Project:<\/strong> Offline Health Module Sharing in Luna mHealth: A Flutter-Based Peer-to-Peer Approach with Google Nearby Connections<\/p>\n\n\n\nThis capstone project presents the design and implementation of a peer-to-peer content sharing system for the Luna mHealth mobile application, an offline-first platform that delivers health education in low-connectivity environments. Luna distributes structured, audio-visual health modules on low-cost smartphones, where internet access cannot be assumed. To support local content dissemination, this work implements a device-to-device transfer system using Google Nearby Connections within a Flutter architecture. The project evaluates prior Bluetooth Low Energy approaches and motivates the selection of a higher-level framework to reduce protocol complexity and improve reliability. The system is modeled using finite state machines to provide predictable connection lifecycles, failure recovery, and multi-peer coordination. A module transfer pipeline with metadata exchange, file streaming, and integrity validation ensures safe ingestion and resharing. The result is a module sharing system that strengthens Luna\u2019s offline distribution model and supports community-driven propagation of health education content.<\/p>\n\n\n\n
AUTUMN 2025<\/h3>\n\n\n\nThursday, November 20<\/h4>\n\n\n\nJAMES TRUONG<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Kelvin Sung
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
11:00 A.M.
Project:<\/strong> Realistic Illumination in AR<\/p>\n\n\n\nAugmented Reality (AR) creates a unique user experience by superimposing digital content onto real-world environments. With the addition of applying realistic visual cues, AR objects can be further involved and properly integrated into a user\u2019s real-world environment to achieve higher immersion and believability. Unfortunately, accurately applying realistic illumination and spatial conditions remains a technically challenging task. Diverse environments and exposure to real-world phenomena often vary in complexity and visual acuity.<\/p>\n\n\n\n
Two primary approaches exist to address this challenge: machine learning and image-based methods. Machine learning based methods can simulate real-world illumination on AR objects by recognizing illumination patterns and features in images, but are reliant on the quality, quantity, and diversity of their training data. In contrast, image-based approaches offer advantages in flexibility and usability where a balance in image quality and hardware performance can still provide a satisfactory result. Unfortunately, image-based approaches remain sensitive to hardware limitations and image quality which can significantly affect the performance and outcome.<\/p>\n\n\n\n
This project focuses on the latter, emphasizing the development of a practical and accessible solution for realistic AR lighting on common consumer devices. The implemented system utilizes two mobile devices: one device gathers real-world environmental data, and the other establishes and renders the AR object. This configuration allows realistic illumination of AR objects in near real-time under dynamic indoor lighting conditions. The resulting system provides a practical and accessible solution for integrating digital objects into real indoor environments allowing them to appear consistent with real physical objects in terms of lighting, shadow, and reflection. However, the performance is limited in complex or rapidly changing environments where noise is more prominent.<\/p>\n\n\n\n
Friday, November 21<\/h4>\n\n\n\nSIQIAO YE<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Michael Stiber
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
3:30 P.M.
Project:<\/strong> Incident Classification of Emergency Services Communication System (ESCS)<\/p>\n\n\n\nIn the context of next-generation 911 (NG911), Emergency Service Communications Systems (ESCS) are becoming increasingly complex, requiring operators to answer calls and then categorize events based on the content of the conversation. This study explores metadata-based event classification methods and how their performance depends on the spatiotemporal structure of incoming calls, providing operators with an initial pre-classification before answering. We synthesized 911-like datasets using a graph-based clustering point process simulator, built a dual-input Long Short-Term Memory (LSTM) network classifier, and compared it with classic non-sequential baseline models. We generated three spatiotemporal separation modes: temporally overlapping but spatially separable; spatially overlapping but temporally separable; and simultaneously spatiotemporally overlapping. The results reveal clear patterns, especially when spatial cues are ambiguous, where temporal memory becomes crucial, and the LSTM model performs better. On a representative test set, the overall accuracy and macro F1 score typically exceed 85%, demonstrating robust performance and highlighting the necessity and option of sequence modeling in metadata-only ESCS classification.<\/p>\n\n\n\n
Tuesday, November 25<\/h4>\n\n\n\nPOULAMI DAS GHOSH<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Min Chen
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
11:00 A.M.
