Mason Journals (George Mason Univ.)
Not a member yet
3256 research outputs found
Sort by
Monocular Visual SLAM Using YOLO-Based Object Detection and Depth Estimation
No full textUnmanned aerial vehicles (UAVs), such as blimps and other lighter-than-air systems, have helpful indoor applications such as monitoring, navigation, and inspection. These vehicles offer advantages like long flight duration and low power consumption but face major challenges in indoor environments where GPS signals are unavailable. Additionally, their limited payload capacity restricts the use of localization systems that rely on heavy sensors, such as LIDAR or stereo cameras. To address these limitations, this project explores a monocular vision-based SLAM method that uses YOLO (You Only Look Once) object detection and monocular depth estimation. A trained YOLO model can identify recognizable objects in real time, while the depth estimation model predicts the distance of each pixel using learned patterns from large datasets. This combination allows the system to estimate the blimp’s position and construct a map that includes both spatial geometry and labeled objects. This system is designed to support navigation and mapping for lightweight indoor UAVs while minimizing computational load and sensor weight. Future testing in simulations and indoor environments will evaluate the accuracy and speed of this combined method for deployment on small aerial platforms
Evaluating the Effectiveness of Persona Simulation in Opinion Prediction with GPT-4.1
No full textPersona simulation involves utilizing large language models to anticipate human choices or interactions based on specific characteristic information. These can be used as additional data points in surveys, addressing the challenges that come with data collection, such as imbalanced sampling or non-response bias. Thus, to further understand current limitations and future directions, we tested persona simulation in opinion prediction using GPT-4.1. Personas with demographic and personality data were taken from Columbia University’s Personas dataset to be used in election forecasting. Using personas from nine U.S. states, GPT-4.1 accurately predicted 2024 election outcomes in eight out of the nine states, only failing in one of the swing states. Yet additional analysis of the American National Election Studies dataset revealed accuracies of 0.648 and 0.610 using logistic regression and GPT-4.1, respectively, indicating great room for voting prediction improvement. We then utilized the American Trends Panel Wave 123 dataset from Pew Research Center, which focused on opinions related to medicine and technologies. GPT-4.1 was able to anticipate beliefs about childhood vaccines with an accuracy of up to 0.94. Furthermore, we applied GPT-4.1 to generate conversations among personas and observed that the simulated dialogues and opinions adhered well to personas' personalities and backgrounds, albeit lacking natural human-like flow. Persona simulation proves to be a promising application of artificial intelligence as long as biases are addressed. In the near future, it will be beneficial to apply it to opinion analysis and reaction prediction in diverse fields ranging from public health to lawmaking to economics
Spatial and Statistical Analysis of Crash Frequency and Severity: An Arlington, VA Case Study
No full textUnderstanding crash data is crucial to identifying patterns that can guide the development of safety infrastructure and countermeasures. This project conducts data-driven analyses, like geoanalysis and frequency analysis, to help understand crash risks, especially in urban areas such as Arlington. However, crashes are complex and influenced by multiple factors, therefore it is crucial to explore the nuances that could predict the relationship between crash severity and independent factors based on crash data provided by the Virginia Department of Transportation covering crashes from 2017 to today. By using tools such as ArcGIS Pro and StataBE 18, we aim to test the correlation between various factors to explain the frequency and severity of crashes. Spatial tools in ArcGIS, such as Near, Join Field, and Summary Statistics, allow us to explore the geolocation of different data layers and identify any potential connections. Comparing multiple fatal crashes in Arlington (between 2017-2025), the AADWT (average annual daily weekday traffic) varied for each, ranging from as many as 47,000 vehicles per day to as low as 740 vehicles per day. Roads with the highest AADWT (around 100,000) often had relatively few crashes, while roads with lower AADWT (below 10,000) had the greatest amount of crash frequencies, peaking at 268 crashes on a roadway with an AADWT of 9,300. These outliers highlight the need to explore infrastructure and behavioral reasons as the relationship between traffic volume and crash occurrence is not proportional but rather random. The findings from this study may help target risk areas that are in need for safety improvements and inform the need for safety features and policies
Assessing the Reproducibility and Computational Costs of Management Science Research Using Intel TDX Secure Environments
No full textAs academic journals increasingly require authors to provide data and code for result verification, concerns around secure and reproducible research execution are growing. Replication packages often contain undocumented or unsafe code, which poses potential threats to system integrity, data privacy, and proprietary algorithms when run in standard computing environments. This study builds upon recent efforts to evaluate secure, hardware-based replication using Intel’s Trust Domain Extensions (TDX), a confidential computing framework that enables secure execution within isolated virtual machines (VMs). Using a curated set of recent Management Science papers, we tested the reproducibility of empirical results from multiple fields including finance, accounting, and economics. Each paper’s code was deployed in a TDX-enabled environment using Google Cloud and Microsoft Azure to evaluate the replication feasibility, security advantages, and resource requirements. Key metrics included runtime and credit-based cost. Our replication benchmarks showed an average usage of approximately 1.45 credits and 2.1 hours per task. Longer runtimes were typically associated with undocumented dependencies, closed-source toolkits, or inaccessible datasets. Our results support the growing applicability of Intel TDX as a cost-effective and secure approach to verifying academic research, especially for well-documented studies in empirical finance and economics.  
