ScholarWorks (California State University)
Not a member yet
87523 research outputs found
Sort by
Oasis Builder: Transforming Urban Heat Islands through a Collaborative AR Experience
No full textUrban Heat Islands (UHIs) elevate temperatures in urban areas, creating both public health and environmental challenges. This thesis presents Oasis Builder, a collaborative Augmented Reality (AR) mobile application developed with Unity and the Niantic SDK, designed to educate users about UHIs. By integrating dynamic target image tracking and real-time temperature data from the Open-Meteo API, Oasis Builder allows users to visualize local heat levels and place cooling elements such as trees and canopies to reduce perceived heat in their surroundings. Experimental evaluations confirm that Oasis Builder reliably detects target images for colocalization and maintains stable AR object placement under diverse lighting and environmental conditions, significantly enhancing usability and educational impact. Through its gamified approach, the application transforms passive learning into interactive, team-oriented experiences that encourage informed decision-making and create awareness of the impact that the users can make in their immediate environment. By translating environmental data into intuitive, AR-based interactions, Oasis Builder bridges the gap between abstract knowledge and tangible action. As a result, it not only highlights the potential of AR technologies as powerful educational tools but also underscores their value in rallying collective action for more climate-resilient cities
XAI-PM: Enhancing interpretability in predictive maintenance models
No full textThe ever-growing reliance of machine learning models in various areas, especially in the predictive maintenance field has called for the need for more interpretable models that could expedite decision-making processes and create a sense of trust in these systems. The aim of this thesis is to apply techniques of Explainable Artificial Intelligence (XAI), to enhance interpretability in predictive maintenance models. The initial step was to understand the data. Hence, the study first explores the AI4I dataset for predictive maintenance using various visualization and analysis techniques. After cleaning and pre-processing the dataset, 7 popular machine learning algorithms- K- nearest neighbors (KNN), Support Vector Classifier (SVC), Random Forest Classifiers, Extra Trees Classifier, Gaussian Naive Bayes, XGBoost and Adaboost are applied for training, testing and validation. The parameters of the models are tuned using the GridSearch algorithm to bring out the best from each one of them. Using Logistic Regression as a benchmark, the performance of these models on both binary and multi-class classification is studied, analyzed and compared in much depth. The metrics used to quantify the performance are- precision, recall, accuracy, AUC score, F-1 score, F-2 score and the confusion matrix. This work aims to demonstrate how complex, black-box models can be made more transparent. A number of XAI tools are used to accomplish this, such as Partial Dependence Plots (PDPs), LIME (Local Interpretable Model-agnostic Explanations), and SHAP (Shapley Additive Explanations). These techniques help users better understand how the models are arriving at their results by offering insights into feature importance, model behavior, and the influence of specific features on predictions. This research underscores the value of incorporating XAI into predictive maintenance workflows, providing actionable insights while ensuring trust in the automated decisions made by machine learning systems
Edge AI on Kria KR260: Implementing ResNet-50 Classification Using DPU TRD and YOLOX Detection with DPU-PYNQ
No full textThe rapid development of artificial intelligence at the edge has created a lot of demand for specialized hardware accelerators capable for supporting deep learning models with minimal latency and power consumption. This project focuses on deploying a custom-designed Deep Learning Processing Unit (DPU) on the Kria KR260 to accelerate the workload of deep learning models: image classification using ResNet-50 and object detection using YOLOX. The primary purpose of the project is to evaluate the easiness in implementation, reusability and performance of DPU across multiple deployment environments specifically petalinux for embedded applications and Ubuntu with Python based interfaces via DPU-PYNQ. A DPU overlay was generated using Vivado with the use of DPUCZDX8G and modifying it for KR260 platform. After synthesis and implementation runs the .xsa , .bit and handoff files were exported for further use. Two deployments were done with the exported files: the first involving building of a petalinux image with integrated Vitis AI 3.0 runtime to execute Resnet-50 model, the second used Ubuntu 22.04 to run the DPU-Pynq 3.5 framework allowing real-time YOLOX interface through a PYNQ interface. Performance evaluation showed that the DPU provided low-latency, high-throughput inference in both workflows. The ResNet-50 model achieved fast classification times on static images, while the YOLOX pipeline supported real-time object detection at approximately 15-20 FPS. Both implementations demonstrated successful integration of DPU architecture. To sum up, this project shows how capable and adaptable the DPU is on the Kria KR260, making it a solid foundation for both low-level embedded tasks and more advanced AI applications. These results point to its strong potential in real-world edge computing scenarios
