolutionizing Efficiency And Resilience In The Maritime Industry, Minodora Badea, Olga Bucovetchi, Adrian V. Gheorghe, Gabriel Raicu

Engineering Management & Systems Engineering Faculty Publications

The maritime industry faces growing challenges in optimizing port logistics due to increasing trade volumes, environmental regulations, and supply chain disruptions. This comprehensive literature review examines the transformative role of artificial intelligence (AI), with particular focus on generative AI, in enhancing efficiency and resilience in port operations. Through systematic analysis of 23 peer-reviewed studies published between 2021-2025, this review synthesizes advancements in real-time data integration, machine learning, digital twins, IoT, and autonomous systems that collectively improve operational decision-making, risk management, and environmental sustainability. Key findings reveal that machine learning applications achieve 90% effectiveness ratings in operational optimization, while predictive analytics …

Leveraging Transformer-Based Ocr Model With Generative Data Augmentation For Engineering Document Recognition, Wael Khallouli, Mohammad Shahab Uddin, Andres Sousa-Poza, Jiang Li, Samuel Kovacic

Engineering Management & Systems Engineering Faculty Publications

The long-standing practice of document-based engineering has resulted in the accumulation of a large number of engineering documents across various industries. Engineering documents, such as 2D drawings, continue to play a significant role in exchanging information and sharing knowledge across multiple engineering processes. However, these documents are often stored in non-digitized formats, such as paper and portable document format (PDF) files, making automation difficult. As digital engineering transforms processes in many industries, digitizing engineering documents presents a crucial challenge that requires advanced methods. This research addresses the problem of automatically extracting textual content from non-digitized legacy engineering documents. We introduced …

T3-Ciders: Train-The-Trainer And Community Building To Increase Cyberinfrastructure Adoption In Cybersecurity Research And Education, Wirawan Purwanto, Mohan Yang, Peng Jiang, Shanan Chappell Moots, Masha Sosonkina, Hongyi Wu

University Administration Publications

T³-CIDERS is a train-the-trainer program to increase the adoption of advanced cyberinfrastructure (CI) and data skills into the fabric of research and education in cybersecurity and cyber-related disciplines. T³-CIDERS trains faculty, researchers, and students as “future trainers” (FTs) with hands-on technical and instructional skills to enable more people to effectively leverage CI in cybersecurity. The program includes a series of technical pre-training modules, a weeklong summer institute, ongoing learning engagements conducted over an academic year; it culminates with the FTs conducting locally tailored CI-infused training events at their respective home institutions. Ultimately, T³-CIDERS aims to build a “CI+cybersecurity” community of …

Point Cloud-Based Diffusion Models For The Electron-Ion Collider, Jack Y. Araz, Vinicius Mikuni, Felix Ringer, Nobuo Sato, Fernando Torales Acosta, Richard Whitehill

Physics Faculty Publications

At high-energy collider experiments, generative models can be used for a wide range of tasks, including fast detector simulations, unfolding, searches of physics beyond the Standard Model, and inference tasks. In particular, it has been demonstrated that score-based diffusion models can generate high-fidelity and accurate samples of jets or collider events. This work expands on previous generative models in three distinct ways. First, our model is trained to generate entire collider events, including all particle species with complete kinematic information. We quantify how well the model learns event-wide constraints such as the conservation of momentum and discrete quantum numbers. We …

Development And Evaluation Of Machine Learning Models For Early Pediatric Sepsis Prediction, Ancita M. Andrade

Browse all Theses and Dissertations

Sepsis is a leading cause of pediatric mortality, claiming more lives in the United States annually than all childhood cancers combined. Early identification in Emergency Departments (EDs) remains challenging, as the current Phoenix criteria establishes an updated international consensus definition for sepsis, however is not designed for use as a screening tool. This study aimed to develop predictive models identifying pediatric patients at risk of sepsis within 24 hours of admission. Multiple tree-based and deep learning models were trained utilizing clinical and laboratory data from the initial four hours of presentation. Both the LightGBM and LSTM architectures demonstrated superior performance, …