Goodbye Vietnam, Good Morning Indochina: Teaching the Indochina Wars in American World History Classrooms
Full text linkEvaluating the Tradeoff Between Predictive Accuracy and Racial Fairness in Machine Learning-Based Recidivism Models
No full textWith the increasing usage of predictive models for criminal risk assessment and predicting recidivism, concerns have grown due to the "black box" nature of these models and their implications with fairness and racial bias. Although these classifiers have shown the potential for high accuracy in terms of predicting criminal risk of recidivism, they may use sensitive factors such as race that can lead to racial unfairness in predictive outcomes. This research explores the impact of race being included or excluded as a feature for machine learning classifiers in predicting recidivism, with a goal of addressing the difficult task of minimizing bias while maximizing predictive accuracy.
Two binary Random Forest classifiers were trained on the COMPAS dataset: one including race as a feature and one excluding it. Both models used demographic and criminal history features and were evaluated using accuracy, precision, recall, F1-score, and flip rate, a measure of counterfactual fairness measuring the percentage of predictions that change when race is altered, where a greater flip rate means larger sensitivity to race changes and lower fairness.
The classifier including race achieved an accuracy of 77% with both the precision and recall being balanced at 77%, but had a flip rate of 36.83%, suggesting high sensitivity to racial variation. Removing race in the second classifier decreased accuracy to 74%, with precision and recall dropping to 74% and 73% respectively. These findings demonstrate the complexity of fairness and race in recidivism prediction, as excluding race reduces direct racial influence but reduces overall predictive performance. Future work will seek to develop a judicial assistant that can provide Chain-of-Thought reasoning to how it arrived at each prediction, enhancing transparency, and Reinforcement Learning with Human Feedback (RLHF) will be used to improve the model’s alignment with ethical judicial decision-making
The Impact of Dust Concentration on the West African Monsoon (WAM) and Regional Climate Patterns
No full textThe West African Monsoon (WAM) is responsible for most of the annual rainfall to the Sahel and Gulf of Guinea, thus affecting the food and economic security of millions. Yet future climate projections remain uncertain, partly because global climate models incompletely represent non-CO2 forcings (e.g. mineral dust) that modulate radiation, cloud patterns, and the land-sea thermal contrast that drives the monsoon. Here we investigate the impact of atmospheric dust concentration on the West African Monsoon using a low-dust scenario derived from a climate simulation of the Miocene epoch (23.03 to 5.333 million years ago). Changes in vegetation between the Miocene and preindustrial simulations led to less dust. Thus, to isolate the role of dust in dust-monsoon coupling, we perform two fixed sea surface temperature (fSST) experiments which differ only with respect to their atmospheric dust concentrations; a pre‑industrial control experiment (modern Saharan dust) and a low‑dust experiment (Miocene conditions). Monthly diagnostics for the core monsoon season (June–September) reveal a seasonally evolving response. We find that dust reduction causes a decrease in precipitation compared to the preindustrial run over the Sahel in June and July, thereby limiting the northward expansion of the monsoon. However, we note this pattern changes in August and September, when Eastern regions of the Sahel are marked by precipitation increase, and thus higher monsoon intensity.
 
Detecting Protein Pathways Affecting Nerve Damage in Rat Spinal Cords with Breast Cancer and Paclitaxel Treatment via RPPA
No full textPaclitaxel (PAC) is a common first-line chemotherapy drug used for breast cancer treatment. However, when treated, severe side effects of chronic pain in the nerves outside the brain and spinal cord, or neuropathic pain, occur. Although many studies explore neuropathic pain and nerve damage caused by paclitaxel treatment, they typically focus on nerve cells and pain relief methods. We examined proteins that may contribute to nerve damage in mouse spinal cords under different conditions of breast cancer and paclitaxel treatment. To identify the different protein pathways present in PAC-treated cancer vs. untreated cancer, vehicle control with cancer vs. control without cancer, and spinal cord with both or neither conditions, we used mass spectrometry and created volcano plots. Using Reverse Phase Protein Array (RPPA), we quantified the number of specific proteins present and identified antibodies for further study. Using immunohistochemistry, we used antibody H2AX to locate this protein in neuroma tissue as a surrogate nerve tissue. RPPA results showed that breast cancer increased histone 4 and IRS-1 Ser612 compared to no cancer. Paclitaxel treatment with cancer increased H2AX and decreased autophagy-related proteins compared to untreated cancer. Currently, we are investigating specific pathways that are altered by the presence of breast cancer and the effects of paclitaxel. We found breast cancer cells can alter spinal cord proteins, which are essential for normal nerve signaling. Additionally, paclitaxel disrupts protein signaling pathways involved in nerve functions, leading to inflammation and neuropathic pain
Rapid Quantum Computation of Ionizable Lipid Acidity Constants
No full textThe COVID-19 vaccine showcased the power of lipid nanoparticles (LNPs) as a novel system for delivering therapies targeting infectious diseases and potentially other illnesses like cancer. Despite the fantastic versatility of LNPs, they require specific changes in physical properties at pHs of around 6.5 to 7.0. Ionizable lipids comprise the only pH sensitive component of the LNP. Matching the pKa of the ionizable lipid to the desired pH range should yield a change in physical properties in that range. However, no current high through-put experimental or computational approach for determining the ionizable lipid’s pKa features practical timeframes (days to months). Using faster but more approximate quantum methods than the literature standard, we found a slightly wider error margin for the predicted pKas but with significantly faster completion times, indicating a viable alternative to the literature standard. Much like the literature standard, our method does not distinguish between effective and ineffective ionizable lipids. Because of the speed increase, this method offers the opportunity to investigate potential beneficial improvements such as including the effects of lipid conformation on predicted pKa