Digital Commons Network^™

Qlorax: Heuristic-Guided Fine-Tuning Of Llama-2 For Domain Adaptation In Entrepreneurship, Gaurob Saha

Theses and Dissertations

This thesis presents a study on the fine-tuning of large language models (LLMs) for domain-specific applications using limited data. We fine-tuned the LLaMA-2 (7B) model on a curated entrepreneurial dataset containing 3,545 human-written question-answer pairs, of which 3,095 were used for training and 450 were reserved for evaluation. A complete fine-tuning and evaluation pipeline was developed, which included clustering human-written answers, generating centroid-based summaries for each cluster, and evaluating the model's generated responses through cosine similarity.

Training was carried out over five epochs, with model performance evaluated after each epoch. The fine-tuned model demonstrated strong semantic alignment with human-written content, …

Video Comprehension Score (Vcs): A Metric For Long-Form Video Description Evaluation, Harsh Dubey

Electronic Theses and Dissertations

Existing video description evaluation metrics fail to capture the long-range chronology and semantic alignment essential for long-form descriptions. An effective evaluation metric for long-form descriptions must (i) assess global thematic alignment, (ii) measure local semantic alignment, and (iii) evaluate chronological alignment while detecting corrupted content. We introduce Video Comprehension Score (VCS), a reference-based metric, which directly addresses these evaluation requirements through three components: Global Alignment Score for thematic alignment, Local Alignment Score for local semantic alignment, and Narrative Alignment Score for chronological alignment with adjustable tolerance. We evaluate VCS on two large-scale synthetic datasets designed to test corruption detection and …

Deep Learning For Irish Garden Bird Identification: Exploring The Role Of Cnn-Lstm In Video-Based Recognition, Antonina Dolynenko

ICT

Bird populations are widely used as indicators of ecosystem health, but traditional monitoring based on manual observation is labour-intensive and difficult to scale. Recent advances in deep learning and low-cost edge hardware offer new opportunities for automated, real-time bird identification in gardens and other local habitats. This thesis investigates whether video-based deep learning models can reliably classify common Irish garden birds from short motion-triggered clips and how temporal modelling compares to image-based models.

A primary dataset of 20-second clips was collected in a private garden in Ireland using a Raspberry Pi with a high-resolution camera and a YOLO-based trigger to …

Kangaroo: Dynamic Fusion Of Branch Instructions In A Pipelined Uniprocessor, Sarah E. Larkin

Dissertations, Master's Theses and Master's Reports

Small pipelined processors are becoming more common as a complement to superscalars in a multi-core chip. However, current uniprocessors offer little in the way of ILP. We present kangaroo, a novel approach to instruction fusion in a pipelined processor. Kangaroo dynamically fuses two adjacent instructions to create a pair that travels through the pipeline as a unit. The instructions re-enter the pipeline as a pair the next time the first instruction is fetched. Unlike in prior art, an instruction, once fused, is not fetched again. Any pair of adjacent instructions can be fused using this technique, including dependent instructions. …

Agentic Ai Systems In Professional Domains: A Probabilistic Framework For Role Suitability And Legal Accountability, Matthew Veach

Master's Theses and Doctoral Dissertations

This thesis presents a cross-sector analysis of agentic AI systems deployed in STEM, education, healthcare, and enterprise domains, with a focus on role suitability, orchestration maturity, and legal accountability. It introduces the Bounded Agentic Suitability Envelope, a dual-bound scoring framework that evaluates deployment viability using weighted assessments of agent capability and decomposed role complexity. Through case studies from Fujitsu, Cleveland Clinic, Carnegie Learning, and Duolingo, the thesis demonstrates measurable gains in efficiency, personalization, and compliance. It argues for an augmentation-first strategy, preserving human roles in high-context domains while enabling targeted replacement in low-complexity workflows. To mitigate risk, the thesis formalizes …

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 …

Digital Forensics & Artificial Intelligence: Senior Honors Project Research Report, Noah Shelton Kasir

Senior Honors Theses and Projects

This project studied how current artificial intelligence large language models could be used to learn digital forensics and anti-forensics techniques compared to traditional search engines such as Google Search. This project aimed to answer the question “Can an ordinary person use AI to learn both anti-forensics and traditional digital forensics skills effectively and efficiently?”. The project research was divided into two distinct phases. Phase one consisted of the creation of a fictional case by acting as a layperson using the help of the Microsoft Copilot AI tool. This case consisted of a layperson “suspect” attempting to learn multiple anti-forensics techniques …

