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

SUMMER 2024

Wednesday, July 24

TASNIM BASHAR

Chair: Dr. Clark Olson
Candidate: Master of Science in Computer Science & Software Engineering
8:45 A.M.; Join Tasnim Bashar’s Online Defense
Project: Enhancing Image Inpainting with a Novel Deep Learning Fusion

Image inpainting, the art of seamlessly filling missing or damaged parts of an image, has seen remarkable progress with deep learning. However, achieving consistently high-quality restorations across diverse image content remains a challenge. This project presents a novel image inpainting framework that harnesses the strengths of partial convolutions, Generative Adversarial Networks, and self-attention mechanisms. Our approach utilizes partial convolutions to effectively address irregular holes and preserve intricate image details. A Generative Adversarial Network architecture is incorporated to encourage the generation of realistic and visually plausible image content. Furthermore, integrating self-attention enables the model to capture long-range dependencies and contextual information within the image, leading to more coherent and higher-quality reconstructions. Evaluations using established image quality metrics demonstrate that our framework achieves superior performance compared to existing state-of-the-art methods, confirming the effectiveness of this innovative fusion in significantly enhancing image quality and restoration precision.


GREESHMA SREE PARIMI

Chair: Dr. Min Chen
Candidate: Master of Science in Computer Science & Software Engineering
11:00 A.M.; Join Greeshma Sree Parimi’s Online Defense
Project: Enhancing Collaborative Language Analysis and Study in MeTILDA

The “Global Predictors of Language Endangerment and the Future of Linguistic Diversity” study predicts that over 1500 languages will become extinct by 2100. Blackfoot is one such endangered language majorly used in the regions of Alberta in Canada and Montana in the U.S.A. Blackfoot is a pitch accent language where the meaning of a word varies depending on the pitch used when speaking it. So, it is very challenging to document and teach the language. To address this, researchers from the University of Washington Bothell (UW) and the University of Montana (UM) have collaborated on an interdisciplinary project named MeTILDA (Melodic Transcription in Language Documentation and Application). It is a cloud-based system that analyzes the pronunciation of individual Blackfoot words, generates Pitch Art, and assists in documenting, teaching, and learning Blackfoot.

This capstone project primarily aims to enhance the MeTILDA application’s analysis, collaboration, and security features. To improve analytical capabilities, we developed the Pitch Art Version Control System, allowing users to work on and save multiple versions of the same Pitch Art. For improved collaboration, we integrated a communication mode within the application, enabling users to interact with both MeTILDA and non-MeTILDA users. To strengthen security, we implemented role-based access control for all features of the MeTILDA application and introduced email verification for new user registrations. Furthermore, substantial improvements were made to the project documentation, particularly in the areas of front-end components and database information.


HARPREET KOUR

Chair: Dr. Wooyoung Kim
Candidate: Master of Science in Computer Science & Software Engineering
1:15 P.M.; Join Harpreet Kour’s Online Defense
Project: Predictive Modelling of Substance Abuse: Analysing Key Features with Machine Learning

Substance abuse remains a critical public health challenge with multifaceted implications for individuals, families, and communities worldwide. This project explores how machine learning techniques can predict and classify substance abuse behaviours, focusing on alcohol, tobacco, marijuana, and nicotine vaping. Leveraging data from the (2021-2022) Behavioural Risk Factor Surveillance System (BRFSS) in 2023, we aim to identify the key predictors of substance abuse and groups at higher risk, taking into account factors like adverse childhood experiences, financial conditions, and mental health status.

Prediction models were developed using four types of machine learning algorithms, including Linear Models (Logistic Regression & SVM), Tree based Models (Random Forest, XGBoost & ADABoost), Neural Networks (MLP & CNN) and Clustering algorithm. Respondents were randomly divided into training and testing samples. The performance of all the models was compared using accuracy, precision, recall, AUC and false positive rate. The study included 31060 respondents of whom, 5867 (19%) were found to be substance abusers. Of the respondents who reported substance abuse 62.93% were between the ages of 18-64, 60.61% were males and 84.76% were non-Hispanic Whites. Random Forest was the best performing model with AUC 0.86, followed by XGBoost (AUC 0.85). The most important factors for substance abuse were BMI, male sex, lower income levels, young adult age group, lower education levels, poor mental health and adverse childhood experiences. Data mining methods were useful in examining patterns across demographics, health conditions and lifestyle behaviours so as to understand the co-morbidities associated with substance abuse.

