Digital Commons Network^™

Towards Secure And Privacy-Preserving Machine Learning Systems, Han Liu

McKelvey School of Engineering Graduate Student Theses & Dissertations

In recent years, machine learning (ML) has advanced at an unprecedented pace, driving the widespread adoption of increasingly sophisticated models across a broad range of real-world applications, including healthcare, finance, autonomous systems, and critical infrastructure. While these models have delivered remarkable benefits and transformed numerous industries, they remain inherently vulnerable to a variety of security and privacy threats. Given their growing role in safety-critical domains, ensuring their security and privacy has become imperative. Systematically addressing these vulnerabilities requires comprehensive adversarial analyses to uncover weaknesses and inform the design of robust defenses. This dissertation systematically investigates ML vulnerabilities in adversarial settings …

Voice Interaction With Conversational Ai Could Facilitate Thoughtful Reflection And Substantive Revision In Writing, Jiho Kim, Philippe Laban, Xiang 'Anthony' Chen, Kenneth C. Arnold

University Faculty Publications and Creative Works

Writing well requires not only expressing ideas but also refining them through revision, a process facilitated by reflection. Prior research suggests that feedback delivered through dialogues, such as those in writing center tutoring sessions, can help writers reflect more thoughtfully on their work compared to static feedback. Recent advancements in multi-modal large language models (LLMs) now offer new possibilities for supporting interactive and expressive voice-based reflection in writing. In particular, we propose that LLM-generated static feedback can be repurposed as conversation starters, allowing writers to seek clarification, request examples, and ask follow-up questions, thereby fostering deeper reflection on their writing. …

Optimizing Small Ai Models For Biomedical Tasks Through Efficient Knowledge Transfer From Large Domain Models, Girish Sundaram

Theses and Dissertations

The PICO (Population, Intervention, Comparison, Outcome) framework is a widely adopted methodology for structuring clinical research questions and extracting relevant information from unstructured medical texts. However, traditional approaches for PICO classification demand computationally expensive domain-specific language models, such as BioBERT and ClinicalBERT, which require extensive training and large annotated datasets. This dissertation introduces Distilled Rapid Embedding Transfer (DRET), a novel knowledge transfer method designed to enable resource-constrained domain adaptation. DRET aims to efficiently transfer biomedical domain knowledge from large, specialized models to a compact, general-purpose model, DistilBERT, thereby enhancing its ability to perform domain-specific tasks without access to the original …

Generative Artificial Intelligence Dependency: Scale Development, Validation, And Its Motivational, Behavioral, And Psychological Correlates, Adalia Yin Hui Goh

Dissertations and Theses Collection (Open Access)

The growing integration of generative artificial intelligence (AI) into everyday life has raised questions about its potential psychological and behavioral consequences. The present research develops and validates the Generative AI Dependency Scale, a multidimensional tool developed to assess individual differences in dependency on generative AI systems. Across six studies involving 1,223 participants from the United States and Singapore, the Generative AI Dependency Scale demonstrated strong psychometric properties, including a stable three-factor structure (cognitive preoccupation, negative consequences, withdrawal) and good test-retest reliability (ICC = .85). Confirmatory factor analysis supported a higher-order dependency construct, and scalar measurement invariance was established across sex …

Usage Of Natural Language Processing And Deep-Learning Techniques On Thematic Apperception Tests To Predict Big Five Personality Traits, Blayten Jones

Electrical Engineering and Computer Science Undergraduate Honors Theses

The usage of personality as a method of behavioral prediction and outcomes of success has grown considerably over the last few decades. This project explores predicting user personality profiles via the Big Five personality index through the integration of advanced natural language processing techniques as well as neural networks. Using a dataset provided by Dr. James W. Pennebaker, participants analyze an image—formally referred to as a thematic apperception test—and write a thorough paragraph describing the details. This free-form text, along with their personality test results, is captured in a structured dataset. Many deep-learning and machine learning models have been used …

Evaluation Of Pre-Trained Vision Language Models In Challenging Contexts, Kankan Zhou

Dissertations and Theses Collection (Open Access)

The rapid advancement and proliferation of pre-trained vision-language models (VLMs) have heralded a new era in the realm of artificial intelligence (AI), opening up unprecedented opportunities and challenges alike. This dissertation sets forth on an ambitious and comprehensive journey to critically evaluate the performance and limitations of pre-trained VLMs, particularly in complex and challenging contexts that test the bounds of their capabilities. Our focus is twofold: to rigorously assess the extent of bias embedded in these models, and to meticulously scrutinize their reasoning abilities, highlighting parallels and disparities between machine and human cognition.