Assessing Iot Intrusion Detection Computational Costs When Using A Convolutional Neural Network, Mathew Nicho, Brian Cusack, Christopher D. Mcdermott, Shini Girija

All Works

IoT systems face vulnerabilities due to their data processing requirements and resource constraints. With 13 billion connected devices globally, this research investigates the economic viability of AI-based intrusion detection systems (IDSs), specifically analyzing the automation costs of implementing a Convolutional Neural Network (CNN) with Long Short-Term Memory (LSTM) for classifying malicious sensor traffic. This study introduces an innovative framework that evaluates six distinct architectural components of CNN and LSTM: image input processing, convolutional layer operations, max pooling layer functionality, fully connected layer characteristics, softmax output activation, and class determination mechanisms. The framework employs six metrics: matrix size, feature vector number, …

Detecting Sars-Cov-2 In Ct Scans Using Vision Transformer And Graph Neural Network, Kamorudeen Amuda, Almustapha Wakili, Tomilade Amoo, Lukman Agbetu, Qianlong Wang, Jinjuan Feng

Electrical & Computer Engineering Faculty Publications

The COVID-19 pandemic has presented significant challenges to global healthcare, bringing out the urgent need for reliable diagnostic tools. Computed Tomography (CT) scans have proven instrumental in detecting COVID-19-induced lung abnormalities. This study introduces Convolutional Neural Network, Graph Neural Network, and Vision Transformer (ViTGNN), an advanced hybrid model designed to enhance SARS-CoV-2 detection by combining Graph Neural Networks (GNNs) for feature extraction with Vision Transformers (ViTs) for classification. Using the strength of CNN and GNN to capture complex relational structures and the ViT capacity to classify global contexts, ViTGNN achieves a comprehensive representation of CT scan data. The model was …

Time-Marching Quantum Algorithm For Simulation Of Nonlinear Lorenz Dynamics, Efstratios Koukoutsis, George Vahala, Min Soe, Kyriakos Hizanidis, Linda Vahala, Abhay K. Ram

Electrical & Computer Engineering Faculty Publications

Simulating nonlinear classical dynamics on a quantum computer is an inherently challenging task due to the linear operator formulation of quantum mechanics. In this work, we provide a systematic approach to alleviate this difficulty by developing an explicit quantum algorithm that implements the time evolution of a second-order time-discretized version of the Lorenz model. The Lorenz model is a celebrated system of nonlinear ordinary differential equations that has been extensively studied in the contexts of climate science, fluid dynamics, and chaos theory. Our algorithm possesses a recursive structure and requires only a linear number of copies of the initial state …

Problem-Based Learning And Artificial Intelligence, Varna Taranikanti, Srineil Vuthaluru

Books and Book Chapters

In the current digital era, significant advancements in Artificial Intelligence (AI) have influenced various fields worldwide, particularly within the health professional education sector. This progress has also percolated into the medical education system, where Problem-Based Learning (PBL) is routinely used as a key pedagogical strategy. In PBL, students are able to fully exert their exploratory skills by solve problems through a student-centered learning approach. PBL also fosters critical thinking and encourages self-directed learning through the inquiry of clinical case scenarios. This chapter investigates the integration of AI, specifically tools like ChatGPT, into PBL frameworks to enhance learning outcomes for medical …

A Spine-Specific Lexicon For The Sentiment Analysis Of Interviews With Adult Spinal Deformity Patients Correlate With Sf-36, Sf-36, And Odi Scores: A Pilot Study Of 25 Patients, Ross Gore, Michael M. Safaee, Christopher J. Lynch, Christopher P. Ames

VMASC Publications

Classic health-related quality of life (HRQOL) metrics are cumbersome, time-intensive, and subject to biases based on the patient’s native language, educational level, and cultural values. Natural language processing (NLP) converts text into quantitative metrics. Sentiment analysis enables subject matter experts to construct domain-specific lexicons that assign a value of either negative (−1) or positive (1) to certain words. The growth of telehealth provides opportunities to apply sentiment analysis to transcripts of adult spinal deformity patients’ visits to derive a novel and less biased HRQOL metric. In this study, we demonstrate the feasibility of constructing a spine-specific lexicon for sentiment analysis …