Another goal of this project was to highlight the importance of collaboration between domain experts and machine learning practitioners and assess the impact it has on the results compared to when domain experts are not involved. Their contributions in feature selection and data interpretability solutions were instrumental in achieving this enhancement. We prioritised model interpretability to foster trust and refine understanding. Additionally, our project introduces a novel approach to interpretability, analysing misclassified data, offering insights into substance abuse dynamics. These results can be used to generate further hypotheses for research, increase public awareness and help provide targeted substance abuse education.

Friday, July 26

JEFFREY MCCREA

Chair: Dr. Munehiro Fukuda
Candidate: Master of Science in Computer Science & Software Engineering
11:00 A.M.; Join Jeffrey McCrea’s Online Defense
Project: Enhancement of Agent Performance with Q-Learning

Graphs store data from various domains, including social networks, biological networks, transportation systems, and computer networks. As these graphs grow in size and complexity, single-machine solutions become impractical due to limitations in computational resources. Distributed graph computing addresses these challenges by leveraging multiple machines to process and analyze large-scale graphs collaboratively.

This capstone project investigates the enhancement of distributed graph computing performance in the Multi-Agent Spatial Simulation (MASS) library by integrating Q-learning for computing shortest path, closeness centrality, and betweenness centrality on distributed large-scale dynamic graphs. This approach is compared to traditional and agent-based graph computing algorithms. Previous approaches in the MASS framework relied on large populations of unintelligent agents to exhaustively traverse graphs to compute solutions, making them inefficient when faced with dynamic graph data. By leveraging Q-learning and the distributed agent-based graph capabilities of MASS, we aim to optimize the decision-making processes of distributed agents, thus improving computational efficiency and accuracy.

Experimental results demonstrate that the adaptive learning mechanism of Q-learning, coupled with the MASS library, allows agents to dynamically adjust to changing graph structures, leading to a more robust and scalable distributed graph computing solution. This research contributes to the field of distributed systems and artificial intelligence by providing an innovative approach to enhancing multi-agent intelligence for graph computing tasks.

Monday, July 29

SAURAV JAYAKUMAR

Chair: Dr. Dong Si
Candidate: Master of Science in Computer Science & Software Engineering
11:00 A.M.; Join Saurav Jayakumar’s Online Defense
Thesis: GraphConv: Geometric Deep Learning for Multiple Conformation Generation from Electron Density Images

In the realm of cryo-electron microscopy (cryo-EM) structural analysis, the precise prediction of molecular conformations within datasets stands as a fundamental endeavor. Despite strides made in deep learning methodologies, existing solutions often yield volumes of suboptimal quality. Addressing this critical limitation, our research introduces GraphConv, an innovative encoder model designed to embed particle images into a latent space, thereby supplanting the conventional encoder utilized by CryoDRGN. This novel approach employs a Graph Neural Network (GNN) architecture featuring multiple GraphConv and Convolutional layers, aimed at capturing richer information from particle images and faithfully reconstructing corresponding 3D volumes. Rigorous testing across two authentic datasets and three simulated datasets underscores the efficacy of our model, showcasing marked enhancements in reconstruction quality. Notably, our findings reveal enhancements in resolution by up to 20\% compared to CryoDRGN. By harnessing the power of GNNs, our methodology heralds significant advancements in the fidelity and accuracy of output volumes, thereby contributing to the ongoing refinement of cryo-EM structural analysis methodologies.

Tuesday, July 30

SUGAM JAISWAL

Chair: Dr. Dong Si
Candidate: Master of Science in Computer Science & Software Engineering
8:45 A.M.; Join Sugam Jaiswal’s Online Defense
Thesis: Development of Personality Adaptive Conversational AI For Mental Health Therapy Using LLMs

Many individuals with mental health issues cannot get access to professional help due to reasons such as lack of awareness, limited availability, and high costs. Conversational agents present a viable alternative to deliver mental health support that is accessible, affordable, and scalable. However, the effectiveness of these agents can vary among users, as different users have different personality types such as extroversion, agreeability, etc. which influence how users interact with chatbots. Therefore, it is important to develop therapy chatbots that adapt to individual personalities. In this study, we highlight the significant role of Personality Adaptive Conversational Agents (PACAs) in mental healthcare. We designed an architecture around traditional ML models and open-source LLMs to build a PACA for mental health (based on the existing iCare project). We built a functional prototype based on it and conducted a user study, which concluded that personality adaptability is a critical feature for mental health chatbots.

