UTSA Runner Research Press (Univ. of Texas at San Antonio)
Not a member yet
6846 research outputs found
Sort by
A Two-Stage Hybrid Federated Learning Framework for Privacy-Preserving IoT Anomaly Detection and Classification
Get PDFThe rapid surge of Artificial Internet-of-Things (AIoT) devices has outpaced the deployment of robust, privacy-preserving anomaly detection solutions suitable for resource-constrained edge environments. This paper presents a two-stage hybrid Federated Learning (FL) framework for IoT anomaly detection and classification, validated on the real-world N-BaIoT dataset. In the first stage, each device trains a generative Artificial Intelligence (AI) model on benign traffic only, and in the second stage a Histogram-based Gradient-Boosting (HGB) classifier labels flagged traffic. All models operate under a synchronous, collaborative FL architecture across nine commercial IoT devices, thus preserving data privacy and minimizing communication. Through both inter- and intra-benchmarking against state-of-the-art baselines, the Variational Autoencoder–HGB (VAE-HGB) pipeline emerges as the top performer, achieving an average end-to-end accuracy of 99.14% across all classes. These results demonstrate that reconstruction-driven generative AI models, when combined with federated averaging and efficient classification, deliver a highly scalable, accurate, and privacy-preserving solution for securing resource-constrained IoT environments.Mechanical, Aerospace, and Industrial Engineerin
The Discourse of Multicultural Literacies by Novice Bilingual Teachers in Dual Language Classrooms
Get PDFThe ongoing teacher shortage crisis in the United States, recognized by the U.S. Department of Education in 2020, impacts all students—particularly emergent bilingual students—and intensifies the demand for qualified bilingual educators. Increasingly, novice bilingual teachers are entering Dual Language Bilingual Education (DLBE) classrooms through various pathways, including traditional undergraduate programs, alternative teacher certification programs (ATCPs), and visiting international teacher (VIT) exchange agreements. This dissertation explores how these novice educators engage with and articulate the discourse of multicultural literacies in their early years of practice.
Framed within Critical Multicultural Literacies Pedagogies—grounded in Critical Pedagogy, Funds of Knowledge, and Funds of Identity—this exploratory case study draws from interviews, classroom observations, and educator-generated artifacts. The findings reveal that novice bilingual teachers draw on their lived linguistic and cultural experiences to negotiate and construct their bilingual teaching identities. Through intentional choices in text selection, lesson planning, and the integration of songs and multilingual resources, these educators contribute to the construction of an inclusive environment that resists dominant monolingual and assimilationist ideologies, striving instead for equity-driven instruction that represents all students as part of their dual language community.
Novice bilingual teachers reflect on their instructional practices and demonstrate a deep commitment to validating students’ identities and representation through the use of culturally relevant pedagogy. However, their efforts encounter systemic barriers, including insufficient professional development that addresses their primary needs during the novice years. In addition, tensions and challenges arise from institutional mandates and the ongoing gentrification of DLBE programs. Despite these challenges, the study highlights the transformative potential of multicultural literacies to foster both teacher identity development and equitable learning environments. It calls for sustained and explicit institutional support, critical pedagogical training, and policy reforms that empower novice bilingual educators to sustain and deepen their engagement with multicultural literacies.
