Mason Journals (George Mason Univ.)
Not a member yet
3256 research outputs found
Sort by
Looming Detection in Athletic Motion Using AI-Driven Visual Stimulus Modeling
No full textVisual stimuli indicating rapid motion are critical to perceptual systems for initiating defensive or anticipatory responses. In high-speed sports such as tennis, detecting quickly moving objects across multiple viewpoints can provide valuable insight into attention, motor preparation, and reaction timing. This study presents a brain-inspired system for identifying motion events within video footage of tennis players, using motion dynamics to model visual patterns. Videos are processed through a custom pipeline that includes foreground segmentation, motion trajectory extraction, and event-based encoding. Detected stimuli are further analyzed for motion peaks and direction changes to infer perceptual salience. The system also supports detection of the tennis ball from multiple perspectives in real-world scenarios. Additionally, a stimulus generator is included to simulate expanding objects with structured backgrounds, enabling the creation of training data that reflects naturalistic conditions. Frames are transformed into dynamic vision sensor (DVS) representations, capturing positive and negative motion events for frame-by-frame analysis. The model aims to support real-time perceptual systems in both neuroscience research and applied human-computer interaction, offering a framework for tracking and interpreting complex motion cues in dynamic environments
Computationally Driven De Novo Design and Engineering of a β-D-Glucose Binding Protein
No full textDe novo protein design, the computational development of proteins from the ground up, bypasses the evolutionary complexity of biological systems, simplifying studies on cellular pathways and proposing various applications. This study presents the development of a novel protein that binds to β-D-Glucose, a cyclic sugar in plant and animal tissues integral to glycolysis and metabolic pathways. Expanding upon a structurally stable beta-barrel scaffold, an iterative protocol was utilized to construct a robust binding pocket. Sidechain residues were modified through the ChimeraX protein editing software, which were then fed into AlphaFold3/Chai Discovery, an application that assesses the folding and viability of the protein, and LigandMPNN, a deep learning application that generates optimized protein sequences given a protein-ligand input. VMD (Visual Molecular Dynamics), which models the protein-ligand complex in solution over time, was then utilized for validation and further iteration. The final protein exhibited modest protein-ligand confidence in Chai Discovery with an ipTM (interface predicted template modeling score) of 0.74. Regarding molecular dynamic simulations, the protein RMSD (root mean square deviation) converged to 3.75A, implying overall protein instability, and determined 13 h-bonds, with at least 4-6 stable bonds with an occupancy of at least 16.47%. These data indicate that the protein-ligand complex was relatively stable at the expense of a compromised backbone structure. Further validation encompasses fluorescence titration and SPR (Surface Plasmon Resonance), an optical method to quantify protein-ligand kinetics and affinities. Ultimately, optimization of this novel β-D-Glucose-binding protein proposes numerous applications in biotechnology and studies on glucose metabolism and protein-ligand behavior
Quantitative Comparison of 3D Microglial Morphology in Control and 5xFAD Alzheimer’s Model Mice
No full textMicroglia, the brain’s resident immune cells, play a critical role in Alzheimer’s disease (AD) progression through phagocytosis of amyloid-beta (Aβ) proteins. In the hippocampus, a region heavily impacted by AD, microglia undergo morphological changes as they attempt to engulf Aβ plaques, contributing to chronic neuroinflammation. Current literature lacks a quantitative understanding of how microglial morphology differs in AD. This study addresses this gap by comparing microglia in control and 5xFAD Alzheimer’s model mice.
We utilize data from NeuroMorpho.Org, an open-source repository of digitally reconstructed neurons. Using the Python NeuroM package, a library for extracting morphometric features from SWC files, we quantified hippocampal microglia and conducted statistical and computational analyses to identify distinguishing features between the groups. Principal Component Analysis revealed that soma and structural complexity metrics contributed most to group variance. Welch’s two-sample t-test confirmed all features differed significantly (p < 0.0001). Supervised machine learning models, specifically logistic regression, random forest, and gradient boosting, identified soma-based metrics, average segment length, and average section length as top predictors of group classification. These results support the hypothesis that AD alters microglial morphology in measurable ways.
