ScholarWorks (California State University)
Not a member yet
87523 research outputs found
Sort by
Inferring at-home gait parameters of older adults using floor vibrations
No full textAs the aging population in the United States grows, there is an urgent need for inventive tools to help healthcare professionals better support older adults. One critical area of focus is detecting health deterioration after hospital stabilization, which could lower readmission rates and enhance patient results and healthcare system efficiency. Research has shown that gait parameters—walking patterns assessed in clinical settings—can perform as vital health indicators. Developing this capability to monitor older adults' mobility within their homes during recovery presents critical potential benefits. Automated gait tracking in daily living environments could enable early detection of health changes, reduce the need for frequent medical interventions, and support more useful health management, ultimately benefiting patients and the healthcare system. This research paves the way for improved home-based health monitoring for older adults by combining cutting-edge algorithms with real-world testing environments. These advancements could be crucial in reducing hospital readmissions and costs, but they could also bring benefits effectively, enhance independent living, and improve overall healthcare outcomes. A comprehensive data acquisition system has been created to advance this effort, simplifying the collection of walking experiment data. This system has developed over 700 datasets, helping algorithm development and establishing a strong research foundation starting at San Francisco State University, Science Building, Room 130, and continuing at ThorntonBuilding, Room 117. A key invention in this work is the Floor Accelerations for Gait Estimation and Detection (AGED) algorithm, which accurately identifies gait events and extracts cadence from floor vibration data. Validated against APDM wearable sensors—the current gold standard—the AGED algorithm has demonstrated an impressive accuracy of 99.7%. The Probabilistic-FEEL (pFEEL) algorithm has been designed to precisely locate events such as footsteps or falls, using a transfer function model that defines impact responses as a hyper-surface over time and space. These technological improvements establish a robust framework for improving gait analysis. Before deploying this system in real-world residential scenes, it must be tested in a practical environment. To achieve this, a wooden walking path copying standard residential layouts has been designed and constructed in a laboratory setting. Sensors have been installed along the route and are experiencing system identification tests to ensure structural consistency and data reliability. This setup will allow researchers to assess the effectiveness of the developed algorithms in real-world conditions. Furthermore, in the future, it will enable a review of how different flooring materials—such as ceramic or porcelain tiles in kitchens and bathrooms, and wood or carpet in living areas—affect system measurements.https://doi.org/10.46569/k3569f09
Curtain Call
No full textWhen their beloved high school drama teacher dies suddenly, six of his most ardent students can barely cope. Led by freshmen, Christy, the group engages in a DIY seance over the course of one night to bring him back for a proper goodbye. However, they soon learn their hero is not who they thought he was. The man they've put on a pedestal kept many secrets, some of them unforgivable.https://doi.org/10.46569/6m312010
Connecting Policy and Practice: The Implementation of The Family Medical Leave Act
No full textIn today's workforce, where employee benefits like job-protected leaves are vital to employee satisfaction and organizational success, the Family Medical Leave Act (FMLA) serves as an important part of legislation supporting work-family balance and influencing workplace culture. A qualitative analysis of secondary data is used to analyze the FMLA and its elements that affect its implementation in public organizations. The research finds that while workplace policies like FMLA can boost an employee's well-being, their effectiveness is limited by inconsistent implementation and accessibility. However, this shows both the challenges and opportunities associated with the implementation of the FMLA. These findings indicate that a cultural shift in public organizations is needed to equally support all employees and strengthen the impact of the FMLA
A Comparative Analysis of Traditional Machine Learning and Retrieval-Augmented Generation (RAG) for Stock Price Prediction
