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
SPRING 2024
Tuesday, May 14
NAIMA NOOR
Chair: Dr. Afra Mashhadi
Candidate: Master of Science in Computer Science & Software Engineering
11:00 A.M.; Discovery Hall 464
Thesis: Fairness in Continual Federated Learning
Continual Federated Learning (CFL) is a distributed machine learning technique that enables multiple clients to collaboratively train a shared model without sharing their data, while also adapting to new classes without forgetting previously learned ones. Currently, there are limited evaluation models and metrics for measuring fairness in CFL, and ensuring fairness over time can be challenging as the system evolves. To address this, our study explores temporal fairness in CFL, examining how the fairness of the model can be influenced by the selection and participation of clients over time.
We introduce novel fairness metrics—Delta Accuracy Fairness (DAF) and Delta Forgetting Fairness (DFF)—specifically designed to ensure temporal fairness in a CFL context. Additionally, we propose a set of client selection strategies that enhance the temporal fairness of the CFL model by addressing disparities in knowledge retention. Through comprehensive analysis, we demonstrate that while no single strategy guarantees perfect temporal fairness, the Low Participation and Low Average strategies consistently outperform others in terms of stability and equity. Furthermore, our findings underscore the adaptability of the Dynamic strategy, which shows significant promise in certain tasks. These insights pave the way for refining client selection strategies, enhancing CFL’s fairness, and fostering more equitable learning environments.
Wednesday, May 15
SHENYAN CAO
Chair: Dr. Munehiro Fukuda
Candidate: Master of Science in Computer Science & Software Engineering
11:00 A.M.; Discovery Hall 464
Project: An Incremental Enhancement of Agent-Based Graph Database
In the domain of big data analytics, graph Database (DB) is vital for managing complex data structures. This project focuses on enhancing an existing agent-based graph DB within the MASS Java framework. Motivated by the limitations of the existing agent-based graph DB, this project aims to enrich its capability to handle data with more detailed property information, aligning with the Property Graph Model. Through a comparative analysis of popular industry graph DBs such as Neo4j, RadisGraph, JanusGraph, and ArangoDB, this project establishes design principles focusing on the adoption of the Property Graph Model, Cypher query language, in-memory distributed graph structures, and agent utilization. The project provides detailed insights into the design and implementation processes, including parsing Cypher queries to Abstract Syntax Tree (AST), planning execution strategies, and comprehensive testing to ensure system functionality and reliability. Overall, the project demonstrates the successful extension of the agent-based graph DB to handle complex and interconnected data structures, accurate execution of CREATE and MATCH cypher queries, and outlined plans for future development.
VEDANTI PAWAR
Chair: Dr. Brent Lagesse
Candidate: Master of Science in Computer Science & Software Engineering
1:15 P.M.; Join Vedanti Pawar’s Online Defense
Project: Adversarial Defense: Implementing and Evaluating Multi-Layered Strategies Against Adversarial Attacks
Deep learning (DL) has become a cornerstone in image classification tasks across various industries, notably in the development of autonomous driving systems, where it significantly enhances vehicle perception and decision-making capabilities. However, reliance on single defense mechanisms often falls short in safeguarding these models against sophisticated adversarial attacks. This research investigates the potential of combining various defense strategies to enhance the robustness of DL models, focusing on the ResNet34 and ResNet50 architectures. By employing widely-used attack methods, this study simulates real-world threats to assess whether these combined defenses can improve model accuracy and security. Testing these strategies on different network architectures across various datasets, the analysis determines the impact of each defense combination along with their computational costs. The findings provide valuable insights into which strategies are most effective in different settings, guiding the development of more resilient DL systems against sophisticated attacks.
Thursday May 16
WUBE ALEMAYEHU TUFFA
Chair: Dr. William Erdly
Candidate: Master of Science in Computer Science & Software Engineering
3:30 P.M.; Join Wube Alemayehu Tuffa’s Online Defense
Project: Transfer Learning in Neural Machine Translation for Low-Resource Languages
This project paper explores the impact of transfer learning and pre-trained models on improving Neural Machine Translation (NMT) between the low-resource language, Amharic, and the resource-rich language, English. Given the unique challenges associated with NMT for low-resource languages, this study proposes to use two innovative architectures: the Concerted Training NMT (CTNMT) and a Bert-fused NMT model, aimed at improving translation quality. These models are evaluated against a conventional transformer model to determine their ability to effectively leverage pre-trained knowledge for language translation tasks.
The experimental approach employs the fairseq and neurST toolkit to conduct controlled experiments, with translation accuracy assessed through BLEU scores. The research consolidates two smaller corpora into an expanded Amharic-English dataset, ensuring robustness and integrity for model training and evaluation while safeguarding against data leakage into the test set. The CTNMT architecture utilizes rate-scheduling and dynamic switch to maximize learning from BERT through sophisticated training methodologies. Meanwhile, the Bert-fused model leverages BERT’s capabilities by embedding it within a custom-build sequence-to-sequence encoder-decoder framework.
