Mason Journals (George Mason Univ.)
Not a member yet
3256 research outputs found
Sort by
Some Tears of Religious Aspiration: Dynamics of Korean Suffering in Post-War Seoul, South Korea
No full text
The Successes & Challenges of Teaching World History in the Twenty-First Century: Two Case Studies from Western Michigan
No full textGenerating Extended Regular Expressions: A LLM-Based Approach and a Systematic Synthesis Guide
No full textRuntime-verification (RV) tools built off of the Monitoring Oriented Programming framework (MOP), specifically TraceMOP and JavaMOP, verify that traces from executing a piece of software satisfy formalised runtime properties (e.g., “It is unsafe for two threads to use the same Appendable object”). These runtime properties are written in MOP files and typically have five components: a natural language (NL) specification that describes the property, parameters (e.g., Appendable a), events (e.g., safe_append, unsafe_append), handlers (e.g. @fail, @match), and a logic property (e.g., ere : safe_append*). MOP files require logic properties to be written in a supported formal language, one of which is extended regular expression (ERE). ERE is a popular choice, as it is built on the widely used regular expression (RE), but because of additional constructs (e.g., negation, epsilon/empty events), EREs can be more complex and more time-consuming to synthesize than REs. First, we attempt to use large language models (LLMs), specifically GPT-4o, to automate the creation of EREs. We repeatedly refined six LLM prompts, performing preliminary testing on all six and then extensively testing the best performing prompt. Next, we propose a systematic guide developers can follow to produce EREs in a setting that has the NL specification, parameters, events, and handlers defined. This guide transforms parts of the inherently creative procedure of ERE generation into an iterative verification task, and preliminary testing of the guide has shown promising results. Based on the guide, we plan to create an improved system for prompting LLMs. Then, we plan to create a tool that compares semantic equivalence of EREs for the purpose of automating validation. These developments are the first steps in formulating a system that autonomously generates the parameters, events, the logic property, and handlers solely from the NL specification
Hijacking the Host: Parasitic Manipulation of Estuarine Mud Crab Sexual Morphology
No full textParasites can alter the morphology of their hosts to enhance their own survival and reproductive success, manipulating host structures to better accommodate their life cycles. Parasitic barnacles (Rhizocephalans) can alter their crab hosts’ morphology through male feminization. We examined if male Chesapeake Bay mud crabs, Rhithropanopeus harrisii, increased their apron width when infected by the Rhizocephalan Loxothylacus panopaei. Crabs were dissected to confirm infection (N=495), and an image processing software measured the apron area and calculated the relative abdominal width (RAW) (e.g., ratio between the widths of apron segments 3 and 6). A low RAW score represents a broad apron, typically seen in females, while a high RAW score represents a narrow apron, typically seen in males. Infected males showed a significant increase in apron area (e.g., a decrease in RAW score) compared to uninfected males. Additionally, as the stage of infection progressed from virgin infection to sexually mature parasite, RAW scores in male crabs significantly decreased. In females, the RAW score and apron area did not change after infection. The feminization of the male apron may be correlated to the parasite’s need for protection of its egg case, which a male crab’s narrow apron does not provide
Evaluating the Reproducibility of Articles in Management Science
No full textWhen researchers do publish, especially in peer-reviewed journals such as Management Science, there is an unspoken presumption that others can reproduce those findings with the same code and data. But in reality, that doesn't typically happen — or get attempted at all. By actively testing whether replication packages actually work—and how difficult or costly it is to get them running—the purpose of this experiment is to expose the gap between the ideal of reproducible science and the reality researchers face. Articles with available replication packages or supplementary materials were identified and processed through TDX to assess the extent to which reported results could be independently verified. The experiment systematically tracked performance metrics like run time(Avg: 2.087 hours), computational power utilized, expense(Avg: $1.149), and success rates. For results and accessibility verification, each replication package was run on Intel TDX and Google Cloud. Particular attention was paid to monitor replication challenges such as big file sizes, broken code, excessive runtimes, or exorbitantly high computational costs. Faced with a 75% reproduction success rate so far, using TDX and Cloud, the results show that while most Management Science studies with replication material available can be reproduced, a large proportion of it—25%—cannot be reproduced, showing persistent challenges with research transparency and reliability
Adaptive Multimodal RGB–LWIR Sensor Fusion and YOLO Backbone Architectures for Real-Time Object Tracking in Autonomous Turret Systems Across Variable Illumination Conditions
No full textAutonomous robotic platforms are playing a growing role across the emergency services sector, supporting missions such as search and rescue operations in disaster zones and reconnaissance. However, traditional red-green-blue (RGB) detection pipelines struggle in low-light environments, and thermal-based systems fail to capture object characteristics such as color and texture. These complementary limitations suggest that combining thermal and visible imaging may yield more reliable performance across diverse conditions. To address these challenges, this study introduces a unified, adaptive framework that fuses long-wave infrared (LWIR) and RGB video streams at multiple fusion ratios, and dynamically selects the optimal detection model based on the illumination conditions. Using a library of 33 custom-trained you only look once (YOLO) models, we identified the top-performing architectures for the three illumination conditions: no-light (<10 lux), dim light (10–1000 lux), and daylight (>1000 lux). Fusion was performed by blending pixel intensities from aligned LWIR and RGB frames at eleven predefined ratios, from full RGB (100/0) to full LWIR (0/100) in 10% steps. We created a dataset of over 22,000 annotated images across varied illumination using a 75/25 split for model training and validation. Preliminary findings suggest that adaptive RGB–LWIR fusion produced noticeably higher average confidence and firing success rates compared to our baseline models (YOLOv5 and YOLOv11) operating on single modalities. This work lays the foundation for adaptive multimodal perception, improving the reliability of autonomous robotic vision in diverse environments
Comparative Analysis of Small and Large Language Models for Natural Language-based Destination Selection on Real-World Geospatial Data
No full textNatural language interfaces serve as a revolutionary alternative to traditional waypoint entry systems seen in robotics, especially in scenarios where user accessibility and hands-free, rapid path input are critical. Although there have been numerous recent, substantial advancements in language models, there is little research regarding how these models can be used for navigation in real-life geospatial contexts.
