Thesis/Project Final Defense Schedule
Join us as the School of STEM master’s degree candidates present their culminating thesis and project work. The schedule is updated throughout the quarter, check back for new defenses.
View previous quarter schedules
Select a master’s program to navigate to candidates:
Master of Science in Computer Science & Software Engineering
WINTER 2025
Monday, February 24
SHIVAM GANESH PAWAR
Chair: Dr. Brent Lagesse
Candidate: Master of Science in Computer Science & Software Engineering
11:00 A.M.; UW2 (Commons Hall) Room 327
Project: Spy-Shield: Hidden Wi-Fi Camera Detection through Light Manipulation and Traffic Analysis
Wi-Fi cameras are optical devices which can record/stream a video of a scene to a remote device. The term ’Wi-Fi’ indicates that they can be directly accessed over the internet. Advancements in Internet of Things (IoT) has made it easier for them to be small and affordable. These devices are wireless, easy to install, and can stream video in real time, making them popular for security and monitoring purposes. However, these cameras has raised serious concerns about privacy infringement. These discreet devices facilitate clandestine surveillance and are difficult to detect manually. This raises privacy concerns for third-party occupants, such as a guest in a hotel room who may not be aware of the deployed spy cameras. According to current methods of detecting hidden cameras, by analyzing the responsive Wi-Fi traffic flow, the detection accuracy is approximately 97%-99%. However, Wi-Fi cameras can intentionally cause delays in data transmission to avoid real-time correlations between the environment and the Wi-Fi transmission. To address this challenge specifically, Spy-Shield presents a delay-proof method for detecting Wi-Fi cameras. Spy-Shield builds on the research of Dr. Brent Lagesse in the field of “Detecting Streaming Wireless Cameras with Timing Analysis”. Using environmental light manipulation and machine learning algorithms, Spy-Shield looks for a causality between patterns in observable Wi-Fi traffic and a predetermined pattern of light flashing (signature). The proposed detection methodology can successfully discover hidden Wi-Fi cameras with an accuracy rate of 97.5%, even with a stream delay. Notably, this approach can detect without requiring connection to the same Wi-Fi network or prior knowledge of the camera’s location.
Wednesday, February 26
Austin M. Yao
Chair: Dr. Kelvin Sung
Candidate: Master of Science in Computer Science & Software Engineering
5:45 P.M.; Join Austin M. Yao’s Online Defense
Project: Infinite 2D Map Generation with Nested Wave Function Collapse
Procedural Content Generation (PCG) enables scalable, dynamic virtual worlds by automating the creation of game assets such as textures, maps, and environments. Among PCG techniques, 2D map generation, particularly for expansive, continuously generated worlds, stands out as a critical application, balancing functional design and visual coherence to shape gameplay experiences. The Wave Function Collapse (WFC) algorithm is particularly suited for this task, offering strengths in constraint satisfaction and the ability to expand small input samples into diverse, coherent outputs. This project investigated variations of WFC to identify those most suitable for real-time scalability and result coherency. This project implements a hybrid WFC design that combines a lightweight process developed by “Game Dev Garnet” for efficient runtime, a hierarchical structure from “Y. Nie” et al. for the on-demand generation, dynamic weighting as implemented by “ChrisHanna” for map content control, and important visual variables as used by “M. Kleineberg” for map aesthetics.
Additionally, the system integrates an efficient User-Controlled Travel System, enabling dynamic content assembly through weighted tile controls and real-time adjustments. The results from this project provide four key accomplishments: 1) supporting efficient infinite 2D map generation in real-time, 2) enabling designers to influence map design through weight control on tiles, 3) showcasing the impact of tile variation and decoration on map aesthetics, and 4) providing guidelines for tile and fallback tile design to maintain map coherence. Although the project demonstrates WFC’s ability to create seamless and diverse maps, challenges encountered included backtracking inefficiencies and computational overhead. These challenges serve as a reference for future investigations into 2D infinite map generation.
