UTSA Runner Research Press (Univ. of Texas at San Antonio)
Not a member yet
6846 research outputs found
Sort by
Recursive Cognition in Practice: How AI Dialogue Generated and Analyzed Its Own Methodology
Get PDFThis mixed-methods study investigates the cognitive and methodological structure of AI-mediated recursive dialogue as a generative process for scholarly inquiry—and as the analytic engine behind this very study. Using nine full-length transcripts from an unsupervised 56-day research sprint, I tracked how dialogic engagement with ChatGPT catalyzed theory emergence, narrative coherence, and publication-ready academic work. ChatGPT autonomously parsed, coded, and sequenced the transcripts in real time, while I provided the conceptual framework and interpretive synthesis. Through thematic coding and quantitative analysis, I identified patterns of reflection, synthesis, and conceptual pivot points, emphasizing the recursive rhythm of: Prompt → Reflection → Clarification → Synthesis. This study demonstrates how recursive dialogue with AI can produce replicable cognitive insights and contribute to emergent models of methodological co-construction. I argue that when intentionally scaffolded, AI dialogue is not only a valid method of inquiry, but one capable of generating new epistemologies and expediting their study from within.AI, Cyber and Computin
Predicting Eviction Status Using Airbnb Data in the Absence of Ground-Truth Eviction Records
Get PDFThe eviction of tenants is a pressing problem, which is prevalent among low-income renters in the USA, and has devastating consequences. Despite the presence of various measures to combat evictions, identifying high-need regions and tenant groups is highly challenging in many regions due to a lack of access to eviction records (partly because of some infrastructural/policy constraints). In response to this information gap, this paper proposes a solution driven by Machine Learning (ML) to monitor eviction status at various spatial resolutions using Airbnb data when ground-truth eviction data is inaccessible. In particular, we begin by demonstrating the potential of utilizing Airbnb data to build ML-driven methods for distinguishing different neighborhoods across different spatial resolutions with respect to eviction status. We then proceed to develop an ML model capable of learning eviction status levels from Airbnb data, even in the absence of ground-truth labels. Empirical evidence is presented, showcasing the model's performance on par with several robust fully-supervised ML models that had access to ground-truth labels during training. Finally, we conduct a set of cross-region tests to comprehensively study the generalizability of the achieved performance across various unseen regions in the USA that were not used during model training. The code of this project can be accessed via https://github.com/maryam-tabar/Airbnb-Eviction.Information Systems and Cyber Securit
An Empirical Evaluation of Low-Rank Adapted Vision–Language Models for Radiology Image Captioning
Get PDFRapidly growing medical imaging volumes have increased radiologist workloads, creating demand for automated tools that support interpretation and reduce reporting delays. Vision-language models (VLMs) can generate clinically relevant captions to accelerate report drafting, yet their varying parameter scales require systematic evaluation for clinical utility. This study evaluated ten multimodal models fine-tuned on the Radiology Objects in Context version 2 (ROCOv2) dataset containing 116,635 images across eight modalities. We compared four Large VLMs (LVLMs) including LLaVA variants and IDEFICS-9B against four Small VLMs (SVLMs) including MoonDream2, Qwen variants, and SmolVLM, alongside two fully fine-tuned baseline architectures (VisionGPT2 and CNN-Transformer). Low-Rank Adaptation (LoRA), applied to fewer than 1% of selected model parameters, proved optimal among adaptation strategies, outperforming broader LoRA configurations. Models were assessed on relevance (semantic similarity) and factuality (concept-level correctness) metrics. Performance showed clear stratification: LVLMs (0.273 to 0.317 overall), SVLMs (0.188 to 0.279), and baselines (0.154 to 0.177). LLaVA-Mistral-7B achieved the highest performance with relevance and factuality scores of 0.516 and 0.118, respectively, substantially exceeding the VisionGPT2 baseline (0.325, 0.028). Among the SVLMs, MoonDream2 demonstrated competitive relevance (0.466), approaching the performance of some LVLMs despite its smaller size. To investigate performance enhancement strategies for underperforming SVLMs, we prepended predicted imaging modality labels at inference time, which yielded variable results. These findings provide quantitative benchmarks for VLM selection in medical imaging, demonstrating that while model scale influences performance, architectural design and targeted adaptation enable select compact models to achieve competitive results
The Examined Life and the Uncritical Mind: Testimony, Fanaticism, and the Duty to Reflect
Get PDFThis thesis examines the epistemic roots of fanaticism, arguing that it arises from an uncritical reliance on testimony. While testimony is indispensable for human knowledge, allowing individuals to access information beyond personal experience, it also creates the potential for epistemic vulnerability. When testimonial trust becomes unreflective, particularly under conditions of ideological or institutional authority, it can transform into a mechanism of manipulation and dogmatic control.