Project:<\/strong> Visual Question Answering System Using Gated Bidirectional Cross-Attention<\/p>\n\n\n\nThis capstone project presents a cost-effective and reliable Visual Question Answering (VQA) system designed to improve accessibility for visually impaired users. VQA is a multimodal AI task that requires joint reasoning over image and text inputs to answer open-ended questions. The delivered solution leverages a novel gated bidirectional Cross-Attention mechanism that integrates BERT-based text embeddings with spatial and global image representations from ResNet-50 to generate answers. In bidirectional Cross-Attention, the AI model dynamically fosters combined learning by allowing the text to attend to visual features and vice versa, enabling a richer multimodal understanding. Meanwhile, the gating algorithm enables the model to filter out irrelevant information by regulating the flow of data between attended and original features, thereby reducing the overall computational overhead of the VQA system. Using this approach, we achieved an accuracy of 58% on the VQA v2 validation set, a 10% improvement over the hierarchical co-attention-based baseline work. Furthermore, the system is deployed as a user-friendly and intuitive web application that supports both voice and text input\/output for enhanced usability.<\/p>\n\n\n\n
Wednesday, November 26<\/h4>\n\n\n\nDHURKA ROHINI MADASAMY<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Wooyoung Kim
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
11:00 A.M.
Thesis:<\/strong> Predictive Robustness and Biomarker Discovery in Rare Cancers: An Interpretable Machine Learning Approach<\/p>\n\n\n\nRare cancers such as Glioblastoma Multiforme (GBM, a brain rare cancer) pose persistent challenges in computational oncology due to severe data scarcity, biological noise, and the difficulty of identifying disease specific molecular signatures. This thesis develops an interpretable machine learning framework to evaluate the predictive robustness of GBM gene expression signals and to isolate biologically meaningful biomarkers under extreme class imbalance. Across four classification settings and multiple resampling strategies, the models achieve consistently strong performance (F1 and MCC values often above 0.95), demonstrating that the GBM transcriptomic signature remains highly separable even under challenging experimental conditions. To refine this signal, a novel Cascade Learning approach systematically removes common cancer pathways and reveals biomarkers specific to the rare cancer. SHAP based interpretability identifies a stable set of genes aligned with known glioma biology, and a complementary Tab2Image analysis offers visual validation of class separability through structured image based embeddings. Alongside these methodological results, the thesis also addresses ethical considerations inherent to rare disease research by emphasizing fairness, representation, and accountability when working with structurally imbalanced datasets. Overall, the study provides a robust, transparent, and ethically grounded pathway for biomarker discovery in rare cancers.<\/p>\n\n\n\n
\n\n\n\nARAVIND TALLAPRAGADA<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Erika Parsons
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
3:30 P.M.
Project:<\/strong> Fish Species Identification System for Enhanced Fishing Experience<\/p>\n\n\n\nThe goal of this project is to research and set the foundations to develop an ML-based fish species identification system to assist recreational anglers while supporting marine research and species preservation in Washington State. The system employs a ResNet-50 convolutional neural network fine-tuned on 2291 images of eight fish species commonly found in Washington\u2019s coastal waters. The project explores several experimental setups and configurations, leading to a 94.25% classification accuracy.<\/p>\n\n\n\n
The model is integrated into a full-stack web application that aims to enable users to upload fish images and receive prompt identification (under 5 seconds). The system also captures and stores environmental metadata including location, timestamp, and historical weather conditions via the OpenWeatherMap API. Species-specific information is retrieved through the iNaturalist API, while conservation status is provided via the IUCN Red List API.<\/p>\n\n\n\n
By providing a platform for accessible species identification along with taxonomic and environmental context, this system aims to deliver a practical tool recreational fishers and an infrastructure for citizen science contributions for ecological monitoring. This project demonstrates practical application of machine learning and cloud technologies to create a functional application that helps with ecological conservation research while serving as a tool for more conscientious recreational fishing.<\/p>\n\n\n\n
Monday, December 1<\/h4>\n\n\n\nNICOLAS POSEY<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Michael Stiber
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
8:45 A.M.