Thursday, February 27
JOHN FISCHER
Chair: Dr. Kelvin Sung
Candidate: Master of Science in Computer Science & Software Engineering
3:30 P.M.; Join John Fischer’s Online Defense
Project: A Reusable Inventory System
Video Games which commonly need an Inventory system use them to satisfy specific requirements for a set of mechanics. These systems can be complex due to the mechanics, their item data management, and the nuances of the game engine that they are developed for. The common game mechanic functionality of adding, removing, and swapping items, along with data management for them, is shared across many different Inventory Systems. Identification of these similarities highlights the potentials for their reuse. The focus of this work was to create a unified and coherent interface to reduce the inherent complexity of Inventory Systems by providing a generic software module that can perform the common game mechanics and manage related item data.
Numerical, Allocated, and Spatial (NAS) categorizations were adopted for Inventory Systems which allowed for common vocabulary, game mechanics, and logical requirements to be established related to the multitude of implementations available. Design guidelines were established to focus the direction of the implementation efforts that were organized through an outlined scope. The initial approach was to create a backend independent of game engine functions, which would be consumed by a frontend to instantiate the display and manage user input.
Verification of the success of the interface throughout the development process was accomplished by explicit demonstrations. The item data management, NAS common functionality and unique restrictions were used to confirm the working backend. The frontend game engine portion was verified by having visual displays that a user could interact with for each of the NAS categorizations as well as a combination of them. The generic software module, to use as the foundation of an Inventory System, was successfully created using the backend and frontend specific API approach. Future work considerations could include incorporating 3D Spatial configurations, providing specific examples on how to add niche mechanics, and generalizations for logic implementation requirements for any game engine frontend.
Monday, March 3
WILLIAM SHEN HAO
Chair: Dr. Min Chen
Candidate: Master of Science in Computer Science & Software Engineering
11:00 A.M.; Join William Shen Hao’s Online Defense
Project: Extending the Algorithm and Enhancing the User Interface of a Rhyme Detection Application for English Rap Lyrics
Rhyme is a key component in hip-hop music. Over the course of the genre, it has evolved from simple patterns such as end rhymes to more complex internal and multisyllabic rhymes. Although it is possible to highlight rhymes in lyrics by hand, this project demonstrates that an automated system for highlighting rhymes is useful for helping people understand the role of rhyme in rap music. Survey results also indicate that automatic rhyme detection aids in lyric composition. Most existing literature focuses on simpler rhyme patterns whereas research on more complex rhyme types tends to focus on lyric generation instead of detection. This project uses one of the few published works that detect more complex rhyme patterns and extends the application’s algorithm to detect a different rhyme type. It also makes improvements to usability, primarily by migrating the application from a Java GUI to a deployed website with a Spring Boot server and a React frontend. A usability survey was conducted with participants of varying amounts of background knowledge on the hip-hop music genre and responses are generally favorable. Further improvements can be made to the usability of the website, including increasing the load that the website is able to handle. Future work also includes highlighting identical patterns with the same color and making another extension to the algorithm.
NIDHI KUMAR
Chair: Dr. Min Chen
Candidate: Master of Science in Computer Science & Software Engineering
1:15 P.M.; Join Nidhi Kumar’s Online Defense
Project: Rhythm Art in Melodic Transcription for Indigenous Language Documentation and Application (MeTILDA)
Linguistic diversity is under significant threat, with over 1,500 languages projected to become extinct by 2100. One such language is Blackfoot, spoken in Alberta, Canada, and Montana, U.S.A. Blackfoot is a pitch-accent language, where variations in pitch can change word meanings, posing challenges for documentation and teaching.
This project enhances MeTILDA (Melodic Transcription in Language Documentation and Application), a cloud-based system for analyzing Blackfoot word pronunciation and generating Pitch Art. MeTILDA is a collaborative effort between researchers from the University of Washington Bothell and the University of Montana, combining expertise from linguistics, computer science, and language documentation.
