Mason Journals (George Mason Univ.)
Not a member yet
3256 research outputs found
Sort by
Creating an Interface for Fluent-Based Task Planning for Robotic Object Search in Household Environments
No full textRobot navigation and object search within household environments are foundational tasks in the robotics community. Current research focuses on improving the planning algorithms that robots use to complete such tasks. However, there is a need in the RAIL Group for an interface between other disparate technologies: the planner, which expects an abstract representation of the state, and the environment itself—either a simulation of a household environment or a physical robot operating in the lab. Therefore, the goal is to establish a connection between the planner and the robot by deriving the up-to-date symbolic representation of the environment as the robot explores. Such a connection allows the planner to be used in closed-loop deployments, affording use of the RAIL Group’s novel approach to learning-informed decision making. Abstract planners require that the state be described via a union of "fluents," each a predicate function representing some aspect of what's true about the world. Our interface determines which fluents are active in a given state via the robot’s partial map of the environment. Based on the active fluents, the planner will determine which action it should take in the given state, before returning that action to the interface. Upon receiving the planner's chosen action, the interface executes it in the environment, then proceeds to recalculate the active fluents. This interface will make it possible to use this planning tool, critical to the RAIL Group’s aims for effective task planning under uncertainty
Analyzing the Effects of Wildfires on Air Quality Across the United States
No full textWildfires have become increasingly frequent and severe across the United States, releasing substantial amounts of fine particulate matter (PM2.5), a pollutant known to pose serious risks to human health and the environment. Elevated PM2.5 levels during wildfire events can affect air quality far beyond the fire’s origin, with implications for public health, environmental monitoring, and policy. This study investigates the impact of wildfires on air quality by integrating advanced visualization tools with spatial and spatiotemporal statistical modeling. Using nationwide wildfire and air quality datasets, we examine pollutant dispersion patterns, identify the most affected regions, and assess regional vulnerabilities to wildfire smoke. Our methodology captures both spatial heat signatures and temporal dynamics of PM2.5 concentrations in relation to wildfire activity. The results provide critical insights into the geographic and temporal variability of wildfire-driven air pollution. By quantifying these effects, our findings support improved air quality forecasting, inform targeted public health responses, and contribute to the development of more effective environmental and emergency preparedness policies
Identifying Consumer Behavioral Patterns in Livestreaming E-Commerce
No full textProbabilistic selling, where consumers purchase a product without knowing its exact identity, has traditionally served as a strategy for price discrimination and inventory clearance. One related growing trend is the global sale of blind boxes, where companies such as Pop Mart expand on this strategy through unique designs and emotional engagement, oftentimes through livestreams. As a result, many shoppers now turn to livestream shopping, where influencers unbox blind boxes in real time for their audiences. This study examines the effects of livestream unboxing on viewer spontaneous purchasing decisions using a direct analysis of livestream data. Preliminary testing included comparing seven different audio transcription models for potential use, with OpenAI’s Whisper Large V2 Model achieving the highest accuracy. Additionally, custom words that the model failed to initially identify, such as common names of blind boxes, were added to initial prompting to increase the accuracy of the model. Livestream videos from a leading Chinese livestreaming platform were then converted to audio transcriptions, and phrase and emotion pairs were subsequently identified through a combination of human tagging and Natural Language Processing (NLP) models. Analysis on these phrases, in combination with sales and engagement data, suggest that phrases identified as urgent or persuasive correlate to spikes in viewer purchases. Overall, these results support the work of previous studies that identified a positive correlation between influencer and buyer consumption choices
PyMOCAT-MC: A Python Implementation of the MIT Orbital Capacity Assessment Toolbox Monte Carlo Module
No full textThis research presents a comprehensive conversion of the MIT Orbital Capacity Assessment Toolbox - Monte Carlo (MOCAT-MC) from MATLAB to Python, maintaining full functional compatibility while leveraging modern Python scientific computing ecosystems. The conversion encompasses over 150 core algorithms and supporting functions, including orbital propagation, collision probability calculations, fragmentation modeling, and atmospheric density modeling. The Python implementation preserves the original MATLAB architecture while introducing vectorized operations, improved memory management, and enhanced modularity through object-oriented design. Key conversion challenges addressed include: (1) Indexing system transformation from MATLAB's 1-based to Python's 0-based indexing across all matrix operations, (2) Complex orbital mechanics algorithms requiring precise numerical accuracy preservation, (3) Large-scale Monte Carlo simulations with memory-efficient vectorization, and (4) Integration with Python scientific computing libraries (NumPy, SciPy, Astropy) while maintaining computational performance. The converted codebase includes complete example scenarios with output validation within 0.9% of the original MATLAB implementation. Performance benchmarks demonstrate faster execution times while providing enhanced accessibility through open-source Python dependencies. The conversion enables broader adoption in the space situational awareness community, supporting orbital capacity assessment, agent-based modeling for satellites, and megaconstellation impact analysis through a more accessible and extensible platform. 
Comparing the Social Networks of Students Enrolled in Inclusive Postsecondary Education Programs and their Peers Enrolled in Traditional College Programs
No full textWe examined differences in the social networks, social supports, and college-related anxiety and distress of 42 college students in a traditional degree-seeking (TDS) program and an inclusive postsecondary education (IPSE) program on the same campus. We found that students in IPSE programs had smaller, denser social networks. However, anxiety and levels of support differed across social, academic, and daily living domains. These findings provide a deeper understanding of the structure and function of the social networks of students with and without disabilities as they begin their college journey, and have important implications for K-12 transition programming and IPSE programs.