No full textPredicting stock prices is one of the most challenging problems in financial data analysis. Markets are usually shaped by both historical trends and unexpected events in the real world, which makes accurate forecasting especially difficult. Traditional machine learning models, including those built on Long Short-Term Memory (LSTM), are proven to perform well in capturing patterns in historical data and modeling time-based relationships. However, because they have a limitation relying entirely on past information, often struggle to adapt when something new suddenly shifts the market, such as an earnings report or a major economic development. As a result, their predictions often fall short during periods of volatility. This study explores a new approach by adapting Retrieval Augmented Generation (RAG), a method originally developed for natural language processing, to the task of stock price forecasting. The central idea is to enhance LSTM models with a retrieval component that brings in relevant external information such as recent earnings reports and macroeconomic indicators. This added context allows the model to make more informed predictions, especially when the market reacts to new developments. Four models are compared to test this approach. The first is a traditional LSTM trained only on historical stock data. The second and third models augment this baseline with earnings data and macroeconomic data, respectively. The fourth model combines both types of external information. Historical price data is sourced from Kaggle's S&P 500 dataset, while real-time external data is retrieved through the Alpha Vantage API. All models are evaluated using standard metrics including Root Mean Squared Error and Directional Accuracy, with special attention given to performance during impactful market events. The results show that models enriched with external data provide more accurate and responsive forecasts. Beyond improving prediction performance, this work demonstrates that retrieval-based methods like RAG can be effectively applied outside of language tasks. By combining the strengths of traditional time series models with dynamic external context, this approach offers a more flexible and realistic way to forecast stock prices in today's fast-moving markets
Automating Casting Notices (a.k.a. Breakdowns) for Talent Agencies
No full textCasting agents, managers and other entertainment industry representatives spend huge amounts of time manually scanning casting breakdowns for various clients. I intend to use my understanding of computer concepts and methodologies and apply the knowledge to automate this process. My project aims to cross-match a stream of ongoing casting notices, also known as breakdowns, with existing client data to generate alerts to their reps when there's a match with the ongoing daily dynamic data. This will ultimately streamline the current, time-consuming manual casting process, wherein reps must read every casting breakdown with limited filters to cover a broad range of clients. My project utilizes search functions, mapping, data mining algorithms such as data retrieval, Database Management Systems (DBMS), K-Nearest Neighbors (KNN), AI for quick matches and other coding processes. I will demonstrate that my project can handle the tedious tasks by separating the required data from the mountains of noisy data. It will reduce the time and attention required to complete the daily work done by talent representatives. It will free these agents from awaiting the next breakdowns to pop up as a major benefit of automation, allowing for time to work on talent deals and even expand talent rosters. My "QuickMatch" project can immediately notify agents via email when a match is found and keeps records of all transactions. My approach is to use an integrated development environment (IDE) called RStudio to perform the clustering algorithm and the Apriori practice to search frequent itemsets and extract them to an area where an agent can locate and be notified. This can be done by creating an automation function to generate specific outputs and notifications, then apply verification and validation to ensure my program runs correctly. For my project, I began by creating dummy data in a database repository rather than access the licensed breakdowns. Using these techniques will expedite a perfect match for a role while providing agents with user-friendly clean data. These tools will save them a considerable amount of time evaluating roles that may not apply to all their actors, increase efficiency in talent search and review
Seismic Reliability Analysis with the Direct Probability Integration Method Utilizing Adaptive Sampling
No full textStructural failures during earthquakes highlight the need for accurate dynamic response predictions. Standard methods yield accurate results but can be computationally expensive. As such, it is necessary to employ techniques that can produce similar results with higher efficiency. This study proposes two adaptive sampling methods, Adaptive GFD-Based Point Selection and Adaptive Kriging-based Point Selection, in conjunction with the Direct Probability Integration Method to estimate the probability of failure more efficiently. Adaptive GFD demonstrated higher accuracy and better replicability, while Kriging adapted well to model behavior but showed more conservative estimates. These methods provide tailored options depending on ground motion characteristics, but the accuracy of each method is ultimately dependent on the ground motion case being analyzed.https://doi.org/10.46569/n8710168
The Withering Eye
No full textThis is a short story collection of short feminist gothic horror stories. With these stories as a portal, I will attempt to delve deep into the psychological alienation of the female experience while also making the familiar unfamiliar. Taking folktales and myths that are well known, I want to subvert previous moralistic undertones to give us something new to take away.https://doi.org/10.46569/n8710167
Blue Light Effects on Cognitive Arousal in EEG during the Stroop Task