Key improvements include the development of a Rhythmic Visualization tool for better representation of linguistic rhythm, a redesigned Learn page for improved usability, and a more modular codebase to facilitate long-term maintainability. These enhancements support ongoing efforts to document and revitalize the Blackfoot language while contributing to broader linguistic preservation efforts.
Friday, March 7
RITHI AFRA JERALD JOTHI
Chair: Dr. Geethapriya Thamilarasu
Candidate: Master of Science in Computer Science & Software Engineering
11:00 A.M.; Discovery Hall Room 464
Project: Hybrid Distribution Framework
User data is a double edged sword. While the analysis of user data has provided valuable insights across various domains, its security risks have posed significant threats on individual privacy. In case of individual level security, Differential Privacy has emerged as the de-facto standard to protect sensitive information by introducing calibrated noise. However, conventional DP noise injection mechanisms struggle with the privacy-utility tradeoff. The Laplace mechanism, with its zero-centered probability density and sharp kurtosis, minimizes distortion for most data points but is ineffective against outliers due to its symmetric light tails, increasing the risk of adversarial reconstruction attacks. The Pareto distribution’s heavy tails enhance outlier privacy by injecting amplified noise, but its steep central truncation degrades utility for non-outliers. The Exponential distribution provides smooth, consistent perturbation, preserving statistical continuity, yet its rapid tail decay weakens privacy for extreme values, making it vulnerable to membership and attribute inference attacks. This research introduces a Hybrid
Noise Distribution (HND) that combines the key strengths of these traditional mechanisms. By integrating the sharp central peak of Laplace for minimal utility loss, the heavy-tailed property of Pareto for robust outlier privacy, and the smooth decay of Exponential for stability, the HND balances privacy and utility. The project involves creating a probability density function, computing the normalization constant, deriving the cumulative distribution function and inversing it to generate noise. The proposed framework has been implemented and tested in mobile environments using Kotlin Multiplatform (KMM), enabling seamless integration across Android and iOS devices. On empirical evaluation The Hybrid Noise Distribution achieves up to 6.3 times better Attribute Inference Error Rate than Laplace, 5.6 times better than Pareto, and 5.9 times better than Exponential, demonstrating its superior privacy protection while maintaining data utility. Future work will focus on adaptive noise scaling techniques and integration with federated learning architectures.
Monday, March 10
RAJ DIPESH PAREKH
Chair: Dr. Geethapriya Thamilarasu
Candidate: Master of Science in Computer Science & Software Engineering
11:00 A.M.; Discovery Hall Room 464
Project: ExplainFed: An Explainable Gradient-based Framework for Detecting and Characterizing Poisoning Attacks in Federated Learning
Federated learning systems are increasingly vulnerable to poisoning attacks, yet existing defense mechanisms often lack interpretability, making it difficult for system administrators to trust and act upon their detection decisions. We present a novel gradient based detection framework that not only identifies poisoning attacks but also provides comprehensive explanations for its decisions. Our approach combines temporal gradient analysis, layer-wise examination, and pattern recognition to detect three types of attacks: label flipping, distributed backdoor, and model poisoning with activation functions (MPAF). Experiments on MNIST and Fashion-MNIST datasets demonstrate that our method achieves superior detection rates (> 90% TPR) while maintaining low false positive rates (<4%) across all attack types. The integrated explanation system generates detailed insights that reduce administrator response time by 62.4% and improve resolution accuracy by 15.1%. Our framework maintains high accuracy (>91% on MNIST, >83% on Fashion-MNIST) under attack conditions, significantly outperforming existing defenses while providing actionable explanations for detected anomalies.