Addressing High-Risk Alcohol Consumption Among CSUN Students: Development and Implementation of a Digital Peer-Led Alcohol Education Program
No full textThe Klotz Student Health Center (Klotz center) is located at California State University, Northridge (CSUN), serving students and the community. This paper focuses on CSUN students with autism. Those with autism are often challenged when trying to communicate or engage in everyday interactions, even more so for those that are nonverbal and/or deaf. The proposed program will help to determine the best intervention methods to improve communication and social interactions, so college students who are diagnosed with autism are more successful. Improving communication skills will lead to possible lifesaving communicate when in crisis and improvements to quality of life. This paper investigates why such a program is not yet available and how accessing these types of programs can lead to improvements in the overall health of this population. This program, at the Klotz center, will allow for better understanding of the different perspectives that persons with autism experience when trying to communicate with others. The Klotz center can be used to eliminate barriers to communication and social skills for college students with autism
Identifying and Mitigating Misinformation Spread on Social Media
No full textIn today's age of social media, false information spreads faster than ever, creating some serious challenges. Misinformation has even influenced main events such as election outcomes, making it a serious threat to society. One of the major challenges in detecting misinformation is its rapid spread. The circulation of false information has escalated to the point of creating a "misinformation crisis" in crucial areas such as elections, healthcare, and social justice. Conventional fact-checking methods struggle to keep up with the sheer volume and speed of fake news spread, necessitating the adoption of machine learning driven approaches. This project explores an advanced method for detecting misinformation using natural language processing and machine learning. The proposed method integrates real-time data processing, contextual analysis, and improved automated fact-checking. It utilizes deep and supervised learning models like BERT along with knowledge-based verification techniques to increase the scope of finding out misinformation
Psychedelic assisted therapy (PAT) content analysis: Discourses in United States and Netherlands
No full textIndividuals with addiction, anxiety, depression, existential distress, and post-traumatic stressdisorder (PTSD), have experienced rapid and long-lasting relief of debilitating symptoms afteringestion of psychedelics. After experiencing treatment resistance or dropout, patients experiencedsuccess with psychedelics they had not with prescribed medicine. The Netherlands (NL) and theUnited States (US) are considering psychedelic-assisted therapy (PAT) as a novel treatment. Ianalyzed the reasons for this, and they will be described and explained in this thesis. NL has beenknown for its legislative tolerance toward drugs while the US is known for the anti-drug policiescommonly referred to as the War on Drugs. I conducted a qualitative content analysis of texts fromNL and the US using open and thematic coding. I found NL's reasons to explore PAT are relatedto harm reduction and a desire to globally lead research, however, barriers to expansion includebureaucracy, public opinion, and funding. In the US, the primary motivations for PAT are helpingveterans with PTSD, tech billionaire funding and addressing the nationwide mental health crisis.In the US, barriers to the expansion of PAT include bureaucracy and limitations on populationsdeemed legitimate for access to pharmaceuticals. Further, I found that while sexual trauma andmilitary sexual trauma (MST) make up criteria for PTSD and other diagnoses, they are not oftendiscussed in conversations about PAT for PTSD or other diagnoses that PAT may be useful for,contributing to the silence around sex, sexual trauma and military sexual trauma (MST).https://doi.org/10.46569/t722hk61
Math Identity, Persistence, and Academic Outcomes