Are We Ready For Synchronous Conceptual Modeling In Augmented Reality? A Usability Study On Causal Maps With Hololens 2, Anish Shrestha, Philippe J. Giabbanelli

VMASC Publications

(1) Background: Participatory modeling requires combining individual views to create a shared conceptual model. While remote collaboration tools have enabled synchronous online modeling, they are limited to desktop settings. Augmented reality (AR) offers a new approach by potentially providing the sense of presence found in physical collaboration, which may better support participants in achieving the sense of presence found in physical locations, thus supporting them in negotiating meaning and building a shared model. (2) Methods: Building on prior works that developed technology, we performed a usability study with pairs of modelers to examine their ability at performing key conceptual modeling …

Design And Evaluation Of A Compiler Architecture For First-Class Pattern Matching, Timothy Colaneri

Open Access Master's Theses

This thesis presents the design, implementation, and evaluation of a compiler for Asteroid, a multi-paradigm programming language that introduces first-class pattern matching as a core feature. Unlike conventional pattern matching systems - typically restricted to syntactic constructs like match or case expressions - Asteroid elevates patterns to first-class entities, enabling them to be assigned to variables, passed as arguments, returned from functions, and composed dynamically.

To support this novel abstraction, the compiler is structured as a two-phase system: a Python-based frontend for parsing and syntax analysis, and a Rust-based backend that compiles the language into executable code via a custom …

Scalable Parallel-In-Time Integration For Equations Of Motion, Nathan W. Chapman

All Master's Theses

Physical simulations always need to balance accuracy and run-time. This work implements the Parareal Algorithm using graphics processing units across a distributed system to accurately simulate time-dependent physics while attempting to minimize runtime. Data-transfer latency is identified as the primary bottleneck, for which mitigation methods are provided. Benchmarks comparing single-GPU and distributed implementations on a spectrum of coarse and fine discretizations are analyzed.

Ensemble Learning For Mri-Based Brain Tumor Classification: A Weighted Voting Approach, Ha Anh Vu

All Master's Theses

MRI is essential for detecting and diagnosing brain tumors, where accurately distinguishing glioma, meningioma, and pituitary tumors is vital for effective treatment planning. However, tumors' complex morphology and MRI imaging variations present significant challenges for reliable classification. Deep learning models, particularly Convolutional Neural Networks (CNN) and ResNet architectures, have demonstrated impressive performance in medical image analysis but often struggle with generalization across different datasets. On the other hand, traditional classifiers such as Support Vector Machines (SVM) and K-Nearest Neighbors (KNN) leverage handcrafted features like Histogram of Oriented Gradients (HOG), which can effectively capture structural details but may lack the adaptability …

J-Necora: A Framework For Optimal Resource Allocation In Cloud-Edge-Things Continuum For Industrial Applications With Mobile Nodes, Marco Pettorali, Francesca Righetti, Carlo Vallati, Sajal K. Das, Giuseppe Anastasi

Computer Science Faculty Research & Creative Works

In the Industrial Internet of Things (IIoT) landscape, where the Cloud-to-Things Continuum (C2TC) paradigm is now a reality, industrial applications need to cope with highly heterogeneous network and computing resources. Moreover, many industrial applications also involve Mobile Nodes (MNs). Efficient allocation of network and computing resources to meet the stringent requirements of such applications is often a very challenging task. In this paper, we propose J-NECORA (Joint NEtwork and COmputing Resource Allocation), a comprehensive analytical framework to derive the optimal joint allocation of network and computing resources in the C2TC, that guarantees the application requirements, even in the presence of …

Correction: Yolo-Based Miner Detection Using Thermal Images In Underground Mines (Mining, Metallurgy & Exploration, (2025), 10.1007/S42461-025-01249-6), Cyrus Addy, Venkata Sriram Siddhardh Nadendla, Kwame Awuah-Offei

Computer Science Faculty Research & Creative Works

In the original published article, Figure 3 appears with the Fig. 1 caption, Figure 1 appears with the Fig. 2 caption, and Figure 2 appears with the Fig. 3 caption. The article has been updated to correct this error.