Preliminary unsupervised machine learning with K-means clustering is ongoing to explore natural data groupings. Future directions include incorporating age-based categorization and applying SHapley Additive exPlanations (SHAP) to further interpret machine learning model feature importance. By identifying distinguishing features in disease and healthy states, this research contributes to early diagnosis and understanding of AD pathology, supporting the United Nations Sustainable Development Goal #3: Good Health and Well-being
Epidemiological-Inspired Model of Post-Operational Chronic Pain in Scoliosis Patients
No full textChronic post-surgical pain (CPSP) is a common and often overlooked complication following surgical correction of idiopathic scoliosis, impacting long-term patient wellbeing despite improvements in surgical outcomes. This project introduces a novel epidemiological framework to model the progression of CPSP using a Susceptible-Infected-Treatment-Recovered-Chronic Pain (SITR-C) compartmental structure. By applying a coupled system of nonlinear differential equations, we simulate pain trajectories over time and assess the effectiveness of surgical interventions. The model is implemented for a single-cohort population and extended to a two-cohort design to compare outcomes between Posterior Spinal Fusion and Vertebral Body Tethering procedures. Further stratification by patient age enables us to explore demographic influences on chronic pain dynamics. Stability analysis confirms the existence of disease-free equilibria, while numerical simulations provide insights into long-term recovery versus chronic outcomes. This work highlights the importance of integrating mathematical modeling into clinical decision-making and pain management strategies. In alignment with the United Nations Sustainable Development Goal 3: Good Health and Wellbeing, the SITR-C model offers a data-driven approach to improve patient care, predict recovery paths, and reduce the long-term burden of CPSP
Robust Statistical Learning for Neuroimaging: Mitigating Artifacts in Conditional Generative AI Data Augmentation
No full textGenerative AI–based data augmentation can address sample size and privacy limitations in neuroimaging, but models like Conditional Variational Autoencoders (CVAEs), often produce artifacts (e.g., checkerboards). These artifacts act as outliers, violating Gaussian-error assumptions of least-squares regression and biasing statistical inference. To mitigate this bias, we developed a simulation framework to evaluate different loss functions for distributed surface-based regression. The framework generates noisy synthetic neuroactivity maps using a CVAE, combines them with original data at varying ratios, and performs distributed image-on-scalar regression (DISR) with bivariate penalized splines over triangulation (BPST), comparing least-squares and Huber loss functions. Results, measured by Mean Integrated Squared Error (MISE) across Monte Carlo replicates, show that synthetic artifacts inflate estimation errors in least‑squares regression. Conversely, Huber regression consistently outperformed least‑squares on both synthetic and mixed datasets, retaining accuracy with minimal efficiency loss despite artifact contamination. These findings establish Huber regression as a robust solution to artifact-induced outliers and underscores the need for robust statistical learning when using generative models in neuroimaging. The proposed method provides a scalable framework for improving statistical power in data-scarce biomedical imaging applications
Bridging Theory and Practice: Developing An Interactive Web Interface for MCTS Algorithm Exploration Implementing OCBA Selection Operator in Othello
No full textMonte Carlo Tree Search (MCTS) algorithms have become integral to artificial intelligence in complex and sequential decision-making scenarios such as Go and Othello by applying multi-armed bandit selection strategies for tree exploration. MCTS selection operators are essential to determining tree expansion in large decision and state spaces. Traditional implementations utilize Upper Confidence bounds applied to Trees (UCT) as the node selection operator, while recent advances suggest Optimal Computing Budget Allocation (OCBA) selection operators provide superior performance by dynamically allocating computational resources based on statistical variance estimates, balancing exploration alongside exploitation over UCT's tendency toward the latter. Despite theoretical advances in MCTS selection strategies, accessible platforms for experiencing these AI systems remain limited. While comparative studies between UCT and OCBA exist in research literature, user-friendly interfaces for MCTS-based Othello gameplay are scarce. This project addresses this gap by developing a comprehensive web-based interface for an Othello AI system implementing MCTS with OCBA selection operators. Built using HTML, CSS, and JavaScript, the responsive interface allows human players to compete against AI with fully configurable parameters. Users can adjust rollout counts (250-5000), select simulation strategies (random/semi-random), choose opponent types, and determine move ordering preferences. The system features real-time game state visualization, score tracking, and intuitive parameter configuration. This work provides a valuable educational and research tool bridging the gap between theoretical AI advances and practical user interaction with sophisticated game-playing algorithms
Imputing Missing Chlorophyll-a Data in the Chesapeake Bay Region Using Random Forest and KNN Machine Learning Models