Get PDFMany students enter their high school math classroom with a negative mindset that influences how they work and perform in the classroom. This study used a survey, student interviews, and document analysis to understand students' math identity and how it influences their persistence in the classroom and their academic outcomes. To explore this, I examined the following research questions: 1) What factors do my students believe shape their math identities? 2) What connections, if any, are there between my students' math identity and their persistence in the classroom? 3) What connections, if any, are there between my students' math identities and their academic outcomes? The major findings are 1) past experiences and gender influence math identity; 2) all students demonstrated persistence regardless of math identity; and 3) math identity and academic outcome have a positive correlation. This study suggests that helping create positive math classroom experiences for all students could improve their math identity, and therefore improve their grade
MABLESim: A Tool for Simulating Indoor Accessibility
Get PDFChallenges arise for people with disabilities when built environments do not accommodate their needs, resulting in significant accessibility issues. Regulations and initiatives often struggle to effectively address the diverse requirements of these populations. MABLESim (Mapping for Accessible Built Environments Simulator) is a framework designed to tackle these challenges by providing a solution for evaluating the accessibility of indoor environments. By converting floor plans into digital models, MABLESim facilitates thorough testing and analysis of various scenarios for people with disabilities. This section will demonstrate how MABLESim operates, showcasingits ability to replicate real-world conditions and user experiences. This makes it a tool for architects, urban planners, and accessibility researchers. These users can interact with the simulation to explore how different mobility needs impact navigation, allowing them to observe potential barriers andinefficiencies. Ultimately, MABLESim offers insights that inform the design of more inclusive and user-friendly spaces, fostering environments that better accommodate the diverse needs of all individuals
Exploring How Humor Shapes Student Engagement, Learning, and Behavior
Get PDFThis research study used surveys, observations, and document collection methods to examine how humor shapes the engagement, learning, and behavior of my fifth-grade students. The findings suggest that I use four main types of humor in my classroom: stories (quick anecdotes designed to make the students laugh), exaggerated statements (statements about the content designed to be funny through exaggeration), different pitches (using higher and lower pitches based on the context to make the comment humorous), and exaggerated movements (e.g., jumping up and down). In addition, the findings show that while humor influences engagement and behavior, more research is needed to clarify its impact on learning. Implications for teaching include the timing and usage of different types of humor in the classroom
A Comparative Study of Hollow Fiber Filtration with Perfusion and Fed-Batch Strategies in N-1 Stage Bioreactors
No full textThis Semester-in-Residence report is submitted in partial fulfillment of the Master of Biotechnology degree at California State University, San Marcos, and was conducted at Genentech's manufacturing facility. The research focuses on the implementation of hollow fiber filtration (HFF) with perfusion in N-1 stage bioreactors, comparing single, double, and continuous-fed batch strategies. The primary objective is to identify process improvements that enhance production efficiency and product quality. Key findings demonstrate that continuous-fed batch systems using hollow fiber filtration had superior performance over traditional fed-batch strategies, achieving up to 40 times higher viable cell density than the double-fed batch strategy and 10 times higher than the single-fed batch strategy, with significantly greater process efficiency. Overall, HFF-perfusion implementation proved to be the most efficient in yield, offering a strong return on investment and faster deployment. This study highlights the feasibility of using HFF-perfusion systems as a compelling solution for biomanufacturing. The study concludes that implementing HFF-perfusion systems for continuous clarification of high cell density cultures increases production throughput and efficiency. These findings support the broader adoption of continuous processing and HFF technologies in biomanufacturing, aligning with industry trends toward sustainability and multi-product facility flexibility. Further research is recommended to explore the application of HFF and other continuous processing strategies across upstream and downstream bioprocessing stages, including real-time monitoring, advanced analytics, and expanded use in vaccines and viral vectors, to enhance efficiency, scalability, and product quality across a broader range of biologic manufacturing