Dynamic Resource Allocation In Cloud-To- Things Continuum For Real-Time Iot Applications, Marco Pettorali, Francesca Righetti, Carlo Vallati, Sajal K. Das, Giuseppe Anastasi

Computer Science Faculty Research & Creative Works

The proliferation of loT devices and the growing demand for real-time applications have driven a shift in the computation paradigm, from Cloud computing to Edge computing, creating the Cloud-to-Things Continuum (C2TC). Many real-time loT applications involve Mobile Nodes (MNs), which may dynamically join or leave. In addition, in future reconfigurable loT systems, applications with different requirements will coexist, and will be dynamically introduced or removed. All this asks for dynamic management mechanisms to ensure the requirements of different real-time applications, even when the system configuration changes over time. In this paper, we propose DJ-NECORA, an online algorithm for the joint …

Feature Fusion Transferability Aware Transformer For Unsupervised Domain Adaptation, Xiaowei Yu, Zhe Huang, Zao Zhang

Computer Science Faculty Research & Creative Works

Unsupervised domain adaptation (UDA) aims to leverage the knowledge learned from labeled source domains to improve performance on the unlabeled target domains. While Convolutional Neural Networks (CNNs) have been dominant in previous UDA methods, recent research has shown promise in applying Vision Transformers (ViTs) to this task. In this study, we propose a novel Feature Fusion Transferability Aware Transformer (FFTAT) to enhance ViT performance in UDA tasks. Our method introduces two key innovations: First, we introduce a patch discriminator to evaluate the transferability of patches, generating a transferability matrix. We integrate this matrix into self-attention, directing the model to focus …

Ca-Vqvae: Cortical Folding Aware Numerical Representation Of White-Matter Structure, Yanjun Lyu, Jing Zhang, Lu Zhang, Tong Chen, Xiaowei Yu, Minheng Chen, Yan Zhuang, Chao Cao, Tianming Liu, Dajiang Zhu

Computer Science Faculty Research & Creative Works

White matter (WM) serves as a fundamental component of the brain providing essential structural support and facilitating the brain cognitive processes. Thus, an accurate and efficient description of the brain's white matter structure is essential for understanding brain function connectivity and development. In this work we used the deep model to combine the information of the WM fiber bundle shape and its related cortical folding patterns together representing the WM fiber bundle from diffusion MRI tractography into a pre-defined low-dimensional space and generate the numerical representation vector. This cortical-aware vector-quantized variational encoder (CA-VQVAE) framework leverages cortical locations and folding patterns …

Secure Data Relay In Federated Digital Twins Of Iot-Enabled Smart Interconnected Factories, Anusha Vangala, Jack Wyeth, Ashok Kumar Das, Sajal K. Das

Computer Science Faculty Research & Creative Works

Smart interconnected factories allow manufacturing units from physically distanced factory sites to communicate classified information needed for additive manufacturing. Each factory has interconnected digital twins of their equipment autonomously managed by a Point-of-Contact digital twin creating a hierarchical system with federated digital twins. The data sharing between the factories must be directed through an edge node responsible for managing multiple factories. We proposed a novel lightweight protocol to prevent the leakage of classified information at any nodes other than the origin and destination digital twins. It leverages elliptic curve cryptography to design a proxy re-encryption scheme with the edge node …

Alertble: Alert Workzone Hazards Using Hybrid Filtering And Machine-Learning-Enabled Ble, Samuel Akinyede, Sejun Song

Computer Science Faculty Research & Creative Works

Collision hazard detection in industrial work zones faces challenges from signal instability, mobility-induced fluctuations, and nonline-of-sight (NLOS) conditions. While Bluetooth low energy (BLE) offers cost-effective proximity sensing, its received signal strength indicator (RSSI) variability - fluctuating by ±10 dBm even at fixed distances - limits reliability in safety-critical applications. This article presents AlertBLE, a hybrid BLE-based hazard detection system that combines extended Kalman filter (EKF) and adaptive moving average (AMA) algorithms to achieve up to 94% RSSI variance reduction in static NLOS conditions. The system introduces speed-aware safety thresholds based on reaction time and braking distance models, dynamically expanding hazard …

V-Usdt: Vision-Based Uav Swarm Detection And Tracking By Leveraging Swarm Formation Constraints, Md Hasibur Rahman, Sanjay Madria

Computer Science Faculty Research & Creative Works

The rapid proliferation of Unmanned Aerial Vehicles (UAVs) and UAV swarm technologies has raised critical concerns about security and safety in low-altitude airspace. In response, we propose a vision-based system for detecting and tracking UAV swarms, which combines a novel UAV detection mechanism with a swarm tracking strategy. Our UAV detector incorporates parallel receptive field blocks alongside an attention mechanism to enhance detection performance. This design effectively captures multiscale features of UAVs while prioritizing salient features, ensuring robust detection under diverse conditions. For swarm tracking, we leverage the inherent formation constraints typically maintained by UAV swarms. These constraints allow us …

