Matching Items (723)
Filtering by
- Creators: Computer Science and Engineering Program
Description
Machine learning has increasingly played a pivotal role in societal decision-making. In such contexts, ensuring the fairness of models becomes critically important. Unfortunately, prior studies have shown that without intervention, models often inherit and even amplify biases present in the datasets. Existing group fairness metrics, such as true positive rate parity and equalized odds, primarily focus on the relationship between model predictions and ground truth labels stratified by sensitive attributes. However, most existing notions of fairness overlook whether the underlying rationale of a model’s decision-making process varies across different subgroups. To fill this gap, we propose a novel metric, FIDSHAP, which evaluates fairness through explainability by quantifying discrepancies in the model’s decision rationales across groups. Subsequent experiments and optimization procedures validate the effectiveness of this metric and underscore the potential of addressing fairness from the perspective of explainability.
ContributorsQiu, Ziyue (Author) / Choi, YooJung (Thesis director) / De Luca, Gennaro (Committee member) / Barrett, The Honors College (Contributor) / Computer Science and Engineering Program (Contributor)
Created2025-05
Description
Many small to mid-sized businesses face significant challenges in integrating artificial intelligence (AI) effectively, primarily due to a lack of internal expertise that bridges technical capabilities with core business operations. This gap leads to operational inefficiencies, increased costs, and potential competitive disadvantages. CAIO Labs addresses this critical issue by providing specialized AI consulting coupled with tailored training programs designed to empower internal team members as dedicated Chief AI Officers (CAIO). Through detailed operational audits and strategic training, CAIO Labs enables sustainable AI-driven transformations, significantly reducing dependency on costly external consultants. By specifically targeting businesses within professional services and e-commerce sectors, CAIO Labs positions itself uniquely in the market, offering personalized solutions that yield measurable improvements in productivity and scalability. This thesis outlines CAIO Labs' strategic approach, validates its core business assumptions, and demonstrates the viability and impact of embedding internal AI expertise within growing businesses.
ContributorsBhangale, Parth (Author) / Byrne, Jared (Thesis director) / Dearman, Jeremy (Committee member) / Barrett, The Honors College (Contributor) / Computer Science and Engineering Program (Contributor)
Created2025-05
Description
Paris, renowned for its culinary heritage and multicultural landscape, offers a unique setting where diverse gastronomic traditions intersect. Among these, Asian food options occupy a significant and growing presence, inviting questions about how cultural identity is both showcased and perceived within the framework of French society. This research seeks to answer the question: How do visitors perceive Asian food options in Paris as reflections of cultural identities within the context of France's secular ideals? By examining visitor perceptions, the study aims to uncover the ways in which food becomes a site of cultural dialogue, adaptation, and sometimes tension, within the broader context of France’s approach to diversity and secularism.
ContributorsMarria, Shreya (Author) / Briggs, Georgette (Thesis director) / Foy, Joseph (Committee member) / Barrett, The Honors College (Contributor) / Computer Science and Engineering Program (Contributor)
Created2025-05
Description
In an age of quickly growing technology, a persistent need exists for code that, while functionally similar, is tailored to specific use cases across websites, applications, and software systems. With the rise of powerful large-language models (LLMs) such as ChatGPT, Claude, Copilot, and more, there is growing potential to reduce redundant coding by reusing and adapting existing solutions. This project explores Microsoft’s PowerApps, Azure, and Google Cloud, which are three platforms that have suites of services that can create complex full-stack software using a more functional approach with an emphasis on programmatic simplicity, ease of connection, utilizing endpoints and API calls, and visual programming. Specifically, the variety of essential building blocks that Azure, Google Cloud and PowerApps provides and their use-cases to emulate an application built for our Capstone project: a chatbot teaching assistant that scrapes information from canvas shells; vectorizing them and utilizing a similarity search to feed relevant pieces of information into the large language model to provide contextually relevant answers related to available information within the course. Throughout the process of building the application through PowerApps, differences between how various functions are implemented in a traditionally built application using mainly Python, Qdrant vector database, PostgreSQL, and other API libraries will be studied, compared, and contrasted to the functionalities provided by the services offered by PowerApps and cloud platforms. This analysis reviews the viability of no/low-code development in the modern software development scene and demonstrates that core-level functionality can be fully integrated using these Low-Code software development environments.
ContributorsSingh, Archit (Author) / Feng, Gregory (Co-author) / Chen, Yinong (Thesis director) / De Luca, Gennaro (Committee member) / Barrett, The Honors College (Contributor) / Computer Science and Engineering Program (Contributor)
Created2025-05
Description
In this project, my partner Archit and I looked to explore the future of no-code/low-code development. Throughout this project, we initially researched low-code/no-code development environments before settling on Microsoft Power Platform and Google Cloud to implement our capstone project, a retrieval augmented generation large language model teaching assistant that uses course-specific content pulled directly from Canvas and fed into the model. At the end of this project, we were able to fully replicate our capstone project's base frontend and the majority of our project's backend within two weeks, with a functional chat screen, two-factor authentication using a verification code, and a built-in user statistics report as compared to our capstone project, which took a whole semester to create.