ALAN LAI
Chair: Dr. David Socha
Candidate: Master of Science in Computer Science & Software Engineering
3:30 P.M.; Join Alan Lai’s Online Defense
Project: Learning Software Engineering Principles Through Designing a Validation Framework for Luna mHealth’s Authoring System
This paper describes two main aspects from my capstone project: designing a validator framework for Luna mHealth’s authoring system, and the software engineering principles learned along the way. Luna mHealth is a mobile health education framework that empowers communities to access essential educational information in an offline setting. Luna content modules are converted from PowerPoint, which are made by authors.
I created a validator framework to help alert content module authors of potential issues with their PowerPoint files. I designed and implemented validator and validation issue interfaces, and created an issue renderer class to render an issue in text for authors to read.
Along the way, I learned about my fast and ambiguous coding habits. The Luna project helped me learn about the importance of targeted refactoring to unify naming conventions, reduce duplication, and separate responsibilities. This led to a codebase that was more comprehensible and maintainable. I applied key improvements to the Luna codebase such as splitting multi-responsibility classes and systematically renaming vague variables to clarify intent. I created an onboarding document and a pull request checklist to promote iterative, test-driven development and enforce modular, reliable coding practices.
Thursday, March 13
THUAN TRAN
Chair: Dr. Min Chen
Candidate: Master of Science in Computer Science & Software Engineering
1:15 P.M.; Join Thuan Tran’s Online Defense
Project: Calendar Retrieval and Organization Using Smart Home Assistant
For calendar users, existing processes of using hands and relying on physical devices to interact with digital schedules are inefficient and inaccessible. Event organizers are not consistent in ways to create and distribute schedules for users, which makes it difficult for users to keep track of.
This project presents the development of an Alexa Skill voice assistant designed to make querying schedules more efficient and accessible, along with a set of APIs for event organizers to create and manage such schedules. Users can manage their personalized schedules with voice commands, which is ideal for visually impaired or those who want hands-free interaction. The system is backed by a back-end infrastructure where event organizers can create, modify, and manage schedules via API.
By having a common platform for both organizers and users to manage schedules, this project significantly helps reduce duplication of efforts and ensures consistent and improved experience for both organizers and users when interacting with schedules. Using this platform allows organizers to update their schedules when needed, and users to receive the latest schedule details.
Master of Science in Cybersecurity Engineering
WINTER 2025
Wednesday, March 5
CHRISTIAN BERGH
Chair: Dr. Marc Dupuis
Candidate: Master of Science in Cybersecurity Engineering
1:15 P.M.; Join Christian Bergh’s Online Defense
Thesis: Cybersecurity is Stressful: The Impact of Stress on Identifying Phishing Attacks
Stress is an emotion that impacts everyone, it can have profound physical impacts on health and cognitive functions. The responsibilities placed on average workers in many fields can lead to increased workplace stress. As technology continues to be integrated into all facets of life and work, many industries have become de facto technology companies with large or valuable technical datasets. Many of these industries are facing an onslaught of cyber-attacks in an attempt to gain access to those large or valuable datasets. One of the most common cyber-attacks, and often found to be the entry way for many data breaches, remains to be phishing attacks. This combination of factors places a large amount of responsibility for a business’s digital security on the shoulders of every employee, while often security is not the primary responsibility of this employee. Malicious actors know that humans are the most vulnerable portion of any security network and understand that employees who may not understand how their company’s digital network functions and do not understand the value of access credentials to a malicious actor can be the easiest to target. Phishing attacks are designed and created to illicit stress emotions and a sense of urgency from the target in order to trick that target into clicking a link, downloading a file, or entering their credentials somewhere they should not. This study uses the framework and core concepts developed in the Trier Social Stress Test to create an acute instance of increased stress on participants. Participants are then shown several screenshots of emails and asked to determine if the email shown is a phishing attack. By analyzing the participants stress levels before and after the phishing attack identification test along with their performance on the test we are able to determine if there Is a significant impact on a participant’s ability to identify phishing attacks under increased stress.