No full textHarmful algal blooms (HAB) have become an increasingly pressing environmental issue in the Chesapeake Bay Region, posing a threat to surrounding ecosystems and human health through releasing toxins or sheer biomass. As the nation’s largest estuary, Chesapeake Bay is not only a center of agricultural activity such as shellfish farming but also supports a wide range of diverse habitats. Consequently, the prediction of algal blooms is essential to protecting water quality. Satellite data collection of chlorophyll-a (Chl-a) values has allowed for accessible and consistent collection of data to train prediction models; however, data is often obscured by cloud cover and other environmental factors. A lack of complete and continuous data has compromised the accuracy of predictions and thus it is essential to use machine learning for the imputation of data gaps. We used a full-coverage dataset from the year 2025 to compare the performance of Random Forest (RF) and K-Nearest Neighbors (KNN) models for imputing Chl-a values in the Chesapeake Bay Region. The best model was found to be RF with a coefficient of determination (R2) of 0.916, a notable improvement over KNN which yielded an R2 value of 0.588 when 30.46% of data was missing. The accuracy of the RF models was enhanced through iterative imputation to improve the accuracy of RF from 0.85 initially, to 0.916. Not only was the RF model more accurate, but it also exhibited a significantly lower runtime that proved essential when processing large amounts of satellite data. This study provides valuable insight into methods of adjusting RF models to better improve the quality of Chl-a data to advise water quality management
Improving PM2.5 Sensor Calibration Using Transformer Model Architecture
No full textAir pollution has a significant effect on human health and the environment, with PM2.5 (fine particulate matter with diameter less than 2.5 μm) causing respiratory issues due to its ability to enter an individual’s lungs and blood stream. Low-cost PM2.5 sensors, such as those from Clarity and integrated via the OpenAQ platform, offer widespread spatial coverage but often lack the precision of regulatory-grade monitors. Recent studies for PM2.5 calibration have used statistical methods or neural networks, but these approaches are either too simple and can’t notice non-linear, complex relationships, or are risk for overfitting. Transformer models have had success in natural language processing, but remain underexplored in the context of PM2.5 and sensor calibration. This project addresses this gap by implementing a Transformer model for calibrating low-cost PM2.5 sensor data. The data contains meteorological data, such as temperature, humidity, wind speed, and PM2.5, as well as temporal features such as date and time. The model uses positional encoding for time-series data, multi-head attention to find trends and patterns, and an Encoder-Decoder structure for aligning sequences. In addition to evaluating model accuracy, the study aimed to determine how much impact the length of historical data (24-hour vs. 48-hour input times) had on model performance. Model performance was assessed using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R-squared (R²) metrics. This research fills the gap by providing a robust Transformer model approach for accurately calibrating low-cost PM2.5 sensor data, which can be impactful in environmental monitoring
Design of a Dragonfly Inspired Lighter-than-air Vehicle
No full textLighter-than-air vehicles, or blimps, are robotics platforms that possess several benefits over traditional aerial vehicles such as energy efficiency and safety in proximity to humans. While most existing blimp designs utilize traditional propellor based propulsion to generate thrust for locomotion, this project explores a bio-inspired approach to lighter-than-air vehicle design, drawing inspiration from the unique flight patterns of dragonflies. Dragonflies exhibit remarkable control, agility, and stability by independently modulating their forewings and hindwings, capabilities that guide this blimp’s mechanical design. Previous research indicates that wing-flapping mechanisms can offer greater energy efficiency and range; however, many blimps inspired by aquatic or avian creatures lack the ability to hover while carrying payloads or to execute tight maneuvers in confined spaces. This design investigates the potential increased utility and energy efficiency of dragonfly inspired flapping and articulating wings as an alternative means of lift and maneuverability
Using an Intelligent Tutoring System (ITS) to Teach Sequencing and Pattern Recognition
No full textComputational thinking skills are being increasingly recognized as critical life skills in the 21st century because they offer people the ability to uniquely approach and solve problems. Currently, CT skills are taught programmatically through activities like coding, block coding, and robotics. While this can be a helpful way to learn, it only teaches CT skills in technical contexts, making the CT skill set not as widely applicable. Most existing research looks at creating frameworks for incorporating CT skills in the classroom, and these frameworks are typically geared towards integrating CT skills in STEM classrooms. To fill this gap an Intelligent Tutoring System (ITS) was designed to help students develop two specific CT skills: sequencing and pattern recognition. OpenAI’s GPT-4.1-nano model was used to create the assistant, and was customized with instructions and sample problems the ITS could use to teach these skills through Language Arts and Math based practice problems. To teach the CT skills mentioned above, the ITS generates practice problems with increasing difficulty to ensure mastery of each topic before moving on. The topics included are Linear Sequencing, Conditional Sequencing, Repetitive Sequencing, Number Patterns, and Input-Output Tables. Each topic is broken into three difficulty levels that the ITS provides five practice problems each for. For future work I hope to test the effectiveness of this tool on elementary students by giving them a pre-test on sequencing and pattern recognition problems, allowing them to work with the tool, and then giving the students a post-test on the same skills to measure growth