Drawing on classical and contemporary epistemology, this work explores the tension between epistemic trust and epistemic responsibility. Chapter 1 establishes testimony as a foundational yet precarious source of knowledge. Chapter 2 investigates how uncritical acceptance of testimony leads to fanaticism, using historical analysis and a detailed case study of La Luz del Mundo to illustrate how authority can monopolize testimonial credibility. Chapter 3 develops a virtue-theoretic account of epistemic responsibility, arguing that intellectual humility, open-mindedness, and reflective skepticism are essential for resisting epistemic entrapment. Chapter 4 extends this analysis to social and institutional contexts, identifying educational reform, media literacy, and open discourse as structural conditions for responsible belief.
By connecting analytic epistemology with applied social analysis, the thesis proposes that epistemic responsibility is both an intellectual virtue and a civic necessity. In an age of misinformation and ideological polarization, cultivating reflective engagement with testimony is vital not only for individual understanding but for the health of democratic knowledge systems. The examined life, as Socrates suggested, remains the foundation of truth and freedom.Philosoph
Do Authoritarians Support Political Violence?
Get PDFResearch has linked the authoritarian personality with support for political violence, including violence against the government. However, support for political violence is simultaneously a measure of and an outcome of the authoritarian personality, and one key component (submission to authority) is the antithesis of one key measure of political violence (violence against authority). This article makes three contributions. First, we accentuate the importance of using exogenous measures of the authoritarian personality when estimating its effect on support for political violence. Second, leveraging data from an original survey and the American National Election Studies, we find that the relationship between authoritarianism and support for violence is conditional: it can be positive, negative, or null, depending on who is in control of government and the specificity of political-violence measures. Third, we argue that another concept—the securitarian personality—might better predict support for violence. Access to firearms—which we argue is downstream from securitarianism—consistently predicts support for political violence.Political Science and Geograph
Towards Efficient Continual Learning With Neuronal Importance and Alignment
Get PDFContinual learning, the ability to sequentially learn tasks from changing input distributions, is a challenge for deep neural networks. The key issue is attributed to a phenomenon known as catastrophic forgetting. It is defined as the tendency of the network to forget previously learned knowledge while experiencing new inputs. Biological systems do not suffer from this problem and have inspired researchers to explore techniques inspired by the brain to mitigate catastrophic forgetting. These techniques can alleviate forgetting, but they require additional memory and/or computational overhead. In addition to this, most of the continual learning mechanisms that are biologically inspired and the evaluation frameworks employ backpropagation as a baseline training rule. Backpropagation, on the other hand, relies on instantaneous weight transport and global
updates, both of which are computationally expensive and call into question its neural plausibility. In this work, we introduce continual learning mechanisms that reduce the memory overhead and investigate neuro-inspired training rules instead of backpropagation. We propose a novel regularization approach named NEO that combines neuronal activation-based importance measurement with neuron-state-dependent learning mechanisms to alleviate catastrophic forgetting in both task-aware and task-agnostic scenarios. NEO is a neuronal state-dependent mechanism driven by neuronal activity traces and selective learning rules. For NEO, the storage requirements for regularization parameters grow slower with network size compared to schemes that calculate weight importance, whose storage grows quadratically. The proposed model, NEO, is capable of achieving a performance comparable to other state-of-the-art regularization-based approaches to catastrophic forgetting while operating with a reduced memory overhead.