This study brings forward the perspectives of novice bilingual teachers and contributes to ongoing conversations in bilingual teacher preparation, multicultural education, and efforts to advance equity in dual language programs.Bicultural-Bilingual Studie
Phenotypic Characterization of pilA, pilB, and pilD Mutants of Acinetobacter baumannii 5075: Impacts on Growth, Biofilm Formation, and Tazobactam Response
Get PDFBackground/Objectives: The Type IV pilus assembly system in <i>Acinetobacter baumannii</i> is a major determinant of its pathogenicity, playing a role in surface-associated functions via the biogenesis of Type IV pili (T4P). Tazobactam (TAZ) is a well-characterized &beta;-lactamase inhibitor, primarily used in combination with &beta;-lactam antibiotics such as piperacillin (PIP) to counteract bacterial resistance mechanisms. While <i>A. baumannii</i> resistance to &beta;-lactam antibiotics has been well studied, the influence of T4P on its susceptibility to TAZ remains largely unexplored. For this reason, we investigated how multidrug-resistant <i>A. baumannii</i> 5075 (AB5075) responds to TAZ by assessing the roles of <i>pilA</i>, <i>pilB</i>, and <i>pilD</i> in bacterial growth and biofilm formation under direct TAZ exposure, with a focus on phenotypic characterization rather than molecular mechanisms. Methods: Bacterial growth kinetics were quantified by measuring the optical densities of cell suspensions and the colony forming units per volume (CFUs/mL) at different time intervals. Time-kill assays and microtiter dish biofilm formation assays were used to evaluate how effectively TAZ can inhibit growth and biofilm formation, respectively. Results: Time&ndash;kill assays confirmed that 32 &micro;g/mL of TAZ inhibited growth in both wild-type (WT) and mutant strains, with the <i>pilD</i> mutant showing initial resistance before eventual inhibition. Biofilm assays showed that the <i>pilA</i> mutant had the highest biofilm formation at 8 h, surpassing the WT strain. A prolonged 32 &micro;g/mL of TAZ exposure (24&ndash;36 h) significantly reduced biofilm production across all strains, with inhibition rates reaching 89% for the WT, 82% for the <i>pilA</i> mutant, 91% for the <i>pilB</i> mutant, and 86% for the <i>pilD</i> mutant. Conclusion: These findings deepen our understanding of the strain-specific roles of T4P components in growth and biofilm regulation in AB5075, and highlight the potential of TAZ as a therapeutic strategy against biofilm-associated infections.Biomedical Engineering and Chemical Engineerin
Investigation of Interaction Between Emergency Management Agencies and Public During Crisis Situations
No full textThe full text of this item is not available at this time because the author has placed this item under an embargo until August 26, 2026.Over the last decade, platforms such as Twitter have reshaped how emergency management agencies warn the public, answer questions, and monitor sentiment during crises accelerating uncertainty reduction, combating rumors, and coordinating relief. Yet three critical questions remain unanswered. First, we have little systematic insight into which official messages people choose to share whether they respond more to concrete instructions or to emotional appeal. Second, the evolution of tone and style in both agency and public posts across the four crisis phases (buildup, impact, chronic response, recovery) has not been fully mapped. Third, the timing interplay between proactive alerts and reactive replies and its influence on how quickly people respond and how long conversations endure lacks precise modeling.
To bridge these gaps, this dissertation presents three complementary studies grounded in the Dual Pathway Model, Speech Act Theory, and Uncertainty Reduction Theory. In chapter 2, we combine topic modeling with negative binomial regression on tweets from Hurricanes Florence (2018), Michael (2018), Nicole (2022), and Winter Storm Uri (2021) to show that utility focused, action oriented messages drive resharing when audiences expect the threat, whereas affect laden, emotional replies better engage those confronting unexpected hazards. In chapter 3, manual coding of speech acts mapped to Fink’s four stage framework reveals a clear “stylistic composition”: directives lead in early phases, solidarity expressions peak at impact, resource centered commitments dominate chronic response, and reflective memorials close out recovery. In chapter 4, mediation analyses demonstrate that, under low ambiguity, authoritative proactive alerts prolong response intervals, while under high uncertainty, interactive reactive posts hasten replies and truncate discussion lifespans.
Together, these studies offer a layered framework mobilization mechanics, rhetorical sequencing, and temporal dynamics to help emergency managers craft phase appropriate, audience tuned, and time sensitive social media strategies that maximize reach, engagement, and community resilience when it matters most.Information Systems and Cyber Securit
Archaeological Report, No. 526
Get PDFFrom November 2024 through January 2025, the Center for Archaeological Research (CAR) of the University of Texas at San Antonio (UTSA) conducted archaeological monitoring of trenching for the San Antonio Parks Police Headquarters Transformer Relocation Project in Hemisfair Park in San Antonio, Bexar County, Texas. The trenching was in response to a request from the Hemisfair Park Area Redevelopment Corporation (HPARC). Excavation, conducted by Klecka Electric Company, Inc., relocated a transformer from behind the Kusch House (41BX579) at 600 Hemisfair Plaza Way #336, to a new location behind the former Park Police Headquarters building at 600 Hemisfair Plaza Way #337. The work was done to accommodate the expansion of the Kusch House to repurpose the structure into a restaurant.
Because the Project Area is on public municipal property, and because there exists the potential for planned work to affect archaeological or historical resources, the project is subject to regulatory review under the City of San Antonio’s (COSA) Unified Development Code (UDC) (Article 6 35-630 to 35-634) and requires review by COSA’s Office of Historic Preservation (OHP). The project also requires review by the Texas Historical Commission (THC) under the Antiquities Code of Texas. Work was performed under THC Texas Antiquities Permit (TAP) No. 31989 issued to Principal Investigator Leonard Kemp. Heather
O’Neal served as the Project Archaeologist.
