Mason Journals (George Mason Univ.)
Not a member yet
3256 research outputs found
Sort by
Crime around public transit: Analyzing spatio-temporal variations in crime rates near D.C. Bus Stops
No full textPublic transportation is a cost-effective and sustainable alternative to driving. Yet, crime in large cities can discourage people from using public transit, especially in places like Washington, D.C. This project examines the spatiotemporal variations in crime rates around bus stops throughout the Washington, D.C. area. Crimes were grouped into four categories: assault, theft, homicide and sexual abuse. We explored which crime types were most prevalent around public transportation and how their occurrence patterns vary across time of the day. To study this, we collected location based crime data for 2024 from Opendata DC and bus stop locations from Mobility Database. We mapped all the bus stops and crime locations and created small zones, or “buffers,” of 0.25 miles around each bus stop and counted how many crimes occurred within each zone. We performed this analysis by crime types and time of day (day, evening and midnight). We also repeated this crime pattern analysis using 0.125 and 0.4 mile buffers to understand the sensitivity of our findings at various spatial scales. The analysis was performed using ArcGis Pro and Python. We found that assault was the most frequent crime and that more crimes occurred at night than during the day. We also found that certain crimes, such as theft, were more common downtown, while other crimes, such as homicide and sexual abuse, were more common near bus stops farther from central D.C. Specifically, we found that in the mornings there were zero homicide cases within 0.25 miles of a bus stop; however, in the night time, there were multiple bus stops with 10 or more homicides. This study shows how the time and location of crimes are connected to public transportation which in return could help city leaders make D.C. safer for the public
Evaluating Secure Virtualization in Academic Research: Replicating Economics Studies Using Microsoft TDX
No full textReplication is a cornerstone of scientific progress, but replicating empirical studies often raises challenges related to environment variability, data access, and reproducibility. Trusted Execution Environments (TEEs), like Intel’s TDX (Trust Domain Extensions), offer a promising framework for secure and standardized replication in academic research. This project evaluates Microsoft’s implementation of TDX as a platform for replicating empirical papers in economics, specifically targeting publications from Management Science. Despite widespread use of cloud computing, many replication efforts lack isolation from host environments and leave security vulnerabilities that TDX seeks to solve. We focused on deploying replication packages for peer-reviewed studies using Microsoft TDX virtual machines. Each package was evaluated on runtime, credit usage, reproducibility success, and environment compatibility. Early results show that while most Stata-based workflows run successfully with minor adjustments, some packages involving licensed software or large datasets require specialized scripts and manual intervention. We developed standardized logging templates and troubleshooting protocols to improve efficiency and team-wide reproducibility. Our methodology emphasizes independence and team collaboration, two students replicate each paper in parallel and compare results before final reporting. This work demonstrates that with appropriate setup and documentation, Microsoft TDX can streamline reproducibility efforts while preserving computational security. These findings may help journals and conferences evaluate TDX as a scalable solution for future replication standards
Digitizing Commodities: A Look at Real-World Asset Tokenization
No full textPrecious metals have long been considered safe assets, yet their physical nature often limits liquidity, divisibility, and global accessibility. With the emergence of tokenized commodities, a fundamental shift can be seen in how traditional assets are digitized and traded. However, systematic analysis of these products remains limited. This study applies a rigorous due diligence framework to analyze eight tokenized commodity products listed on rwa.xyz. These products are issued across diverse jurisdictions, including the U.S., U.K., Liechtenstein, UAE, and British Virgin Islands. Each product was evaluated across legal structure, regulatory compliance, redemption rights, and peg mechanisms. The analysis combined a qualitative legal review with on-chain analytics from Etherscan, revealing significant differences in transparency and market engagement. For example, Paxos Gold showed high adoption with over 68,000 holders and nearly $900 million in monthly transfer volume, while products such as tPlatinum and VNX Gold had minimal usage. The findings indicate that the RWA tokenization of precious metals creates efficiency through 24/7 trading and fractional ownership while introducing new regulatory and custodial requirements, showing that successful commodities require legal frameworks and transparent operations to bridge traditional and decentralized finance ecosystems. This research and methodology lay the groundwork for future academic and industry assessments of tokenized commodity markets
Evaluating Intel TDX for Secure Verification of Replication Packages
No full textReplication packages are a collection of raw data and coding scripts that can be used to recreate the results of a research paper. Without secure protections, these packages are vulnerable to tampering and data leakage, which can lead to misinformed decisions and reduced trust in academic findings. This study evaluates Intel Trust Domain Extensions (Intel TDX), a confidential computing technology that runs code in secure, isolated environments, in order to verify replication packages from Management Science (2024). Each package was tested using a TDX-enabled virtual machine on Google Cloud, and the results were compared to published outputs. Intel TDX successfully protected against unauthorized access, but technical limitations, such as file input and output restrictions, limited software support, and resource constraints, led to replication times ranging from 0.8 to 3.3 hours and an 80 percent success rate. While Intel TDX provides strong security benefits, broader adoption will require improved automation and more standardized replication packages
Performance, cost, and effectiveness of verifying academic paper replication packages via Intel TDX
No full textReplication packages are a collection of files and data that allows other researchers to reproduce the results reported in an academic paper. It is a core feature that establishes integrity within the modern world of research, prominently in economics. As such, recreating these packages requires a secure, confidential, and regulated environment that can efficiently handle varying sizes of data. We explore the efficiency and cost effectiveness of recreating academic replication packages within a TDX (Trust Domain Extension), which are isolated virtual machines that protect the confidentiality of its contents. Of the 27 replication packages recreated within a TDX, 19 were successful. The average cost of successful replications was $1.30 with an average run time of 2 hours and 10 minutes, with most cases producing accurate results. Running replication packages within TDX provides authenticity and security benefits with the issue of certain cases failing due to software limitations. However, given the current cost, runtime, and rate of success, it can be assumed that with further development in software, running replication packages within TDX’s can be plausible.