Dataset Generation For Routing Policy Study In Ad Hoc Wireless Networks, Vishnu Vishnu Priya

Browse all Theses and Dissertations

Ad Hoc wireless networks, with their decentralized architecture and dynamic topology, present challenges in reliable and energy-efficient routing. While machine learning (ML) and reinforcement learning (RL) offer promising solutions, progress is limited by the lack of realistic, high-fidelity datasets. This research introduces a simulation-based framework for generating four diverse datasets representing combinations of node mobility (mobile vs. static) and spatial distribution (random vs. clustered). Each dataset captures critical metrics such as Signal-to-Interference-plus-Noise Ratio (SINR), bottleneck rate, and power consumption across multi-hop paths. A lookahead-based greedy routing algorithm with scenario-aware power control is implemented to emulate practical behavior. Supervised ML models, …

A Secure Ml-Assisted Framework For Resilient And Efficient Prediction Of Physiotherapy Sequence In Bilateral Carpal Tunnel Syndrome, Pratik Pandurang Kharat

Browse all Theses and Dissertations

Bilateral idiopathic carpal tunnel syndrome (CTS) is a neuromuscular disorder characterized by compression of the median nerve at both wrists, leading to symptoms such as pain, numbness, tingling, and muscle weakness. Unlike unilateral cases, bilateral idiopathic CTS presents distinct therapeutic challenges due to the simultaneous involvement of both hands and the lack of an identifiable underlying cause. This study explores the application of machine learning techniques to predict the optimal sequence of physiotherapeutic interventions Stretching followed by Myofascial Mobilization (S/M) or the reverse (M/S) in female patients with bilateral idiopathic CTS and right hand dominance. Data were drawn from a …

Learning Under Data Scarcity: Reasoning And Negative Distillation For Texts And Graphs, Calvin T. Greenewald

Browse all Theses and Dissertations

Modern machine learning (ML) models rely on large amounts of high-quality labeled data to achieve optimal performance. However, in many real-world domains, such as cyber security, acquiring sufficient labeled data is often infeasible due to cost, privacy concerns, and the rapid evolution of underlying phenomena. This challenge underscores the importance of learning under data scarcity. This thesis addresses this challenge by proposing distinct, modality-specific techniques for text and graph domains, which allow models to generalize effectively with minimal data. For text classification task, we incorporate distilled rationales from large language models and adversarial perturbations into the input space to improve …

Digital Health Using Data Science, Padmapriya Velupillai Mekandan, Paramita Basak Upama, Masud Rabbani, Amity Ali, Sheikh Iqbal Ahamed

Computer Science Faculty Research and Publications

The digitization of healthcare has led to an unprecedented growth in health-related data, offering new opportunities to transform clinical decision-making, disease prediction, and patient engagement. However, extracting actionable insights from diverse data sources such as electronic health records, wearable devices, and mobile health apps requires a fusion of domain knowledge in healthcare and technical expertise in data science. This paper presents a structured, interdisciplinary framework for digital health that emphasizes practical strategies for data acquisition, preprocessing, feature engineering, machine learning, and ethical data use. The model promotes a holistic understanding of digital health challenges and opportunities, preparing future professionals to …

The Integration Of Artificial Intelligence And Ontologies: Transformations In Knowledge Representation And Application, Grazia Serratore, Julaine Clunis

STEMPS Faculty Publications

Artificial Intelligence (AI) is reshaping the landscape of knowledge representation. There is an increasingly strong bidirectional relationship, between AI techniques and ontologies. AI techniques revolutionized traditional, manual ontology development and contribute to automated ontology construction, while ontologies enhance the performance of AI systems and their semantic accuracy. Through a comprehensive review of current literature, this paper aims to examine: i) how Machine Learning (ML) techniques contribute to the automated construction, refinement, and validation of ontologies; ii) the most widely used and effective ML approaches for ontology construction; iii) how domain-specific requirements influence the selection and adaptation of AI techniques for …

How Artificial Intelligence Will Shape Securities Regulation, Gabriel Rauterberg

Other Publications

How will the increasing prevalence and sophistication of artificial intelligence (AI) change the doctrine and practice of securities law? My main thesis is that it will push securities regulation toward a more systems-oriented approach. This approach will replace securities law's emphasis, in areas like manipulation, on forms of enforcement targeted at specific individuals and accompanied by punitive sanctions with a greater focus on ex ante rules designed to shape an ecology of actors and information.