During this research, we were able to build a personality classifier that achieved an average F1-score of 0.96 across the Big Five personality dimensions – Agreeableness, Extraversion, Openness, Conscientiousness, and Neuroticism, and successfully integrated that with an open-source LLM to generate adapted responses. The remote user study demonstrated that 95% of the test users found the responses from the adaptive chatbot relevant to their situation compared to 30% for the non-adaptive chatbot, and 55% of users agreed that the responses felt suited according to their personality as opposed to 15% for the non-adaptive chatbot. With this study, we have shown that it is feasible to create free, accessible, and personalized mental healthcare solutions and that the adoption of PACAs could represent a pivotal step toward making mental healthcare more personalized and widely available.


ANI AVETIAN

Chair: Dr. Annuska Zolyomi
Candidate: Master of Science in Computer Science & Software Engineering
1:15 P.M.; Join Ani Avetian’s Online Defense
Project: The Views Application: Introducing Augmented Reality into Travel Applications

The travel and tourism industry is growing worldwide and experiencing a resurgence following the COVID-19 global pandemic travel restrictions. When people travel to unfamiliar places, they need to orient themselves to their surroundings, find their way around, and gain information about the local culture. Traditional travel applications, like websites and smartphone apps, are for arranging travel and providing information. However, the information and services provided by websites and apps are presented out-of-context with the real-world surroundings of the traveler. Emerging technology, specifically, Augmented Reality (AR), presents an opportunity to enhance the traveler’s experience with real-time information that is easy to access and contextually relevant.

Here we introduce a novel AR travel application called “Views.” This app uses image detection to allow users to easily learn about the landmarks they visit. Once detection is complete, users will have an AR environment where they can interact with their surroundings. A fact will be displayed in AR with the help of Artificial Intelligence (AI) working in the background. This provides an experience that a user can effortlessly incorporate into their travel plans and activities. Additionally, this project entailed identifying potential cybersecurity risks that come with these kinds of applications and developing a potential solution to them. We looked at attacks of clickjacking in AR environments and developed a solution to combat these attacks with image detection. As we envision a future where wearable devices may become the norm, applications like this will be at the forefront of these innovations. Using this application as a stepping stone, we begin to understand the important aspects of AR travel user experience and implementation approaches to build usable and secure AR features.

Thursday, August 1

HARIKA CHADALAVADA

Chair: Dr. Min Chen
Candidate: Master of Science in Computer Science & Software Engineering
11:00 A.M.; Join Harika Chadalavada’s Online Defense
Project: I’poyít – Blackfoot Language Learning Application

The Blackfoot language, a vital part of the cultural heritage of the Blackfoot tribes of the Great Plains in North America, is facing imminent risk of extinction with only about 3,000 fluent speakers remaining. This paper details the development of an application aimed at revitalizing the Blackfoot language by leveraging modern technology to make learning accessible and engaging. The application was developed using the versatile Flutter framework, ensuring seamless cross-platform compatibility.

The primary goal of this application was to build a comprehensive solution that addresses and fills the gaps identified in existing Blackfoot language learning applications and provides a more effective and engaging learning experience. To achieve this goal, the application features distinct student and admin login interfaces, with Firebase serving as the backend to manage user data and interactions efficiently. The application provides a comprehensive suite of interactive learning modules that include vocabulary exercises with flashcards and audio pronunciation guides, along with phrase modules that facilitate practical language use in everyday contexts. To enhance the learning experience, the application incorporates gamified elements such as experience points, badges for progress, and quizzes that test and reinforce language skills. A leaderboard and a discussion forum are integrated to foster a community of learners who motivate each other through shared achievements and discussions. Significantly, the content within the application is curated by a linguistics expert known for her extensive research in Blackfoot phonology to ensure that the educational material is authentic and culturally resonant. This project not only offers a practical solution to the preservation of the Blackfoot language but also serves as a model for preserving other endangered languages, ultimately demonstrating how digital technology can be harnessed to safeguard linguistic diversity.