(Replication packages were recreated in TDX by a group
Assessing Climate Impacts on Forage Availability in the Somali Region of Ethiopia Using Remote Sensing
No full textForage conditions in the Somali Region of Ethiopia are a critical concern due to the area’s dependence on livestock and pastoralism. This region contributes significantly to Ethiopia’s total livestock population, with pastoralism making up an estimated 42% of the national livestock economy (Tenaw et al.). However, forage availability is increasingly diminishing, posing risks to food security and livelihoods. Over 80% of Ethiopia’s land is estimated to be moderately to severely degraded (Solomon et al.), with the Somali Region being especially vulnerable. This study investigates the impact of changes over time in land surface temperature (LST) and precipitation on the Normalized Difference Vegetation Index (NDVI), which reflects the general condition of forage in the area.. Temporal trends of NDVI, LST, and precipitation in selected grassland areas of the Somali Region were derived using satellite remote sensing data. A multiple regression analysis was conducted to quantify the significance of the relationships between these climate variables and the NDVI. The resulting findings provide vital understanding of the key elements influencing forage degradation in the Somali Region, aiding in more informed pastoral planning and land management strategies for this vulnerable area.
Tenaw, et.al. Assessment of Place of Delivery and Associated Factors among Pastoralists in Ethiopia: A Systematic Review and Meta-Analysis Evaluation. J Pregnancy. 2023 Nov 9;2023:2634610. doi: 10.1155/2023/2634610
Solomon, et.al. Revitalizing Ethiopia’s highland soil degradation: a comprehensive review on land degradation and effective management interventions. Discov Sustain 5, 106 (2024). https://doi.org/10.1007/s43621-024-00282-
Optimizing UAV Performance In Adverse Flying Conditions Using Reinforcement Learning
No full textIn recent years, reinforcement learning (RL) has emerged as a promising method in making UAVs fully autonomous, in which agents are put in simulated environments and improve their performance through trial and error. However, RL-trained UAVs are prone to issue when adapting to noncurated environments different from their training conditions. These limitations can undermine reliability and safety in critical scenarios, especially in real-world scenarios where conditions can change. Using the Pyflyt library to simulate the environment, sensor noise was introduced to emulate unfamiliar and harsh environments and observe the change of performance of the UAV. Flight trajectories, stability, and navigation success rates were analyzed across varying noise conditions to assess the model’s adaptability and response to novel conditions and train an RL model with these variables in mind. Preliminary results are indicative of restriction of contemporary models. At a baseline noise level of 0.1, the simulation yielded a mean reward of 6.497 with a standard deviation (σ) of 106.838; at 0.3, the mean dropped to -64.552 (σ = 119.743); and at 0.5, it fell further to -154.132 (σ = 137.479). The increasingly negative rewards and rising standard deviations indicate poor model guidance and growing instability in UAV flight paths. The simulation data stipulates the importance of testing learning algorithms in imperfect conditions and exhibits the faults in many current RL models when applied to unpredictable UAV settings. The experiment builds on the framework for developing RL models that maintain UAV simulation performance in more realistic scenarios, such as disaster relief and surveillance, by conditioning such projects in more disruptive, noise-intensive environments
A Conversational AI System for Daily Support and Risk Detection in Workplace Stress
No full textThis paper presents the design and implementation of a conversational AI-based chatbot application for the early detection and support of workplace-related psychological stress. The system leverages natural language processing (NLP) through Google’s Gemini large language models, integrated into a custom web-based interface developed using Python, JavaScript, and HTML. The user-centered frontend emulates a familiar coffee shop environment to reduce perceived stigma and foster user engagement in a relaxed digital space.