Ai Innovations In Rppg Systems For Driver Monitoring: Comprehensive Systematic Review And Future Prospects, Soha G. Ahmed, Katrien Verbert, Nazar Zaki, Ashraf Khalil, Hamad Aljassmi, Fady Alnajjar

All Works

Advanced technologies, notably camera-based systems using remote photoplethysmography (rPPG), are increasingly used in automotive safety to non-invasively monitor driver well-being and fatigue by measuring physiological metrics like heart and respiration rates. This review examines recent advancements in machine learning algorithms and signal processing for rPPG in driver monitoring. A literature search up to April 2, 2024, across major databases, identified 344 studies; 29 were analyzed in depth, focusing on: 1) rPPG signal extraction and heart rate estimation, where deep learning improved accuracy; 2) fatigue detection, showing benefits of multimodal data fusion; 3) mental state monitoring, with machine learning classifying cognitive …

Automating The Amino Acid Identification In Elliptical Dichroism Spectrometer With Machine Learning, Ridhanya Sree Balamurugan, Yusuf Asad, Tommy Gao, Dharmakeerthi Nawarathna, Umamaheswara Rao Tida, Dali Sun

Electrical & Computer Engineering Faculty Publications

Amino acid identification is crucial across various scientific disciplines, including biochemistry, pharmaceutical research, and medical diagnostics. However, traditional methods such as mass spectrometry require extensive sample preparation and are time-consuming, complex and costly. Therefore, this study presents a pioneering Machine Learning (ML) approach for automatic amino acid identification by utilizing the unique absorption profiles from an Elliptical Dichroism (ED) spectrometer. Advanced data preprocessing techniques and ML algorithms to learn patterns from the absorption profiles that distinguish different amino acids were investigated to prove the feasibility of this approach. The results show that ML can potentially revolutionize the amino acid analysis …

T3-Ciders: Fostering A Community Of Practice In Ci-And Data Enabled Cybersecurity Research Through A Train-The-Trainer Program, Wirawan Purwanto, Mohan Yang, Peng Jiang, Masha Sosonkina

Electrical & Computer Engineering Faculty Publications

We present a training program named T³-CIDERS, the Train- The-Trainer approach to fostering cyberinfrastructure (CI)- and Data-Enabled Research in CyberSecurity. T³-CIDERS is a train-the-trainer program for advanced cyberinfrastructure (CI) skills that is designed to be synergistic with research, teaching, and learning activities in cybersecurity and cyber-related disciplines. The participants, termed 'future trainers' (FTs), are trained in effective instructional design and CI hands-on materials from DeapSECURE, developed in a previous CyberTraining program. T³-CIDERS aims to enhance cybersecurity research and education through broader adoption of advanced CI techniques such as artificial intelligence, big data, parallel programming, and platforms like high-performance computing (HPC) …

Detecting Anomalous Srf Cavity Behavior With Unsupervised Learning, Hal Ferguson, Jiang Li, Adam Carpenter, Chris Tennant, Dillon Thomas, Dennis Turner

Electrical & Computer Engineering Faculty Publications

We present an unsupervised learning framework for detecting anomalous superconducting radio-frequency (SRF) cavity behavior at the Continuous Electron Beam Accelerator Facility (CEBAF), emphasizing its initial performance and effectiveness. Key to the system’s success was the development of data acquisition systems (DAQs) that capture fast-sampled, information-rich signals, essential for detecting transient effects. The approach involves creating daily cavity-specific models using principal component analysis to handle variations in rf signal behavior and mitigate performance degradation from data drift. This unsupervised method eliminates the need for expensive labeling by continuously updating models with recent data. Deployed and operational for 3 months before a …