Following NEO, the role of learning rules that avoid the weight-transport problem is examined in the context of continual learning. We investigate weight estimation approaches that use linear combinations of local and non-local regularization primitives for alignment-based learning. These approaches are coupled with parameter regularization and replay mechanisms to demonstrate robust continual learning capabilities. The layerwise operations observed in alignment-based learning help to boost performance in complex task-aware and task-free scenarios on multiple image classification datasets. We study the dynamics of representational similarity for the learning rules and provide its mapping to the knowledge preservation capabilities of the models. Finally, a framework (ECHO) is released to rapidly understand and compare the compute and resource costs for different continual learning models. ECHO enables training models with alignment-based learning rules and generates meaningful and easy-to-understand visualizations with a few lines of code. In summary, this research examines continual learning from a co-design perspective and provides a tool to understand neuro-inspired efficient strategies for deployment.Electrical and Computer Engineerin
ScooterLab Workshop 2025 - Advancing Research and Collaboration using a Micromobility-supported Sensing and Research Infrastructure
Get PDFThe 2025 ScooterLab Workshop on Advancing Research and Collaboration using a Micromobility-supported Sensing and Research Infrastructure, held in person on May 19, 2025, at the Weston Conference Center in San Antonio, TX, brought together 65 attendees from academia, government, and state research institutions. The event was the first in-person ScooterLab workshop and the first public demonstration of ScooterLab’s integrated micromobility-supported sensing capabilities, including both its on-scooter sensing platform and supporting software systems.
ScooterLab, a community research testbed supported by the National Science Foundation (NSF) under awards #2234516 and #2234517, aims to enable data-driven research across fields such as urban mobility, smart cities, environmental sensing, and privacy/security. The primary goals of the 2025 workshop were to: (1) showcase the operational maturity of the ScooterLab infrastructure, (2) facilitate hands-on demonstrations and discussions with researchers, (3) solicit new collaborative research proposals, and (4) advance interdisciplinary partnerships around mobility-centric data collection and experimentation.
The full-day program included keynotes from national experts in micromobility and research infrastructure, invited talks from NSF and university leadership, seven research presentations across two technical sessions, a panel discussion on micromobility-supported sensing, and a poster and demo showcase. In total, eight new research proposals were submitted in response to the open Call for Collaborations, leading to new project partnerships on the ScooterLab testbed.
The workshop also hosted the first in-person meeting of the Community Advisory Board (CAB), where members reviewed recent hardware and software updates, evaluated progress from early field deployments, and provided feedback on testbed durability, data governance, safety, and scalability. The CAB reaffirmed its support for expanding ScooterLab as a shared infrastructure for national-scale micromobility research.
As a next step, the ScooterLab team is working directly with selected collaborators to begin implementing and supporting their proposed experiments on the testbed. This growing community effort reinforces ScooterLab’s role as a programmable, participatory, and impactful research platform for advancing urban mobility and sensing at scale.