In this work, we present an end-to-end working implementation that converts a voice prompt into input for a language model to derive an intended destination and create a route to accurately navigate a rover. We evaluate two different models — a fine-tuned Large Language Model (LLM), Gpt-4o-mini-2024-07-18, and a Small Language Model (SLM), all-MiniLM-L6-v2. While the models each produce their own perceived destination, the route-finding backend is unified under the use of Dijkstra’s algorithm to output a latitude-longitude shortest path from a specified start point to the determined endpoint. We use a real-world OpenStreetMap (OSM) data snapshot of the George Mason University (GMU) Fairfax campus for the scope of our experimentation.
Our evaluation methodology used 50 curated test prompts to establish a baseline for comparison. We surveyed the GMU community with these same prompts, as their response would be representative of the highest likelihood of an intended destination on campus. We used efficiency (execution time) and accuracy (intended destination compared to the baseline) as metrics for evaluating the two models. Our studies showed that the LLM was 4% more accurate and 1.55 seconds slower per prompt on average compared to the SLM
Projecting Ocean Acidification using Ordinary Differential Equations
No full textAs Earth’s atmospheric carbon dioxide (CO₂) levels continue to rise due to industrialization, fossil fuel combustion, and deforestation, the consequences extend far beyond polluted air. A significant portion of this excess CO₂ is absorbed by the world’s oceans, triggering a series of chemical reactions that increase the acidity of seawater. This growing acidity disrupts marine ecosystems, threatening organisms such as coral reefs, shellfish, and plankton that rely on stable carbonate chemistry to survive. This project presents a dynamic mathematical model that uses a a coupled system of ordinary differential equations (ODEs) to track the concentrations of key chemical species involved in ocean acidification, including aqueous CO₂, carbonic acid (H₂CO₃), bicarbonate (HCO₃⁻), carbonate (CO₃²⁻), and hydrogen ions (H⁺). The model captures the reversible reactions that govern the carbonate system in seawater and incorporates temperature-dependent rate constants informed by collision theory and empirical Arrhenius relationships. The model further accounts for how increasing ocean temperatures may also accelerate the rate of acidification, and the ocean’s natural buffering capacity that has been thought to maintain pH stability. By simulating varying levels of atmospheric CO₂ and ocean temperatures, the model explores the limits of this buffering mechanism under future climate scenarios. Our results demonstrate that as atmospheric CO₂ concentrations rise, the buffering capacity of seawater decreases, resulting in a rapid and sustained drop in ocean pH. This decline in pH not only reduces the availability of carbonate ions required for calcifying organisms but also suggests that the ocean’s ability to self-regulate its acidity is weaker than previously assumed. The findings underscore the urgency of reducing anthropogenic CO₂ emissions to protect marine life and preserve the chemical balance of the ocean. This project relates to United Nations SDG goal 14, target 14.3: “minimizing and addressing the impacts of ocean acidification.
Analyzing the Dispersion of Tire Wear Particles in Urban Runoff through Computational Modeling
No full textTire wear particles (TWPs), generated from vehicle tires by friction, are dangerous but understudied pollutants in urban runoff which induce freshwater mortality at even minute concentrations. These particles accumulate in stormwater systems and release toxic byproducts that imperil aquatic ecosystems. While their toxicology and concentrations in stormwater have been documented, the physical mechanisms driving TWP transport across urban surfaces remain poorly understood. Existing research focuses on emission rates and chemical analysis, leaving a gap in predictive modeling on how TWPs migrate into and through stormwater runoff. To address this, we developed a one-dimensional computational model coupling the advection–diffusion equation for pollutant concentration with a modified Burgers equation for runoff velocity. We implemented a stable Crank–Nicolson finite difference scheme and trained an inverse physics-informed neural network to infer parameters of the partial differential equation system under distinct boundary conditions, enabling future calibration with real-world data. Preliminary results suggest that even modest TWP buildup can reshape velocity fields over time, creating downstream concentration hotspots. Our model captures the two-way feedback between particle accumulation and runoff behavior, offering a data-adaptive, physics-based tool to forecast TWP dispersion under real-time weather conditions. This research serves to inform targeted mitigation strategies in line with United Nations Sustainable Development Goals 6 and 14 on Clean Water and Sanitation and Life Below Water, respectively