The final Project Area encompassed 973 m2 (0.24 acre) and included four backhoe trenches. A new site (41BX2670) was identified as well as a modern brick feature and a small assemblage of nineteenth and twentieth century artifacts. Located 94 cm below the surface in the northern end of Trench 1, site 41BX2670 consisted of a deposit of cobbles, concrete, a wooden beam, and associated artifacts. The feature aligns with a one-story wood frame dwelling at 306½ South Street shown on the 1896 Sanborn map. Collected artifacts include charcoal, a clear piece of bottle glass, and a fragment of painted plaster. Following documentation, CAR recommended that site 41BX2670 had no remaining research value. As such, the site is not recommended as eligible for listing on the National Register of Historic Places (NRHP) or as a State Antiquities Landmark (SAL). Both COSA-OHP and the THC agreed with that recommendation and granted permission for the removal of the portion of the site exposed within the trench.
All records generated during the course of this project and all collected artifacts are permanently curated at the CAR curatorial facility. They are accessioned as number 3016.Hemisfair Park Area Redevelopment CorporationCenter for Archaeological Researc
LLMs vs Agents: Repairing Inter-Procedural Vulnerabilities in Real-World Code
Get PDFAutomated vulnerability repair has emerged as a promising alternative to traditional rule-based and static-analysis approaches, which are often limited by high false-positive rates and incomplete coverage. This thesis investigates the comparative effectiveness of fine-tuned large language models (LLMs) and agentic systems for both vulnerability detection and repair, with a particular focus on inter-procedural vulnerabilities. Given that commonly used datasets (e.g., BigVul, CVEFixes) suffer from duplicated and mislabeled samples, the study prioritizes higher-quality datasets, emphasizing improvements introduced in MegaVul and PrimeVul while acknowledging remaining noise.
The model was fine-tuned CodeLlama-13b on MegaVul and PrimeVul for function-level vulnerability detection and repair, and it was evaluated it on test split of PrimeVul to minimize label inaccuracies. The work addresses three core research questions: (i) how dataset size and quality influence fine-tuning performance, (ii) what is the performance of fine-tuned LLMs and agents in vulnerability repair, and (iii) whether agentic approaches handle inter-procedural vulnerabilities better than fine-tuned LLMs.
Experimental results show that dataset selection significantly affects detection performance: fine-tuning on PrimeVul yields a Precision of 0.49, Recall of 0.53, and F1 of 0.58, whereas MegaVul produces a Precision of 0.64, Recall of 0.52, and F1 of 0.57. Agentic systems demonstrate a consistent trade-off—achieving higher precision but lower recall compared to the LLM. For repair, a multi-agent system outperforms both the fine-tuned LLM and ReAct-based agents, reaching a CodeBLEU score of approximately 0.96 and a ≥95% perfect-match rate of around 0.666, while the LLM shows moderate similarity but almost no near-perfect matches.
Experimental results show consistent trade-offs of precision and recall between the agentic systems and fine-tuned LLMs. The recall accuracy of the LLMs was significantly higher than that of the agents, so it tended to detect more vulnerable functions than the agents. However, the LLM's precision was much lower, so it often incorrectly classified benign functions as vulnerable. Results show that agentic systems are 5-11% more accurate than fine-tuned LLMs at detecting inter-procedural vulnerabilities. In conclusion, LLMs are effective vulnerability screeners (breadth/recall), while agents provide higher-confidence decisions and superior patches.Computer Scienc
Disordering Colonial Schooling: An Ethnography of Educational Language Policy and Planning in Saint Lucia
Get PDFThis ethnography of educational language policy and planning conducted in Saint Lucia, explores the Discourses about language structuring the St. Lucia National Language Policy of 2018 and the educational language policies enacted by two fifth-grade teachers throughout the 2022-23 school-year. Data for this dissertation was collected through the retrieval of multiple educational policy texts, the preparation of descriptive fieldnotes about schooling activities, by taking pictures at the research site, and by transcribing a semi-structured interview with the study’s main participants. Framed on border thinking gnoseology, the text analysis revealed ten Discourses about language, transmodernly intertwined into three matrices. In addition, policy enactments are structured by dissimilar matrices of Discourses ordering colonial schooling throughout the school-year and Creole schooling during October. A deeper analysis reveals that these matrices of Discourses are shaped by similar logics emerging from the loci of enunciations Coloniality of power, Global North research, and Kwéyòlness. The dissertation proposes exploring transmodern approaches to plurilingual education to produce linguistically responsive policies to better serve all learners.Bicultural-Bilingual Studie
AI-Driven Optimization of Economic Policy: Taxation
