7903 research outputs found
Sort by
Wearing Saris and Suits on Capitol HillNavigating Elite White Spaces as South Asian American Women in Politics
No full textThis preliminary research paper explores how South Asian American women navigate the elite white nature of U.S. politics, focusing on how they balance authentically representing their cultures and communities while navigating varying levels of power that prioritize individual advancement. Through initial interviews with South Asian American women in prominent political roles, and by applying established theoretical frameworks on identity-based hierarchies and social justice leadership, this study aims to offer insights into how South Asian American women approach the challenge of conforming to white standards in politics while maintaining their ethnic and cultural identities
Smart End Effector Docking System
No full textThe Smart End Effector Docking System (S.E.E.D.S.) serves as an autonomous docking system to assist with replenishing fluid and charging the Agricultural Robot (AgBot) that has been developed by the Robotic Systems Laboratory. The docking system aims to improve autonomous capabilities by allowing the AgBot to recharge and transfer fluid independently, removing the need for human involvement. The S.E.E.D.S. project can additionally help to address labor shortages and safety hazards associated with agricultural workers by providing several functions including fluid transfer and power transfer for charging
Reimagining School Leadership: Latinx Administrators Leveraging Community Cultural Wealth to Support Latinx Student Success: A Qualitative Study in East San Jose Grounded in Counter-Storytelling, Critical Race Theory, and Latino Critical Theory
No full textThis qualitative study examines how Latina administrators in East San Jose leverage their Community Cultural Wealth (Yosso, 2005) to support Latinx student success in K-12 education. Although Latinx students comprise the largest racial/ethnic group in California public schools (California Department of Education, 2021), Latinx leaders remain significantly underrepresented in educational leadership. Guided by Critical Race Theory (Delgado & Stefancic, 2017), Latino Critical Race Theory (LatCrit) (Solórzano & Bernal, 2001), and Yosso’s (2005) framework of Community Cultural Wealth, this study centers the counter-stories of three Latina administrators to explore how their cultural identities, lived experiences, and relational practices inform leadership.
Data were collected through in-depth interviews and analyzed using counter-storytelling methods (Solórzano & Yosso, 2002). Three themes emerged: (1) aspirational leadership rooted in cultural identity, (2) navigating barriers and biases in educational leadership, and (3) creating culturally affirming educational spaces. Participants drew upon aspirational, familial, resistant, navigational, social, and linguistic capital to resist deficit narratives, expand academic opportunity, and foster inclusive school climates. Their leadership practices represent resistance and transformation, exemplifying what Theoharis (2007) describes as social justice leadership—daily efforts to dismantle inequities and affirm student identity.
The study culminates in the Community Cultural Wealth Leadership Praxis Model, a conceptual framework integrating cultural knowledge, equity-focused practice, and transformational leadership. This work contributes to the scholarship on culturally sustaining leadership and offers implications for school leaders, preparation programs, and policymakers committed to justice and equity in education
2024 State of the Library
Get PDF◼ Nicole Branch, University Library Dean Strategic Activities Budget Staffing Highlights from the Units
◼ Melanie Sellar, Assistant Dean, Learning & Engagement Instruction Research & Student Support Outreach & Programming Access & Delivery Services Learning Commons Space
◼ Lev Rickards, Assistant Dean, Collections & Scholarly Communication Stewarding the Library acquisitions budget Transformative agreements Folio Archives & Special Collections
◼ Spotlight
◼ Staff Appreciations
◼ Questions & Discussio
Applied Auto-tuning on LoRA Hyperparameters
Get PDFThis senior design project explores the application of Bayesian optimization-based auto-tuning techniques on the low-rank adaptation (LoRA) fine-tuning of large language models (LLMs), demonstrating how fine-tuning methods can reduce training times and costs, albeit with a slight trade-off in accuracy. However, little is known about the optimal hyperparameters on LoRA and its variants for those methods. This project addresses this lack of knowledge by analyzing data gathered from auto-tuning LoRA hyperparameters to determine the most optimal parameter configurations for a model’s accuracy and training efficiency.
The team has implemented a pipeline utilizing many different technologies. The main technology driving the Bayesian optimization-based auto-tuning is GPTune, a performance auto-tuner built on Gaussian Process regression. The team selected Llama3 as the base model for testing due to its high-performance capability and relevancy. These models are fine-tuned using the QLoRA framework and evaluated on their training time and loss. GPTune uses Bayesian optimization to efficiently explore the Pareto front of training time and loss for varying QLoRA parameters.
This research has contributed to LLMs by demonstrating the efficacy of Bayesian optimization-based auto-tuning techniques in fine-tuning LoRA. By applying these techniques, we have shown that it is possible to optimize hyperparameters more efficiently, reducing computational resource requirements and improving model performance. Our work provides a proof of concept for integrating advanced auto-tuning frameworks like GPTune with existing fine-tuning tools, paving the way for more accessible and cost-effective use of LLMs in various applications
Daily Digest: A News Aggregation Site
Get PDFA staggering amount of news articles are uploaded every day – approximately 5,000 in the United States alone [1]. That volume of information causes difficulty for many people who try to stay up-to-date with current events. The number of articles and the multitude of sources that they come from can feel overwhelming. In our project, we attempt to tackle this issue. We use web scraping to collect a dataset of news articles and combine it with a Large Language Model (LLM) capable of processing those articles to generate news summaries for the user. The user interacts with the program through a website, which presents them with a list of topics that may interest them. The user can select a few articles, similar or dissimilar to each other, to generate a summary of the key points of the selected articles. Overall, this approach is useful for abbreviating large amounts of information into more digestible pieces. Caution must be taken, however, as summaries generated by the LLM may not be completely factual. The user must remain vigilant as there may be inaccuracies in the generated summaries. As LLMs, training datasets, and the power of computer hardware advance, the accuracy and quickness of the generated summaries will continue to increase