We initiate our exploration with a targeted …

Advancing Drug-Drug Interaction Prediction Using Multi-Modal Feature Integration With Graph Neural Networks, Ernest C. Chianumba

Theses, Dissertations and Culminating Projects

Pharmaceutical treatments are essential for managing medical conditions, but drug-drug interactions (DDIs) pose significant risks to patient safety and healthcare outcomes. This research integrates Knowledge Graphs and Graph Neural Networks to predict DDIs by exploring complex drug relationships. We construct a comprehensive knowledge graph using DrugBank data (1,000 drugs, 155,774 interactions) enriched with molecular features from PubChem. Our methodology introduces a novel multi-modal approach by integrating transformer-based embeddings (ChemBERTa, SPECTER, and SBERT) to create 1152-dimensional feature vectors that capture structural, biomedical literature, and semantic properties of drugs. Formulating DDI prediction as a link prediction task, we compare three Graph Neural …

An Analysis Of Bias Towards Women In Large Language Models Using Likert Scale Evaluations, Sarah T. Fieck

Electrical Engineering and Computer Science (MS) Theses

Closed-source large language models (LLMs) developed by large technology companies continue to grow in popularity. However, ethical conversations surrounding the safety of model outputs have been a prominent topic of discussion. This project aims to assess three leading closed-source LLMs: OpenAI’s ChatGPT, Google’s Gemini, and Anthropic’s Claude, to analyze how their outputs perform when treated as a subject of several psychological evaluation scales measuring biased behaviors against women. The Ambivalent Sexism Index, Modern Sexism Scale, and Belief in Sexism Shift evaluations were used to get descriptions of how the LLMs respond to traditional and modern prompts involving sexism and gender …

Survey On Application Of Large Language Models In Network Attack And Defense, Prisha Purohit

Honors Theses

The emergence of Large Language Models (LLMs) has significantly transformed the technological and cybersecurity landscape, introducing both unprecedented opportunities and formidable challenges. With the public release of ChatGPT in 2022, LLMs have gained global prominence, redefining natural language processing capabilities and enabling advancements across various fields. In cybersecurity, these models represent a dual-use technology: while they offer powerful tools for threat detection, automated analysis, and security training, they also pose risks when leveraged by malicious actors for phishing, social engineering, and the creation of evasive malware. This thesis presents a comprehensive literature review exploring the dual roles of LLMs in …

Phoneme Recognition For Pronunciation Improvement, Matthew Heywood

Theses/Capstones/Creative Projects

This project aims to improve English pronunciation by investigating speech errors and developing a tool to provide precise feedback. The study focuses on creating a new pronunciation tool that offers localized feedback, identifies specific errors, and suggests corrective measures. By addressing the shortcomings of current methods, this research seeks to enhance pronunciation refinement.

Utilizing cutting-edge technology, the tool leverages speech-to-phoneme AI models and modified lazy string matching algorithms to compare the user's spoken input with the intended pronunciation. This allows for a detailed analysis of discrepancies, providing users actionable insights into their phonetic errors. The speech-to-phoneme AI models mark a …

Multimodal Learning For Visual Perception And Robotic Action, Taisei Hanyu

Electrical Engineering and Computer Science Undergraduate Honors Theses

Multimodal learning aims to weave information from images, language, depth, and other sensors into one coherent representation, much as people naturally combine sight, speech, and sound. Progress toward that goal is slowed by three gaps: vision encoders that cannot balance crisp object boundaries with global context, 3-D semantic maps that are computationally prohibitive for real-time, open-vocabulary queries, and vision-language-action pipelines that depend on large token pools with weak relational grounding.

We first introduce AerialFormer, a lightweight hybrid of convolutional and Transformer layers that captures long-range structure without sacrificing fine detail. On the large-scale iSAID benchmark it reaches 69.3% mean IoU, …

Modeling Language And Vision At Human Scales, Clayton Fields

Boise State University Theses and Dissertations

The impressive results that have recently been achieved in natural language processing and artificial intelligence have been primarily driven by the introduction of the transformer deep learning architecture, increasingly large models with many parameters and using enormous datasets. The size of models and their training data requirements present costly demands that freeze many researchers out of training with cutting edge models. Beyond these practical implications, current methods learn from text alone, without the rich array of sensory information that human beings use in learning language. This means that language models are often incapable of reasoning about the concrete world that …

Towards Supporting Children's Metacomprehension During Web Search, Christine Dianne Pinney