Get PDFArtificial Intelligence adoption in areas like health and trading are well advanced as compared to its adoption in Economics, Public Policy and Social good. As many economies seek ways to be fairer, equitable and efficient, Machine Learning and Deep Learning promises to offer better and efficient solutions. We sought to model and understand how AI can be used to individually assign a dynamic tax rate every month depending on the current macroeconomic factors: inflation, unemployment and individual events such as birth and sickness. To address this, we implemented a Reinforcement Learning (RL) algorithm called Deep Q-Network (DQN). Our framework indicated that the AI model was able to dynamically assign tax rate every month optimizing, GDP, Wealth and Welfare (Disposable income), Government revenue, and equity. According to our model a tax rate of 19% was the optimal point where government and households had maximum benefits. At this optimal tax rate, inequality was reduced to approximately 0.301 and the economy was vibrant due to greater consumer spending. Our approach to optimal taxation is different from many other researches and theories in that, rather than an aggregate tax rate for every income bracket on a progressive model, our AI model treated every individual differently considering that individual’s present and past conditions such as sick days and medical bill, births or partner’s death. This is a fairer and equitable way, since two or three persons may be on the same income but with different past and current life happenings. With this approach, individuals are able to get the financial help they need hence less pressure on government programs, translating into less government spending.Computer Scienc
Modeling, Control, and Protection of Modern Distribution Systems with High Penetration of Inverter-based Resources
No full textThe full text of this item is not available at this time because the author has placed this item under an embargo until August 26, 2026.Modern distribution systems are undergoing rapid transformation due to the growing integration of inverter-based resources (IBRs) and high-power loads such as heavy-duty electric vehicles (HDEVs). These developments pose significant challenges to traditional protection schemes, particularly under faulted grid conditions where the behavior of IBRs and HDEVs diverges from conventional synchronous machines. This dissertation develops advanced control methods for both grid-forming (GFM) and grid-following (GFL) inverters to ensure stable fault ride-through (FRT) in IBR-dominated systems and introduces protection-aware modeling techniques that enhance grid resilience and protection compatibility in modern distribution systems. The proposed approaches enable dynamic voltage support and sufficient negative sequence current injection during asymmetrical faults, improving the sensitivity and selectivity of protection relays. Furthermore, the research extends to the modeling and control of HDEV charging stations, equipped with battery energy storage and bidirectional power capabilities. A key contribution is the development of a control method that allows these stations to remain connected during faults, supplying fault current and reactive power to support grid recovery and protection system operation. A current-limiting framework is introduced to manage the inverters and charger fault response, ensuring voltage regulation and relay coordination without exceeding device ratings. The effectiveness of the proposed methods is validated through electromagnetic transient (EMT) simulations in MATLAB/Simulink and real-time hardware-in-the-loop (HIL) testing using Opal-RT and commercial SEL protection relays, demonstrating their applicability to modern distribution systems with high penetrations of IBRs.Electrical and Computer Engineerin
The Role of Superoxide-inducible Adenylate Kinase in the Pathogenesis of Lyme Disease
No full textThe full text of this item is not available at this time because the author has placed this item under an embargo until August 26, 2030.Lyme Disease is the most prevalent arthropod-borne disease in the United States and is caused by the bacterium Borrelia burgdorferi. Standard treatment for Lyme Disease is antibiotics; however, symptoms may persist after antibiotics in what is called Post-Treatment Lyme Disease Syndrome. With no commercially available vaccine for humans and limited treatment options, a novel approach is necessary to combat Lyme Disease. Identifying novel virulence factors can lead to the development of new therapeutics that target them. In this study, we studied the role that BB0417, a superoxide-inducible adenylate kinase, plays in the pathogenesis of Lyme Disease. This protein aids in maintaining ATP levels in the cell and catalyzes several reactions in the synthesis of DNA and RNA. First, we wanted to observe expression levels of BB0417 in B. burgdorferi in different conditions (pH, temperature, oxidative stress). To do this, we employed SDS-PAGE, an oxidative stress assay, and quantitative real-time PCR. Borrelial mutants lacking a functional superoxide dismutase were employed to define the role of BB0417. Moreover, we aimed to test the essentiality of BB0417 to the survival of B. burgdorferi. To do this, we placed bb_0417 under the control of an IPTG-inducible promoter that would allow us to test for essentiality without having to generate a bb_0417 mutant. The possibility of BB0417 being essential to B. burgdorferi adds another potential antigen that can be targeted to combat Lyme Disease.Integrative Biolog