 
Evaluating LLMs as SQL Tutors: A Comparative Study of Adaptive, Feedback-Driven Learning using Commercial and Research-Phase Models
No full textLarge Language Models (LLMs) are being increasingly used in education to help people learn various topics, including SQL. While most research focuses on how well AI models can generate correct SQL queries, there has been less attention on how useful they are as tutors, especially when it comes to helping learners fix mistakes through feedback. In this study, several commercial LLMs (GPT-4, Claude, Grok, Llama, and Gemini) and some experimental models (FISQL, CoSQL, DB-GPT, and S3SQL) were evaluated by grading their feedback based on a rubric.
A set of 10 SQL learning tasks across five key topics: selection, aggregation, joins, set operations, and subqueries, was created. For each task, the models were given an incorrect query, and their responses were graded by one researcher using a rubric of 5 points with four weighted areas: accuracy (15%), clarity (20%), guidance (25%), and adaptiveness (40%). Commercial models were generally accurate and fluent, but their ability to guide learners through mistakes varied. Among them, Grok (4.675) stood out as the best-performing commercial model. Conversely, research models, like S3SQL and FISQL, gave strong, structured feedback, especially when they were aware of the database schema, but weren’t built for full back-and-forth conversations. Among those, S3SQL gave the strongest (4.21) and most structured feedback based on published findings.
Overall, our findings show that while LLMs can support SQL education, they vary in terms of adaptability and giving feedback. This study designs a framework for evaluating adaptability and tutoring functionality in both commercial and experimental LLMs, helping future developments in SQL education. Next steps include proposing a new model that combines the features of the best-performing models to further enhance SQL tutoring
Robust Sleep Stage Classification from Multimodal Physiological Signals Using a Cross-Modal Transformer with Missing Data Resilience
No full textAccurate and robust classification of sleep stages, in particular classifying Wake, N1, N2, N3, and REM stages, is of uttermost importance to diagnose sleep disorders and promote overall health. Traditional laboratory-based polysomnography (PSG) is resource-intensive, while existing at-home monitoring solutions often struggle with noisy or missing data. This research proposes a deep learning framework featuring a Cross-Modal Transformer for sleep stage classification that fuses multimodal physiological data (EEG, EOG, EMG) while maintaining performance in the presence of incomplete sensor streams. Unlike simple attention or gated fusion which proved ineffective, the Transformer architecture forces the model to learn a joint, intertwined representation by treating the modalities as a single sequence of feature "tokens." The methodology involves extracting comprehensive features from the Sleep-EDF dataset, followed by subject-wise train-test splitting and aggressive data augmentation that simulates the random loss of any modality. A baseline attention model demonstrated a significant failure to adapt, with Cohen's Kappa score dropping from 0.74 to 0.64-0.66 when a modality was removed. In contrast, the Cross-Modal Transformer achieved a Kappa of 0.75 and accuracy of 81.3 with all modalities present and demonstrated graceful degradation, with scores of 0.74 when missing EMG, 0.73 when missing EOG, and a strong 0.72 when relying on EEG only. Future work will explore real-time implementation, investigate the framework's adaptability to other physiological signals, and evaluate its performance on more diverse and larger datasets from real-world, at-home monitoring scenarios
Investigating Protein Expression of AKT/mTOR Pathway in HR+/HER2- Breast Cancer Tumors using Laser Capture Microdissection and Reverse Phase Protein Arrays
No full textTumors classified as HR+ and HER2- are most likely to develop resistance to endocrine therapy through over-activation of the AKT/mTOR signaling pathway. Immunohistochemistry staining (IHC) is typically used to identify and quantify protein expression; however, similar to using whole tissue lysates in protein analysis, loss of sensitivity is a critical issue. This study used laser capture microdissection (LCM) coupled with reverse phase protein arrays (RPPA) to isolate tumor cells from twelve biopsies from high-risk breast cancer patients and investigate the expression of AKT/mTOR related proteins compared to whole tissue (WT) lysates from the same biopsies. Using a t-test, the protein expression of both mTOR S2448 (p= 0.042) and Estrogen Receptor alpha (ERa) (p= 0.005) were found to be significantly higher in the microdissected tumor population compared to WT lysates. The mean intensity value of mTOR S2448 expression in WT lysates was only of the microdissected tumor cells and in the same trend the mean intensity value of ERa was only 9% of the microdissected. These results indicate that LCM together with RPPA is a far more sensitive option for the detection and quantification of total and phospho-proteins providing a more accurate read out of their activation state
Optimizing Sequence-to-Graph Alignment
No full textGraph‑based reference genomes preserve population‑level variation that linear references cannot, yet the combinatorial expansion of candidate paths grows exponentially (2k for k variant nodes) with each additional variant node making sequence‑to‑graph alignment computationally demanding. We evaluate four alignment strategies, a greedy heuristic, dynamic programming (DP), exhaustive search, and exhaustive search with early stopping, on synthetic directed acyclic graphs (DAGs) containing 10-20 nodes with a 10% branching probability and simulated short reads. The greedy heuristic completes in <1 ms per read but attains only 43 % alignment accuracy. Early‑stop exhaustive search guarantees 100 % accuracy but requires 2.3s per read, underscoring the cost of exhaustive exploration and the tradeoff between runtime and completeness. DP achieves a balanced profile, delivering 93% accuracy with a runtime two orders of magnitude faster than exhaustive search. These results map the speed‑accuracy frontier in sequence‑to‑graph alignment and motivate banded DP and heuristic pruning to scale analyses to kilobase‑scale genome graphs