Boise State University Theses and Dissertations

As children interact online with search engine result pages (SERPs), children’s comprehension monitoring skills are put to work. Children are known to struggle with utilizing online information as the text content can be misaligned with their reading abilities. Comprehension monitoring skills allow people to assess the comprehensibility of text and is measured by how well comprehension predictions align with performance on comprehension tests. Previous research has shown that augmenting standard SERPs with readability visual cues can be helpful for children searching online. While comprehension predictions are traditionally collected after reading, SERP information provides the opportunity to measure children's perceived comprehensibility …

Exploring The Biasing Effects Of Gender On Personality Disorder Diagnoses Formulated By Artificial Intelligence, Zoe Colclough

Student Theses

Gender bias is prevalent in personality disorder assessments, and while artificial intelligence has been posited as a solution to improve diagnostic objectivity and accuracy, the potential for such technologies to propagate human gender bias in mental health contexts remains underexplored. This study investigated the influences of gender bias on the diagnostic performance of ChatGPT-4o for personality disorders using three factorial research designs, which involved experimentally manipulating patient gender in a combined sample of 360 vignettes and case studies. Vignettes were synthesized through a novel artificial intelligence-assisted methodology established for this research, and case studies were identified from the literature. Significant …

Query Understanding In Llm-Based Conversational Information Seeking, Yifei Yuan, Zahra Abbasiantaeb, Yang Deng, Mohammad Aliannejadi

Research Collection School Of Computing and Information Systems

Query understanding in Conversational Information Seeking (CIS) involves accurately interpreting user intent through context-aware interactions. This includes resolving ambiguities, refining queries, and adapting to evolving information needs. Large Language Models (LLMs) enhance this process by interpreting nuanced language and adapting dynamically, improving the relevance and precision of search results in real-time. In this tutorial, we explore advanced techniques to enhance query understanding in LLM-based CIS systems. We delve into LLM-driven methods for developing robust evaluation metrics to assess query understanding quality in multiturn interactions, strategies for building more interactive systems, and applications like proactive query management and query reformulation. We …

Building Bridges Across Papua New Guinea’S Digital Divide In Growing The Ict Industry, Marc Cheong, Sankwi Abuzo, Hideaki Hata, Priscilla Kevin, Winifred Kula, Benson Mirou, Christoph Treude, Dong Wang, Raula Gaikovina Kula

Research Collection School Of Computing and Information Systems

Papua New Guinea (PNG) is an emerging tech society with an opportunity to overcome geographic and social boundaries, in order to engage with the global market. However, the current tech landscape, dominated by Big Tech in Silicon Valley and other multinational companies in the Global North, tends to overlook the requirements of emerging economies such as PNG. This is becoming more obvious as issues such as algorithmic bias (in tech product deployments) and the digital divide (as in the case of non-affordable commercial software) are affecting PNG users. The Open Source Software (OSS) movement, based on extant research, is seen …

Cross-Dataset Fairness Evaluation Of Transformer-Based Sentiment Models, Sara Zuiran

Theses and Dissertations

With the growing exploration of Natural Language Processing (NLP) systems in decision-making environments, it is essential to evaluate technical and ethical aspects of the dataset and the NLP model to improve fairness. To assess fairness, the thesis examines demographic imbalances in sentiment classification models by evaluating transformer-based models fine-tuned on the Stanford Sentiment Treebank version 2 dataset (SST-2) against the demographically annotated Comprehensive Assessment of Language Model dataset (CALM). This work identifies performance disparities in sentiment prediction across demographic groups by examining sensitive attributes such as gender and race. The study evaluates both the RoBERTa and MentalBERT transformer models using …

Climate Change And The Field Of Farm Labor In The Lower Rio Grande Valley, Texas, Cody Richard Mortell

Theses and Dissertations

Farm workers in the United States are often vulnerable to adverse impacts of climate change, particularly extreme heat. This study explores farm workers’ climate change perceptions (CCP) as a mediating force in climate adaptation. Using latent class analysis, in-person survey data (n=404) from Lower Rio Grande Valley, TX (2024) and NOAA data were analyzed. Results were interpreted using Bourdieu’s field analysis framework and his notions of capital and symbolic power. Findings include: a social hierarchy with three latent classes of workers based on economic, cultural, social, and symbolic capital, which statistically significantly affected CCP. Workers in higher structural positions were …

Toward The Application Of Natural Language Processing In Electronic Health Record Analysis For Taxonomy Development, Latoya Mcdonald