ContributorsFeng, Gregory (Author) / Singh, Archit (Co-author) / Chen, Yinong (Thesis director) / De Luca, Gennaro (Committee member) / Barrett, The Honors College (Contributor) / Computer Science and Engineering Program (Contributor)
Created2025-05
Description
The Computer Science(CS) students who attend Arizona State University(ASU) are required to participate in a capstone project. For two semesters, they work on a team with other students to help a sponsor who has requested some work to be completed by the students. In some cases, this results in a group of young students, typically ages 20-22, creating a project intended for an audience much older than themselves, 45 and above. This can pose specific challenges as younger computer users can overestimate others' familiarity with technology. Students may create designs they feel are well crafted, which fail to account for an older demographic’s detached web browsing experience. With this situation in mind, a survey was created to test the web design sensibilities within a group of college students and a group of people 30 years their elders. When analyzing the results, the answers didn’t display a large discrepancy across age groups, however, the free response sections showed a large divide between the age groups. The 18-24 age block conveyed a greater familiarity with the technology they use. They’re more confident in their ability to use the tools provided to them compared to the 45+ age block whose short answers display a hesitant attitude toward the computer. This disconnect was further exemplified by survey questions, which resulted in short and unhelpful answers.
ContributorsEllis, David (Author) / Malpe, Adwith (Thesis director) / Dorsey, John (Committee member) / Barrett, The Honors College (Contributor) / Computer Science and Engineering Program (Contributor) / School of International Letters and Cultures (Contributor)
Created2025-05
Description
A natural language processing (NLP) chatbot is a program that can communicate with a human by processing their language into understandable commands. While most associate AI with LLMs, these models are not as effective with specific, involved tasks. The goal of this thesis is to demonstrate how NLP can be combined with a small-scale generative AI model to create a chatbot that can complement larger projects. The thesis researches the benefits of a small-scale chatbot in contrast to larger models in cost, time efficiency, and accuracy, and it details an example of the implementation of a small-scale chatbot within a larger project. For the implementation, I have collaborated with my sponsor, Northrop Grumman, to integrate an NLP chatbot into their GSE Frontend project. The chatbot interacts with the user, requesting specific commands related to log history, graphing, and obtaining data from the main program. The result of the implementation is an effective tool that complements the main program’s purpose with little cost and error and has great expandability alongside the program to improve its functionality.
ContributorsKhondoker, Maheeb (Author) / Osburn, Steven (Thesis director) / Arora, Aman (Committee member) / Barrett, The Honors College (Contributor) / Computer Science and Engineering Program (Contributor)
Created2025-05
Description
This work introduces a novel optimal transport framework for probabilistic
circuits (PCs). While it has been shown recently that divergences between
distributions represented as certain classes of PCs can be computed tractably,
to the best of our knowledge, there is no existing approach to compute the
Wasserstein distance between probability distributions given by PCs. In this
work, we propose a Wasserstein-type distance that restricts the coupling mea-
sure of the associated optimal transport problem to be a probabilistic circuit.
We then develop an algorithm for computing this distance by solving a series
of small linear programs and derive the circuit conditions under which this
is tractable. Furthermore, we show that we can easily retrieve the optimal
transport plan between the PCs from the solutions to these linear programs.
Lastly, we explore approaches to parameter learning for PCs that minimize
the empirical Wasserstein distance between a PC and a dataset, and provide
two approaches that minimize this distance.
ContributorsCiotinga, Adrian (Author) / Choi, YooJung (Thesis director) / Byeon, Geunyeong (Committee member) / Barrett, The Honors College (Contributor) / Computer Science and Engineering Program (Contributor) / School of Mathematical and Statistical Sciences (Contributor)
Created2025-05
Description
AdhereWear is a sustainable fashion venture aimed at extending the life cycle of custom apparel through the use of reusable, sticker-based designs. Traditional event-specific apparel often contributes to overproduction and waste, especially on college campuses where custom T-shirts are frequently produced for short-term use. AdhereWear offers an eco-friendly alternative by introducing modular, interchangeable stickers that can be applied to a base garment, allowing for easy customization across multiple events. This project explores product development, market testing, pricing models, and operational strategies to validate the feasibility of this approach. By combining creativity, sustainability, and user-centered design, AdhereWear empowers consumers to reduce garment waste and rethink how identity and apparel intersect.
ContributorsAgrawal, Saanvi (Author) / Gunti, Emilie (Co-author) / Perla, Ameyally (Co-author) / Byrne, Jared (Thesis director) / LaRosa, Julia (Committee member) / Barrett, The Honors College (Contributor) / Computer Science and Engineering Program (Contributor) / School of Sustainability (Contributor)
Created2025-05
Description
AdhereWear is a sustainable fashion venture aimed at extending the life cycle of custom apparel through the use of reusable, sticker-based designs. Traditional event-specific apparel often contributes to overproduction and waste, especially on college campuses where custom T-shirts are frequently produced for short-term use. AdhereWear offers an eco-friendly alternative by introducing modular, interchangeable stickers that can be applied to a base garment, allowing for easy customization across multiple events. This project explores product development, market testing, pricing models, and operational strategies to validate the feasibility of this approach. By combining creativity, sustainability, and user-centered design, AdhereWear empowers consumers to reduce garment waste and rethink how identity and apparel intersect.
ContributorsGunti, Emilie (Author) / Agrawal, Saanvi (Co-author) / Perla, Ameyally (Co-author) / Byrne, Jared (Thesis director) / LaRosa, Julia (Committee member) / Barrett, The Honors College (Contributor) / Computer Science and Engineering Program (Contributor) / Dean, W.P. Carey School of Business (Contributor)
Created2025-05