ARSHEYA RAJ

Chair: Dr. Erika Parsons
Candidate: Master of Science in Computer Science & Software Engineering
1:15 P.M.; Join Arsheya Raj’s Online Defense
Thesis: NeuroGaming: Innovative Rehabilitation for Upper Limb Neurologic Conditions Using Mixed-Reality Simulation Games and EEG/EMG Biofeedback

Recent advances in Augmented Reality (AR), Mixed Reality (MR), Electroencephalogram (EEG), and Electromyogram (EMG) offer significant opportunities in medicine and neuroscience. This research aims to use these technologies to aid stroke patients with upper limb extremity weakness. This work extends the Edge Computing Ecosystem for Neuroscience Patients’ Rehabilitation, part of the ‘Stroke Rehabilitation Project’ by University of Washington Bothell Engineering, University of Washington Seattle Neuroscience, and Rehabilitation Medicine at Harborview Medical Center (UWHM). Recently, UW Bothell’s CSSE has also contributed, focusing on solutions for stroke patient rehabilitation. Traditional motor rehabilitation is costly, resource intensive and often monotonous, reducing patient engagement. We introduce “NeuroGaming”, an interactive approach engaging elements of AR / MR games that can enhance rehabilitation programs and improve patient outcomes. Using augmented and mixed reality technologies, interactive environments can be created on mobile devices, providing engaging and motivating experiences for patients. AR simulates real-world scenarios, offering a safe and fun way to practice tasks and aid in
rehabilitation.

We used EEG and EMG sensors to conduct experiments and to collect data in a controlled environment targeting a reduced set of representative relevant motor tasks. The data were processed using various signal processing and statistical techniques, which in combination with the MR / AR game can be used to build a novel feedback and guidance system. This system is a building block of our “NeuroRehab” ecosystem, which will use various ML models and algorithms for sequential prediction, with the aim of guiding patients through an optimal rehabilitation path within the game environment.

Results showed that the combination of Frequency Filtering, ICA, and ERP with FNN and SVM models yield, so far, the best accuracy for classifying the motor tasks in EEG and EMG data. These findings contribute to the field of stroke rehabilitation of upper limb extremity weakness. It also contributes to a larger project that aims for better understanding and rehabilitation of other neurological ailments by offering insights to different hand gestures using EMG, and EEG data, and creating a framework for data processing, and feedback systems.

Back to top

Master of Science in Cybersecurity Engineering

SUMMER 2024

Wednesday, August 7

MANUEL M. DUARTE

Chair: Professor Mark Kochanski
Candidate: Master of Science in Cybersecurity Engineering
1:15 P.M.; Join Manuel M. Duarte’s Online Defense
Project: Analysis of Enhancing Security Measures in Docker Containers: A Threat Model Approach

This project aims to explore new mitigation strategies and understand the risks
associated with deploying Docker as an architectural solution. We researched various security
issues with Docker container images and reviewed Docker’s architecture and its application
usage. In addition to this analysis, we investigated industry best practices to understand
what can mitigate these threats effectively. We created a Threat Model specifically for
Docker Components and analyzed threats using STRIDE analysis.

Based on our findings, we present a few recommendations for enhancing the secure use of
Docker in application deployment, i.e. when using Docker Hub, always use a security scanner
tool before deploying any images and disable unused services on a Dockerfile; we mention
other good security practices that should be applied in the early stages of development.
Our future work will involve further experiments focusing on using ML real-time anomaly
detection algorithms such as support vector machine (SVM) and Isolation Forrest to detect
vulnerabilities early on, in SSL protocols and other attacks to root access. Also, using
machine learning to improving certificate management. Continued research in this area is
essential to deepen our understanding of Docker container security and to develop robust
defenses against emerging threats.

Keywords: Docker, Dockerfile, Anomaly detection, Encryption STRIDE, DREAD.

Back to top

Master of Science in Electrical Engineering

SUMMER 2024

Back to top