All Dissertations

Electronic health records (EHRs) are pivotal resources for nurse practice because they increase the timeliness and reliability of patient information at the point of care and support access by multiple healthcare providers and the individual patients themselves. However, it is widely recognized that data extraction from EHRs is challenging due to the variability in the language used in clinical care notes and the lack of standardized terminology across healthcare systems. The broad objective of this dissertation is to develop taxonomy-based classification models for nursing care by applying feature engineering approaches to EHRs that include nursing care of ostomy patients following …

Cache-Augmented Generation In Rag Pipelines: Fast And Memory-Efficient Approach To Multi-Agent Knowledge Query Systems, Yaju Gopal Shrestha

Honors Theses

This thesis presents an implementation and evaluation of Cache-Augmented Generation (CAG) for knowledge query systems, building upon the approach introduced by Chan et al. (2024). Traditional Retrieval-Augmented Generation (RAG) systems (Lewis et al., 2020) face challenges including high latency, excessive memory usage, and complex infrastructure requirements. By implementing a cache-augmented architecture that preloads relevant knowledge and eliminates real-time retrieval, our approach significantly improves response time while reducing resource requirements. The research demonstrates the effectiveness of CAG through a comprehensive implementation for The University of Southern Mississippi's chatbot system, achieving a 49.02% improvement in response time compared to traditional RAG approaches. …

A Sociocognitive Perspective On The Activation Of Productive Epistemic Resources In College Students' Understanding Of Chemical Equilibrium, Henk Steven Van Den Bogaard

Open Access Theses & Dissertations

Science education, particularly in higher education, has a content coverage problem: it has long been described as "a mile wide, an inch deep" in various consensus documents from the National Academies. In chemistry education, the most common general chemistry curriculum covers topics as chapters based on the popular Sienko and Plane 1960's general chemistry textbook. The limited time available in such fast-paced coverage may limit the instructor's modeling and assessment of key performances in developing a mechanistic understanding of complex chemistry phenomena. In this sociocognitive, qualitative investigation, general chemistry II students' responses to assessment prompts were used to activate cognitive …

Machine Learning And Time Series Forecasting For Hydropower Predictions, Jose Reynaldo Vega

Open Access Theses & Dissertations

Recent advancements in machine learning have led to the design of many neural network architectures aimed at solving real-world problems. Each network works to make predictions by finding patterns in the provided data. One common application is time series forecasting, where a model predicts future events based on historical time series data. Time series forecasting is used in a variety of fields, one example being in predicting water releases of Hybrid Floating Photovoltaic-Hydropower (HFPVH) systems. As global population growth drives an increase in energy demand, the need for resilient and sustainable energy generation has become urgent. HFPVH systems have emerged …

Towards Multimodal Scene Graph Generation Approaches To Video Understanding, Trong-Thuan Nguyen

Graduate Theses and Dissertations

This thesis advances video understanding by enhancing Video Scene Graph Generation (VidSGG) through improved temporal modeling, the integration of long-range temporal dependencies via continuous updates to interaction histories, and the utilization of Large Language Models (LLMs) for scene graph reasoning. To this end, three novel datasets and corresponding approaches are introduced. First, the ASPIRe dataset incorporates interactivity annotations and leverages the Hierarchical Interlacement Graph (HIG) for hierarchical temporal modeling, providing deep insights into scene changes and effectively capturing intricate interactions. Next, the AeroEye dataset, focusing on drone videos, is paired with the Cyclic Graph Transformer (CYCLO), which establishes circular connectivity …

Single Valued Neutrosophic Hypersoft Set Based On Merec-Moosra Mcdm Methods For Evolution Of Sustainable Strategies For The Circular Supply Chain Based On The Beverage Industry, Alaa Salem, Rayan Hussein, Ahmed Abdelmouty, Mohamed Abouhawwash

Neutrosophic Systems with Applications

Corporate enterprises around the world are facing increasing pressure to operate sustainably due to rising environmental concerns such as climate change, resource scarcity, and ecological degradation. In response, circular supply chain (CSC) practices have emerged as a promising solution, especially within the manufacturing and beverage sectors. CSC focuses on reusing materials, minimizing waste, and creating value from products that have reached the end of their lifecycle. This study investigates the challenges and opportunities of applying circular supply chain management (CSCM) in the beverage industry, which is known for generating significant waste due to high production volumes. The research identifies key …

Strengthening The Bonds Between Us: An Empirical Investigation Of Morale In Human-Ai Teams And The Socially Supportive Ai Teammates Who Empower It, Rohit Mallick