This report documents the workshop agenda, key discussions, and outcomes, and consolidates the next steps identified by participants and the Community Advisory Board. The intent is to use these inputs to guide near-term project onboarding and inform longer-term priorities for scaling ScooterLab and supporting the broader research community.National Science Foundation (NSF), #2234516 and #2234517Computer Scienc
Atmospheric Transport and Composition of Hailstones: Stable Isotopic Trajectories, Organic Matter, and Microplastic Contamination in Texas Supercells =
Get PDFHailstones record the chemical, physical, and anthropogenic processes that occur in the atmosphere acting as a natural archive of environmental composition and storm dynamics. Hailstones from supercell storms in Texas (Burkburnett, 2020; Del Rio, 2020; Uvalde 2021) were analyzed for their stable isotopic composition (δ2H and δ18O), microplastic particulates, and dissolved organic matter chemistry to elucidate how convective storms mediate vertical transport of water, the chemical transformations of atmospheric organic matter, and the chemical species and morphologies of microplastics. Crystallographic analysis coupled with hailstone thin and thick sections revealed the isotopic composition of water controlled by the temperature and water content in storm updrafts. Microplastic analysis using Shimadzu Aim-9000 FTIR spectroscopy indicated 283 particles in all 21 hailstones sampled with morphologies dominated by fragments and fibers, with primary chemical species composed of polycarbonate, epoxy, polyethylene, and nylon; reflecting a diversity of industrial, urban, and agricultural land-use emission sources. 24-hour HYSPLIT back trajectory modeling reflected lower level 500m air masses traversing the Rio Grand Valley manufacturing and population centers. Dissolved organic carbon (TOC = 0.8 to 5.5 mg L-1, mean of 2.1 ± 1.0 mg L-1), fluorescence index (FI = 3.1 to 1.1), and humification (HIX = 0.2 to 4.5) revealing oxygenated, microbially derived, and photochemically processed organic matter characterizing freshly produced secondary organic matter rather than refractory terrestrial humic-like matter. The results show that hailstones retain a record of their isotopic, particulate entrainment and molecular signatures reflecting microphysical processes of biogenic and anthropogenic origin.Civil and Environmental Engineerin
An Evaluation of the Replicable Factor Analytic Solutions Algorithm for Variable Selection: A Simulation Study
Get PDFObserved variable and factor selection are critical components of factor analysis, particularly when the optimal subset of observed variables and the number of factors are unknown and results cannot be replicated across studies. The Replicable Factor Analytic Solutions (RFAS) algorithm was developed to assess the replicability of factor structures—both in terms of the number of factors and the variables retained—while identifying the “best” or most replicable solutions according to predefined criteria. This study evaluated RFAS performance across 54 experimental conditions that varied in model complexity (six-factor models), interfactor correlations (ρ = 0, .30, and .60), and sample sizes (n = 300, 500, and 1000). Under default settings, RFAS generally performed well and demonstrated its utility in producing replicable factor structures. However, performance declined with highly correlated factors, smaller sample sizes, and more complex models. RFAS was also compared to four alternative variable selection methods: Ant Colony Optimization (ACO), Weighted Group Least Absolute Shrinkage and Selection Operator (LASSO), and stepwise procedures based on target Tucker–Lewis Index (TLI) and ΔTLI criteria. Stepwise and LASSO methods were largely ineffective at eliminating problematic variables under the studied conditions. In contrast, both RFAS and ACO successfully removed variables as intended, although the resulting factor structures often differed substantially between the two approaches. As with other variable selection methods, refining algorithmic criteria may be necessary to further enhance model performance.Operations and Analytic
New Permutation-Free Quantum Circuits for Implementing 3- and 4-Qubit Unitary Operations
Get PDFThe article presents the quantum signal-induced heap transform (QsiHT) method of the QR-decomposition of multi-qubit operations. This transform can be generated by a given signal, by using different paths, or orders, of processing the data. We propose using the concept of the fast path of calculation of the QsiHT and applying such transforms on each stage of the matrix decomposition. This allows us to build quantum circuits for multi-qubit unitary operation without permutations. Unitary operations with real and complex matrices are considered. The cases of 3- and 4-qubit operations are described in detail with quantum circuits. These circuits use a maximum of 28 and 120 Givens rotation gates for 3- and 4-qubit real operations, respectively. All rotations are performing only on adjacent bit planes. For complex unitary operation, each of the Givens gates is used in pairs with two Z-rotation gates. These two types of rotations and the global phase gate are the universal gate set for multi-qubit operations. The presented approach can be used for implementing quantum circuits for <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mi>n</mi></mrow></semantics></math></inline-formula>-qubits when <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mi>n</mi><mo>&ge;</mo><mn>2</mn></mrow></semantics></math></inline-formula>, with a maximum of (<inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><msup><mrow><mn>4</mn></mrow><mrow><mi>n</mi></mrow></msup><mo>/</mo><mn>2</mn><mo>&minus;</mo><msup><mrow><mn>2</mn></mrow><mrow><mi>n</mi><mo>&minus;</mo><mn>1</mn></mrow></msup><mo stretchy="false">)</mo></mrow></semantics></math></inline-formula> Givens rotations and no permutations.Electrical and Computer Engineerin