The results suggest that both innovative models are effective, with the Bert-fused model achieving higher BLEU scores in both Amharic-English and English-Amharic translations compared to the baseline transformer. While the CTNMT model performed well in English-Amharic translation, it was not applicable for the opposite direction. These findings highlight the potential of pre-trained models to improve the quality of Neural Machine Translation (NMT), especially for languages with limited linguistic resources. Particularly the success of these models validates the hypothesis that integrating deep bidirectional language understanding can substantially enhance translation quality, presenting a notable advancement in the field of machine translation.
Thursday, May 23
THOMAS PINKAVA
Chair: Dr. Afra Mashhadi
Candidate: Master of Science in Computer Science & Software Engineering
8:45 A.M.; Discovery Hall 464
Thesis: Deep Reinforcement Learning for Data-Agnostic Post-Training Debiasing of Black-Box Machine Learning Models
As reliance on Machine Learning systems in real-world decision-making processes grows,
ensuring these systems are free of bias against sensitive demographic groups is of increasing
importance. Existing techniques for automatically debiasing ML models generally require
access to either the models’ internal architectures, the models’ training datasets, or both. In
this paper we outline the reasons why such requirements are disadvantageous, and present
an alternative novel debiasing system that is both data- and model-agnostic. We implement
this system as a Reinforcement Learning Agent and employ it to debias four target ML
model architectures over three datasets. Our results show performance comparable to data-
and/or model-gnostic state-of-the-art debiasers.
Master of Science in Cybersecurity Engineering
SPRING 2024
Wednesday, May 15
TIMOTHY LUM
Chair: Dr. William Erdly
Candidate: Master of Science in Cybersecurity Engineering
3:30 P.M.; Discovery Hall 464
Project: The Howdu (“How do I…?”) Project: A Knowledge Management System to Support Cybersecurity Implementation
Knowledge capture and distribution has been a perennial human endeavor since prehistory. In the modern era, the internet has facilitated an exponential growth in the volume of information available, however it presents a sub-ideal resource in providing consistent, structured instruction for how best to implement effective cyber defenses.
In this capstone, we evaluate the technical pressures and missing linkages that have prevented guidance from being presented to mitigate cyberattacks. We then build a cyber defense knowledge repository that allows users to create a cumulative snapshot of their experiences and insights.
In limited testing, Howdu facilitated a 61% speedup of an arbitrary and partially obfuscated task for those performing it (Practitioners). It further altered the subjective perception of this task from “”Confusing””, “”Frustrating””, and “”Ambiguous”” to a more positive outlook of “”Easy””, “”Comprehensible””, and “”Fun””. These initial results suggest the application’s ability to improve network defense by aiding defender efficiency, decreasing stress, and reducing burnout.
Future Works include an integration of the system for grading information – called the Trust Index – and implementation of a system for translating knowledge articles across languages – called the Gnosetta. Other long term goals include containerization of the application and reductions in third-party service reliance.
Master of Science in Electrical Engineering
SPRING 2024
Tuesday, May 14
CHERYL KUNG
Chair: Dr. Walter Charczenko
Candidate: Master of Science in Electrical Engineering Engineering
3:30 P.M.; Discovery Hall 464
Thesis: Low-Cost 3-D Printed Helical Antenna with Dielectric Support
Satellite-based internet connection requires high directivity, millimeter wave phased array antennas to be able to receive and transmit signals effectively. Phased array antennas for millimeter waves have historically been very expensive to manufacture. Exploring low-cost methods for manufacturing high directivity antennas may bring down the costs of these systems, allowing more equitable access to internet.
Helical antennas are a type of high directivity antenna that can be used for these purposes. However, helical antennas are difficult to manufacture and scale due to its three dimensional (3-D) shape of the helix conductor. New 3-D printing technology allows the creation of a dielectric support for the helical antenna element. This adds mechanical rigidity to the antenna and is feasible for high volume manufacturing at a lower cost.
This thesis explores the design of a low-cost helical antenna using a 3-D printed dielectric core for mechanical support. The research in this thesis concludes that it is possible to design a helical antenna using low-cost dielectric materials with high relative permittivity at microwave frequencies. As a proof of principle, a 5 GHz helical antenna embedded in a solid dielectric was designed and modeled using electromagnetic field simulation software. At 5 GHz, the software simulations can be compared to helical antennas that are manufactured on conventional 3-D printers and commonly used resin dielectrics. The conclusion and results of the computer simulations show that helical antennas with dielectric support will radiate in the axial mode with high directivity and circular polarization.