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 2024

Friday, January 26

MATTHEW WOERNER

Chair: Dr. David Socha
Candidate: Master of Science in Computer Science & Software Engineering
11:00 A.M.; Join Matthew Woerner’s online defense
Thesis: Identifying and Addressing the Gap Between How Students and Professionals Read Code

This project investigated and addressed the questions of: a) how do students and professional software developers read novel codebases, and b) how can we help students learn to better read code.

Our Spring 2023 study used semi-structured interviews and code reading exercises to identify and quantify several differences in the ways students and professional software developers read novel codebases. Students tended to face more difficulty with these reading tasks than the professionals due to an apparent lack of structured code reading process and an over reliance on making unverified assumptions about the code. We focused on three particular anti-patterns. Our interview data also indicated that the lack of a structured code reading process complicates transitioning into a professional atmosphere post degree, requiring new professional software developers to learn these skills on the job.

Based upon the results, we developed a module to teach students a structured way to read code in novel codebases, and to assess their improvement. The module was integrated into the Fall 2023 quarter of CSS 390 (Software Engineering Studio). Students worked their way through a variety of formative exercises leading up to a final summative assessment where they were evaluated on their performance improvement throughout the module as well as how they compared to a prior group of students given a similar assessment in the Spring quarter. Comparing the number of code reading anti-patterns exhibited by both groups, we found that the students who completed the module were significantly less likely to trace into files outside of the code path, were more likely to follow all stack traces in a code reading challenge, and were less likely to make uncorrected misinterpretations about a codebase.

Wednesday, February 21

JARDI A. MARTINEZ JORDAN

Chair: Dr. Michael Stiber
Candidate: Master of Science in Computer Science & Software Engineering
11:00 A.M.; Discovery Hall 464
Thesis: Graph-based Modeling and Simulation of Emergency Services Communication Systems

Emergency Services Communication Systems (ESCS) are evolving into Internet Protocol (IP)-based communication networks, promising enhancements to their function, availability, and resilience. This increase in complexity and cyber-attack surface demands a better understanding of these systems’ breakdown dynamics under extreme circumstances. Existing ESCS research largely overlooks simulation and the little work that exists focuses primarily on specific cybersecurity threats and neglects critical factors such as the non-stationarity of call arrivals. This paper introduces a robust, adaptable graph-based simulation framework and essential mathematical models for ESCS simulation. The framework uses a graph representation of ESCS networks where each vertex is a communicating finite-state machine that exchanges messages along edges and whose behavior is governed by a discrete event queuing model. Call arrival burstiness and its connection to emergency incidents are modeled through a cluster point process. The model applicability is demonstrated through simulations of the Seattle Police Department ESCS. Ongoing work is developing GPU implementations of these models and exploring the use of simulations in cybersecurity tabletop exercises.

Thursday, February 29

MOHAMMED KHALEELUR REHMAN

Chair: Dr. Kelvin Sung
Candidate: Master of Science in Computer Science & Software Engineering
8:45 A.M.; Join Mohammed Khaleelur Reman’s online defense
Project: Multi-tier System Performance Optimization

The College Affordability Model represents a critical data visualization tool tailored to assess the expenses associated with attending colleges in the United States. This web-based system serves as a valuable resource for policymakers, empowering them with comprehensive insights derived from exploring existing data. The architecture of the tool encompasses three distinct tiers: the database, backend, and frontend, each contributing to its functionality.

The existing College Affordability Model faced performance challenges due to additional processing during plot computations in various scenarios. Consequently, this project aimed at a thorough analysis, understanding, and optimization of the system’s performance. Drawing inspiration from industry case studies on performance optimization, a novel design solution was derived to address the identified issues. The project implemented this solution, incorporating appropriate tools for result analysis. It provides a promising verification system prototype that can be extended to the entire system that ensures marked improvement in user interaction responsiveness.

Friday, March 1

CHENGJUN XI

Chair: Dr. Min Chen
Candidate: Master of Science in Computer Science & Software Engineering
11:00 A.M.; Join Chengjun Xi’s online defense
Project: Web Education Platform for Endangered Languages

Language extinction is a concerning problem in current world. More than 1500 languages may disappear by the end of this century. Melodic Transcription in Language Documentation and Analysis (MeTILDA) is a web toolset developed to help documentation and education of endangered pitch-accent languages. A pitch-accent language is a language where the change of pitch-accent may change the meaning of words. The MeTILDA system can visualize and document the pitch movement using a novel perceptual scale. Nevertheless, it is primitive in education functionalities, only supporting learning by syllable and word.

The project presented by this paper focuses on extending the education functions of the MeTILDA system. It is a Content Management System (CMS) including six sub-systems, namely course, lesson, discussion, assignment, quiz and grading. It supports four major aspects of language education including listening, speaking, reading and writing. Besides, it integrates the pitch art component in the MeTILDA system to facilitate visualization of pitch movements in audio recording. Moreover, it inherents the cloud-based architecture of the MeTILDA system, so it can be easily integrated with the existing MeTILDA system to better support language education. With this project, the MeTILDA system will have a complete functionality of language education with unique advantages of automatic pitch visualization.