All Dissertations

This dissertation investigates how artificial intelligence (AI) can be designed to improve the collective emotion within a team. A team's collective emotion, or morale, describes how motivated, optimistic, and enthusiastic the group is in accomplishing its goals. We conducted four studies that compared different social support strategies that AI teammates can provide to the team. Study 1A found that AI teammates who communicate with emotions can better motivate human team members and promote awareness of team dynamics and environmental changes. Study 1B found that human teammates become more motivated and happier when their AI teammates express joy and are close …

Unlocking The Planning Capabilities Of Llms Through Maximum Diversity Fine-Tuning, Wenjun Li, Changyu Chen, Pradeep Varakantham

Research Collection School Of Computing and Information Systems

Large language models (LLMs) have demonstrated impressive task-solving capabilities through prompting techniques and system designs, including solving planning tasks (e.g., math proofs, basic travel planning) when sufficient data is available online and used during pre-training. However, for planning tasks with limited prior data (e.g., blocks world, advanced travel planning), the performance of LLMs, including proprietary models like GPT and Gemini, is poor. This paper investigates the impact of fine-tuning on the planning capabilities of LLMs, revealing that LLMs can achieve strong performance in planning through substantial (tens of thousands of specific examples) fine-tuning. Yet, this process incurs high economic, time, …

Towards Multimodal Empathetic Response Generation: A Rich Text-Speech-Vision Avatar-Based Benchmark, Han Zhang, Zixiang Meng, Meng Luo, Hong Han, Lizi Liao, Erik Cambria, Hao Fei

Research Collection School Of Computing and Information Systems

Empathetic Response Generation (ERG) is one of the key tasks of the affective computing area, which aims to produce emotionally nuanced and compassionate responses to user's queries. However, existing ERG research is predominantly confined to the singleton text modality, limiting its effectiveness since human emotions are inherently conveyed through multiple modalities. To combat this, we introduce an avatar-based Multimodal ERG (MERG) task, entailing rich text, speech, and facial vision information. We first present a large-scale high-quality benchmark dataset, AvaMERG, which extends traditional text ERG by incorporating authentic human speech audio and dynamic talking-face avatar videos, encompassing a diverse range of …

Seaexam And Seabench: Benchmarking Llms With Local Multilingual Questions In Southeast Asia, Chaoqun Liu, Wenxuan Zhang, Jiahao Ying, Mahani Aljunied, Anh Tuan Luu, Lidong Bing

Research Collection School Of Computing and Information Systems

This study introduces two novel benchmarks, SeaExam and SeaBench, designed to evalu ate the capabilities of Large Language Models (LLMs) in Southeast Asian (SEA) application scenarios. Unlike existing multilingual datasets primarily derived from English translations, these benchmarks are constructed based on real world scenarios from SEA regions. SeaExam draws from regional educational exams to form a comprehensive dataset that encompasses sub jects such as local history and literature. In contrast, SeaBench is crafted around multi turn, open-ended tasks that reflect daily inter actions within SEA communities. Our evalua tions demonstrate that SeaExam and SeaBench more effectively discern LLM performance on …

Towards Explainable And Robust Nlp: Neutrosophic Probability Augmentation In Text Classification, Nabil M. Abdel-Aziz, Mahmoud Ibrahim, Khalid A. Eldrandaly

Neutrosophic Systems with Applications

The rapid growth of textual data necessitates advanced text classification models. However, traditional methods struggle with ambiguity and uncertainty in natural language, reducing classification reliability. To address this, we integrate neutrosophic logic, which explicitly models truth, indeterminacy, and falsity, into a DistilBERT-based text classification framework. Additionally, we employ data augmentation using synonym replacement to enhance generalization. Our approach is evaluated on the AG News dataset, classifying articles into four categories: World, Sports, Business, and Science/Technology. By incorporating neutrosophic attributes, the proposed framework assesses text quality, mitigates uncertainty, and improves robustness against ambiguous inputs. Experimental results demonstrate an accuracy of 94.10%, …

Reverse Modeling In Large Language Models, Sicheng Yu, Yuanchen Xu, Cunxiao Du, Yanying Zhou, Minghui Qiu, Qianru Sun, Hao Zhang, Jiawei Wu

Research Collection School Of Computing and Information Systems

Humans are accustomed to reading and writing in a forward manner, and this natural bias extends to text understanding in auto-regressive large language models (LLMs). This paper investigates whether LLMs, like humans, struggle with reverse modeling, specifically with reversed text inputs. We found that publicly available pre-trained LLMs cannot understand such inputs. However, LLMs trained from scratch with both forward and reverse texts can understand them equally well during inference. Our case study shows that different-content texts result in different losses if input (to LLMs) in different directions---some get lower losses for forward while some for reverse. This leads us …