Tintin: A Unified Hardware Performance Profiling Infrastructure To Uncover And Manage Uncertainty, Ao Li, Marion Sudvarg, Zihan Li, Sanjoy Baruah, Chris Gill, Ning Zhang

Computer Science Faculty Research & Creative Works

Hardware performance counters (HPCs) enable the measurement of microarchitectural events, which are crucial for tracking and predicting program behavior. High-fidelity measurement and precise attribution are essential for accurate profiling. However, existing profiling tools have fundamental challenges in both aspects. In measurement, numerous events compete for limited hardware monitoring resources; while for attribution, applications have diverse requirements, but systems provide limited support. Existing tools mitigate the former limitation through event multiplexing, but this approach introduces non-trivial errors. The latter limitation, however, remains largely unaddressed. This paper introduces Tintin, an HPC profiling infrastructure with a modular three-component design that addresses both challenges. …

Generative Ai And Finding The Law, Paul D. Callister

Faculty Works

Legal information science requires, among other things, principles and theories. The article states six principles or considerations that any discussion of generative AI large language models and their role in finding the law must include. The article concludes that law librarianship will increasingly become legal information science and require new paradigms. In addition to the six principles, the article applies ecological holistic media theory to understand the relationship of the legal community’s cognitive authority, institutions, techné (technology, medium and method), geopolitical factors, and the past and future to understand the changes in this information milieu. The article also explains generative …

Ai Lawyering Skills Trainers: Transforming Legal Education With Generative Ai, Alexandria Serra

Faculty Works

The integration of generative AI (GenAI) tools in legal education is not just an innovation—it's a transformative shift redefying how law students acquire and refine advocacy skills. This article examines AI’s critical role in modernizing legal education, emphasizing its potential to offer personalized, one-on-one coaching that enhances student learning and engagement. As AI reshapes the legal profession, law schools must evolve to prepare students for an AI-driven future. Serving as a practical guide, this article provides a step-by-step framework for educators and institutions to develop AI tools that simulate real-world courtroom scenarios and provide continuous, personalized feedback. It also highlights …

Evaluating The Evaluators: The Role Of Benchmarks In Legal Ai, Jonathan A. Franklin, Sean Harrington, Christine Hye Won Park

Other Faculty Publications

No abstract provided.

Heuristic Approaches For Coordination Of Heterogeneous Robotic Systems In Harvesting Automation With Size Constraints, Hyeseon Lee

Dissertations, Master's Theses and Master's Reports

This thesis presents the development of path planning algorithms for the coordination of heterogeneous robotic systems while considering size constraints. The objective is to generate practical and efficient solutions for real-world applications. The use of heterogeneous collaborative robots is beneficial in many applications, such as transportation operations in warehouses or manufacturing environments, surveillance, and monitoring, and task allocation and path planning are critical techniques that need to be addressed to deploy in real-world applications. This research focuses on automating lavender harvesting, where robots with varying capabilities must collaboratively navigate complex field layouts to efficiently complete harvesting tasks.

The problem considers …

Leaf-Based Varietal Categorization Of Sweetpotato (Ipomoea Batatas L. Lam.), A Potentially Healthful Vegetable, Using Image Processing And K-Means Clustering, Shahidul Islam, Md Towfiqur Rahman, Md Hamidul Rahman, Abdul Momin

Department of Agricultural and Biological Systems Engineering: Dissertations, Theses, and Student Research

Sweetpotato (Ipomoea batatas Lam) leaves contain higher concentrations of phenolic compounds, flavonoids, and carotenoids that are remarkable in health promotion. However, the nutrient content in sweetpotato leaves varies from variety to variety, and leaf shape and color are the key identifying factors for the varietal classification of sweetpotatoes. So, detecting sweetpotato leaves is essential for the in-situ identification of sweetpotato varieties and for developing intelligent agricultural systems. This study aimed to create a leaf-shape-based varietal classification technique for sweetpotato using image processing techniques coupled with a K-means clustering algorithm. 38 leaf images (RGB) of two sweetpotato cultivars were collected …