Monday, March 4

GARY LAM

Chair: Dr. Min Chen
Candidate: Master of Science in Computer Science & Software Engineering
8:45 A.M.; Join Gary Lam’s online defense
Project: Mushroom Identification Application Using Machine Learning

Mushroom foraging is a simple hobby that has few material requirements but demands a great deal of knowledge to perform safely and successfully. Attempting to forage without sufficient experience can lead to serious health consequences, or in rare cases, death. Typically, beginners gain this knowledge by learning from an already experienced guide or consulting extensive field guides. However, with new advances in image recognition and deep learning, a new tool for mushroom identification can be made possible.

This project presents such a tool, consisting of a backend prediction model and a front end application for users to interface with. Using existing databases of labeled fungi images that are further filtered and processed, a convolutional neural network is trained using transfer learning and employed as the prediction model. The front end application allows a user to upload an image and receive the most probable predictions. Unlike in existing fungus identification applications, this prediction is accompanied by detailed descriptions of the predicted species’ physical characteristics and other identifying features. In addition, it allows the user to input their own observations of the specimen and highlights matches within the known features of the predicted species. With this process, a burgeoning forager can learn to spot the distinguishing characteristics of certain species. In addition, more experienced users can employ the application to organize their notes and gain a reference to confirm their own identifications. The user inputs describing the image features are also saved, with user permission, and can be used to improve the machine learning model or build a new model capable of recognizing individual features of fungi.


MATTHEW MUNSON

Chair: Dr. Clark Olson
Candidate: Master of Science in Computer Science & Software Engineering
11:00 A.M.; Join Matthew Munson’s online defense
Project: Real-Time Evaluation of Simulated Aircraft Instruments Using Machine Vision

While digital displays and glass cockpits have become widespread in modern aircraft, analog instruments remain. These gauges can be challenging to digitize or integrate into automated safety systems. This work investigates the application of machine vision to evaluate aircraft instruments of varying complexity. For ease of acquisition, training data was recorded from a flight simulator and used to train neural networks. The resulting models have high accuracy when evaluating single pointer gauges in lighting conditions similar to the training data set, as well as with entirely different lighting conditions. Performance remains robust even with more complex instruments, such as dual pointer airspeed gauges and attitude indicators, although occasional misinterpretations of gauge pointers occur. Attempts to train models to identify instrument positions from panned and zoomed input video using labeled bounding boxes were not successful as the resulting models had low accuracy. Potential future work on this system includes applying it to real-life aircraft and integration with safety systems, including detection of instrument display failures.


XIANG LI

Chair: Dr. Michael Stiber
Candidate: Master of Science in Computer Science & Software Engineering
1:15 P.M.; Join Xiang Li’s online defense
Project: Agile Data Recording Architecture for Complex Scientific Simulations

Simulation development and eScience are driven by the complex questions that scientists and engineers want to answer. A simulation-driven or eScience investigation is an iterative process — as answers are found, new questions are created. Consequently, the development of simulation and eScience software involves rapid iteration, and the data that investigators want to capture from such software frequently changes.

Traditionally, new simulation data characteristics require development of new software modules or modification of existing ones to facilitate the recording of the updated data. This brings two disadvantages. First, scientists and engineers must invest significant time and resources into understanding and addressing data recording nuances with each iteration of their investigation. Second, the procedures developed during each iteration are of limited use in the next. This is particularly problematic in large-scale projects that involve various simulation types, where managing multiple data recording systems becomes a significant overhead. To address these issues, we have developed a flexible and scalable data recording architecture that supports a wide range of simulations and data types. This architecture was realized by redesigning the data recording subsystem within the Graphitti simulator, and we assessed the flexibility and reusability of this redesigned system by evaluating the lines of code (LOC) and examining its maintainability. We observed a complete elimination of lines of code (a reduction of 100 percent) in the updated data recording subsystem compared to the old one, specifically in the context of recording various new variables within existing simulations. This result shows that new architecture significantly reduces development needs for saving and updating simulation data across different simulation projects, as well as modifying variables within existing simulation models. Additionally, we demonstrate that this approach can easily record more data types with minimal changes (2 lines of code), thus broadening its ability to support additional fundamental data types that were not previously accommodated by the data recording subsystem. Overall, our new lightweight data-recording architecture met our project goal of supporting various simulations without requiring the development of additional software.