No full textThis study examines cognitive performance and neural activity during a Stroop task under different lighting conditions, using electroencephalography (EEG) to assess changes in brainwave activity. The research was designed as a 3 × 2 × 2 quasi-experimental mixed factorial design, with chronotype (Larks, Doves, Owls) and testing time (morning vs. evening) as between-subjects factors, and lighting condition (ambient vs. blue light) as a within-subjects factor. While the full factorial design was intended to explore complex interactions between chronotype, time of day, and lighting condition, limitations in participant recruitment prevented sufficient cell sizes across all groups, thereby restricting the feasibility of conducting a fully powered omnibus ANOVA. As a result, targeted within-subjects analyses were conducted to examine the effects of blue light exposure on both cognitive performance (reaction time and accuracy on the Stroop task) and EEG-derived measures of cognitive arousal (Alpha, Beta, and Theta band power). Paired-samples t-tests were used to compare Stroop performance and EEG activity before and after light exposure, and repeated-measures ANOVAs were applied separately to Alpha and Beta bands, and to Theta, to assess lighting effects over time. The central aim of this thesis is to investigate whether blue light can induce a state of cognitive arousal, as evidenced by improvements in Stroop task performance and EEG power changes consistent with heightened alertness reported in the literature.https://doi.org/10.46569/hm50v189
CycPUF: Cyclic Physical Unclonable Function
No full textIn the realm of hardware security, Physical Unclonable Functions (PUFs) have emerged as a critical technology for device authentication and secure key generation. PUFs leverage inherent manufacturing variations to produce unique responses to specific input challenges. However, traditional PUFs, such as Arbiter PUFs (APUFs), Ring Oscillator PUFs (ROPUFs), and Butterfly PUFs (BPUFs), are increasingly vulnerable to machine learning-based modeling attacks, which can predict their responses with high accuracy. This thesis introduces CycPUF, a novel framework that integrates feedback loops into delay-based PUF architectures to enhance their security against such attacks. CycPUF modifies conventional PUF designs by incorporating cyclic feedback mechanisms, which dynamically alter the challenge-response behavior. This approach generates a broader spectrum of output behaviors, including binary, steady-state, oscillatory, and pseudo-random responses, thereby complicating the modeling process for adversaries. Our evaluations demonstrate that CycPUFs significantly improve resistance to various machine learning attacks, including Logistic Regression, Support Vector Machines, Covariance Matrix Adaptation Evolution Strategy, and Deep Feed-Forward Neural Networks. We implemented CycPUFs on Field-Programmable Gate Arrays (FPGAs) and conducted comprehensive assessments of their power consumption, spatial requirements, and functional performance metrics. The results indicate that while CycPUFs incur a modest increase in hardware complexity and power usage, they achieve near-ideal levels of uniqueness, uniformity, and reliability. Specifically, CycPUFs exhibit enhanced uniqueness, making them more distinguishable from one another, and maintain high reliability and uniformity under varying environmental conditions. We subjected CycPUFs to fault injection attacks to evaluate their robustness against hybrid modeling attacks. Even with introduced faults, CycPUFs maintained superior security compared to their non-cyclic counterparts. This thesis also explores the potential applications of CycPUFs in secure device authentication and key generation, highlighting their practical benefits in real-world scenarios. CycPUF represents a significant advancement in PUF technology, offering a robust solution to the growing threat of machine learning-based modeling attacks. By leveraging cyclic feedback, CycPUFs enhance the security and functional metrics of traditional PUF designs, paving the way for more secure and reliable hardware security primitives
Parent Stress Levels When Having a Child with a Disability
No full textPurpose: Having a child/children diagnosed with a disability has impacts that extend beyond the child/children that are affected. Prior research has focused on highlighting the challenges child/children face when diagnosed with a disability and to a lesser extent, their mothers. Not much attention has been directed to the difference between maternal and paternal stress. Research Question: This study will compare the relationship between gender and parental stress, specifically highlighting how maternal and paternal stress levels differ when having a child/children with a disability. Methods: A systematic literature review was conducted to evaluate studies from 2000 to 2024. The subjects of focus are parental mental health, stress, and the gaps in studies revolving around fathers' perspectives. Results: The findings highlighted the need for more research to focus on paternal experiences. It also reveals the importance of providing mental health support and resources for parents of child/children with a disability as they are often overlooked in interventions as the focus is directed on the child/children with a disability. Discussion: By identifying the specific stressors that parents with a child/children with a disability face, we can better understand what contributions, changes and developments need to occur for these parents to receive the support they need