To assess mental health risk, the application incorporates surveys derived from contemporary mental health datasets. These responses are processed through a decision-support pipeline that utilizes machine learning models to classify users into one of three risk categories: low, medium, or high. Risk levels are masked using themed terminology (e.g., coffee-based metaphors) to maintain an approachable user experience. Based on the determined risk level, users are directed to a corresponding AI chatbot experience tailored to their emotional needs. The user is recommended to use this application daily in which they input their situation and daily mood and are given personalized exercises, words of affirmation, and comforting comments, tailored to their current mood in order to boost morale and decrease stress and anxiety.
Preliminary testing indicates high user retention rates and satisfactory classification accuracy across test cases. The proposed system demonstrates the feasibility and value of empathetic, AI-driven conversational interfaces in non-clinical mental health support settings. Future development will focus on multilingual support, integration with corporate wellness platforms, and increased personalization through embedded deployment and user-specific customization features
Modeling and Analysis of the Impacts of the Rise in AI Use on Data Center Growth, Resulting Workforce Displacement, and Environmental Impacts Through a Coupled System of Ordinary Differential Equations Within a Feedback Framework
No full textGenerative AI has widespread impacts on data center growth, environmental emissions, resource consumption, and the workforce. However, the complex, interconnected nature of these systems—and their tendency to evolve through dynamic feedback loops—makes them challenging to accurately model with static or linear methods. To address this, we model these dynamics through a coupled system of ordinary differential equations (ODEs) within a continuous-time feedback framework. Workforce dynamics are both incorporated into the ODE system and modeled explicitly through a compartmental framework—Susceptible (At Risk), Infected (Unemployed), and Recovered (Reskilled). We solve the system numerically to simulate the evolving impacts AI adoption has on variables including data center growth, energy use, carbon emissions, water consumption, and labor transitions. We then examine various scenarios by running simulations that compare increases in the AI adoption rate (r) and the intensity of policy pushback (psi_c) to the peak variable output. Over a simulated 365-day period, AI adoption stabilizes; however, the slight increases in adoption significantly elevate energy use, water consumption, and carbon emissions. We also find that unemployment rises substantially regardless of reskilling efforts, as proportionately, fewer workers reskill successfully. Policy pushback is effective at quickly decreasing the timespan released emissions, even at less intense levels. However, with pushback, emissions are elevated in this short timeframe, exacerbating unemployment. This project aims to inform future AI mitigation policies, particularly when discussing sustainable measures of regulating AI adoption so as to not further exacerbate environmental and workforce concerns. Our work supports UN Sustainable Development Goals #13 (Climate Action) and #8 (Decent Work and Economic Growth). This research also offers a foundation for future data-driven modeling from Physics-Informed Neural Networks (PINNS)
Modeling Contrail Formation Via an SIR-Framework
No full textCondensation trails, or contrails, are streaks of condensed water released behind jet aircraft at high altitudes that can extend to become artificial cirrus clouds. They have become increasingly prevalent in the discussion of aviation climate impact due to their ability to trap heat within the atmosphere, accounting for 35% of all aviation emissions. However, the behavior of contrails during and after production is still poorly understood. We introduce a system of one-dimensional compartmental models motivated by the Susceptible-Infected-Recovered (SIR) framework to model and simulate contrail formation and dissipation. The susceptible compartment consists of air parcels where contrail formation is likely due to a cold and humid atmosphere. Secondly, air parcels become “infected” into contrails once hot aircraft exhaust passes through them. Finally, after a period of time the area “recovers” into the final compartment. Modeling begins with a baseline advection-diffusion equation including a constant wind velocity term (u) that accounts for contrail movement post-production. Using the method of manufactured solutions (MMS), the partial differential equation is solved for each compartment to generate a solution for each parameter β (rate of contrail formation) and ɣ (rate of contrail dissipation). Using a forward time-centered space (FTCS) algorithm and adhering to stability conditions, preliminary results show that predicting contrail rates of formation and dissipation is accurate within 1% using controlled timesteps on both the time and space axes. Future research aims to analyze models with a fluctuating wind velocity and including multi-dimensional frameworks. Doing so will continue supporting United Nations Sustainable Development Goal #13: Climate Action