Wednesday, March 6

NIKHIL CHAUHAN

Chair: Dr. Brent Lagesse
Candidate: Master of Science in Computer Science & Software Engineering
11:00 A.M.; Join Nikhil Chauhan’s online defense
Project: GhostCrowdShare : A privacy preserving video sharing application

GhostCrowdShare introduces a unique approach to secure video sharing. This project implements the client-server architectures described in https://faculty.washington.edu/lagesse/publications/ppvs.pdf to enhance privacy while sharing media online. Its main aim is to allow private video sharing among specific groups based on mutual presence at an event. It ensures confidentiality until co-presence is confirmed, which holds immense value for both personal and professional realms.

Its standout feature is computing similarities between homomorphically encrypted videos, a method grounded in the above research paper. This involves video encryption and maintaining computing similarity between encrypted videos. The system is designed to be infinitely scalable horizontally as well as vertically. It can perform tasks ranging from user authentication, event creation and joining, video compression and video encryption, similarity calculation between encrypted videos and allowing eligible parties to download videos, all communicating via APIs.

This project also solves a major pain point for previous researchers using Microsoft Seal for homomorphic encryption where they had to either write wrappers on top of the original seal library or use py-seal which is a not-so-good implementation of the original library.

In essence, GhostCrowdShare paves the way for future secure video sharing, bridging the gap between privacy needs and content sharing in today’s digital age, and it also acts as a framework for any other application with similar requirements.

Thursday, March 7

PAUL JINWOO LEE

Chair: Dr. Min Chen
Candidate: Master of Science in Computer Science & Software Engineering
1:15 P.M.; Join Paul Jinwoo Lee’s online defense
Thesis: Detecting Nasal, Glottal, and Breathy Tones in Acoustic Speech using Phonetic Features and Machine Learning

Discerning subjective tones in languages with limited speech data presents a considerable challenge. This study utilizes machine learning models to analyze and demonstrate potential trends in nasal, glottal, and breathy tones, aiming for applicability across diverse speakers and languages with limited speech data.

The research leverages larger datasets to uncover patterns in acoustic features related to the tones across an array of word-based speech samples. Categorization of word samples is based on A1-P0 amplitude shift calculations for nasality and harmonic amplitude slopes for glottal and breathy tones. Manual labeling of datasets facilitates the training of machine learning ensembles, which are then tested against both split datasets and external datasets, including those from different languages.

Utilizing a silence-splicing algorithm on the Common Voice’s Development library, we extract a dataset of 7,397 English words for feature calculations related to tone detection. Machine learning models are trained on 70% of the features and tested on the remaining 30%, demonstrating initial accuracy rates of 99.99%, 99.99%, and 96.50% for nasal, glottal, and breathy tones, respectively. Testing against the Common Voice’s Test library of 7,208 English words yielded varying prediction accuracies (68.10%, 94.56%, and 95.38%) for the respective tones. Testing against the limited samples of the endangered Blackfeet language produced prediction accuracies of 55.55%, 100%, and 100% for the respective tones.

While glottal and breathy tones demonstrate more consistent performances across different datasets, nasality calculations exhibit a wide range of standard deviation values possibly due to the subjective nature of this feature. The findings emphasize the nuanced nature of calculating nasality across numerous speakers of even the same language. Challenges persist in precise tone distinctions across different languages, underscoring the ongoing need for refinement and additional audio features.

Friday, March 8

CHLOE MA

Chair: Dr. Dong Si
Candidate: Master of Science in Computer Science & Software Engineering
3:30 P.M.; Join Chloe Ma’s online defense
Project: Beyond Current Boundaries: Integrating Deep Learning and AlphaFold for Enhanced Protein Structure Prediction from Low-Resolution Cryo-EM Maps

Constructing atomic models from cryo-electron microscopy (cryo-EM) maps is a crucial yet intricate task in structural biology. While advancements in deep learning, such as convolutional neural networks (CNNs) and graph neural networks (GNNs), have spurred the development of sophisticated map-to-model tools like DeepTracer and ModelAngelo, their efficacy notably diminishes with low-resolution maps beyond 4 Å. To address this shortfall, our research introduces DeepTracer-LowResEnhance, an innovative framework that synergizes a deep learning-enhanced map refinement technique with the power of AlphaFold. This methodology is designed to markedly improve the construction of models from low-resolution cryo-EM maps. DeepTracer-LowResEnhance was rigorously tested on a set of 37 protein cryo-EM maps, with resolutions ranging between 2.5 to 8.4 Å, including 22 maps with resolutions lower than 4 Å. The outcomes were compelling, demonstrating that 95.5% of the low-resolution maps exhibited a significant uptick in the count of accurately predicted residues. This denotes a pronounced improvement in atomic model building for low-resolution maps. Additionally, a comparative analysis alongside Phenix’s auto-sharpening functionality delineates DeepTracer-LowResEnhance’s superior capability in rendering more detailed and precise atomic models, thereby pushing the boundaries of current computational structural biology methodologies.

Back to top

Master of Science in Cybersecurity Engineering

WINTER 2024

Back to top

Master of Science in Electrical Engineering

WINTER 2024

Back to top