Project:<\/strong> Implementing a GPU Model for Simulating Emergency Services Communication Systems<\/p>\n\n\n\nGraphitti is a general-purpose, high-performance, graph-based simulator designed to significantly speed up the simulation of network event models, such as biological neural simulations, using NVIDIA GPUs. Graphitti also provides an architecture designed to allow for the validation of GPU simulation results against single-threaded CPU simulation results. This enables researchers and other users of Graphitti to be confident in the correctness of their parallel implementations. Previous work at the University of Washington, Bothell\u2019s Intelligent Networks Laboratory (INL) has succeeded in creating and validating a single-threaded CPU implementation of a Next Generation 911 (NG911) Emergency Services Communication System (ESCS) model. This Master\u2019s capstone project builds on this work, specifically, by creating a NG911 GPU implementation to support the simulation of larger networks. Two kernels, advance911VerticesDevice and maybeTakeCallFromEdge, were implemented using the CUDA programming language to parallelize the most computationally intensive components of the NG911 model. The runtimes of these kernels and of a simulation consisting of 1932 vertices, 34,119 calls, and 288,000 timesteps were measured to assess the performance of the implementation. Runtimes of the GPU implementation were slower than those of the CPU implementation, for advance911VerticesDevice (2.63x); for maybeTakeCallFromEdge (1.84x); and for the total simulation time (2.08x). Future work will be needed to achieve performance improvements for the GPU implementation by examining possible causes such as the complexity of the advance911VerticesDevice kernel and the inability to get the advance911VerticesDevice kernel sufficiently busy during the simulation. Because this is the first instance in the INL of creating a GPU implementation from an existing CPU implementation, the difficulty of the creation process using the existing architecture is analyzed to understand if architectural changes should be made to improve this process for future models. A normalized functional code change (NFCC) metric is computed to measure the difficulty. A NFCC value for all files changed during the GPU implementation was found to be 0.1359, demonstrating that the process for creating a GPU implementation from an existing CPU implementation is not difficult.<\/p>\n\n\n\n
\n\n\n\nJEREMY NEWTON<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Wooyoung Kim
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
11:00 A.M.; UW2 (Commons Hall) Room 327
Thesis:<\/strong> Interpretable Machine Learning for Biomarker Identification in RNA Seq Cancer Data<\/p>\n\n\n\nExisting research on RNA Seq gene expression biomarkers has provided various methods to select a small list of genes as cancer biomarkers from a large number of gene expression data. Previous methods for identifying potential gene expression cancer biomarkers have focused on statistical analysis, but other methods have incorporated machine learning, often including Interpretable Machine Learning (iML) techniques. On 16 cancer type cohorts of TCGA data, we used inherently interpretable machine learning models: Logistic Regression, Random Forest, and Linear Support Vector Machine to narrow down lists of potential genes as biomarkers using the trained model’s feature importance rankings. We subsequently applied the model-agnostic iML techniques Shapley Additive Explanations (SHAP) and Permutation Importance to narrow down the subsets even further. We compared classification performance between machine learning models trained on selected features from these methods with features selected by statistical methods, and biomarkers from external research. We cross-checked the highly scored potential biomarkers with biomedical annotations and gene pathway analysis. We found that these biomarker selection methods lead to comparable or better classification performance on these datasets than the biomarkers from outside research, or from statistical analysis alone. Mutual Information estimation was a surprisingly useful technique for initial feature selection, and iML techniques improved the selection of short lists of biomarkers for classification.<\/p>\n\n\n\n
Tuesday, December 2<\/h4>\n\n\n\nAKBARBEK RAKHMATULLAEV<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Munehiro Fukuda
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
11:00 A.M.; UW2 (Commons Hall) Room 327
Thesis:<\/strong> IVF Semantical: Agent-Based Implementation of Vector Search on GPU<\/p>\n\n\n\nVector search plays a crucial role in large-scale similarity search applications, with IVF (Inverted File Index) being a widely used indexing method due to its balance between accuracy and efficiency. However, traditional vector search algorithms that used IVF as an indexing method, such as IVF Flat and IVFPQ, yield results by brute force searching within each cluster\/list. This paper introduces a new IVF-based vector search algorithm, called IVF Semantical, which does the search within each cluster\/list through a different arrangement of data and traversal using the Binary Search. In order to accelerate the development phase, the author used MASS CUDA to handle the searching part, which allowed to leverage the abstraction level of the code. We evaluated our IVF Semantical, implemented for GPUs using MASS CUDA, against two other algorithms, IVF Flat and IVFPQ, demonstrating the significant speed efficiency of the approach. The findings suggest IVF Semantical can make vector search more efficient and robust on infrastructure that requires immediate response, such as navigation systems or robots.<\/p>\n\n\n\n
\n\n\n\nSARAH MARTEL<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Annuska Zolyomi
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
3:30 P.M.
Project:<\/strong> Sail the Ship, Steer the Self: Using Social Accountability and Gamification for Goal Attainment in Users with Attention Difficulties<\/p>\n\n\n\nPeople with attention difficulties, such as those with Attention-Deficit\/Hyperactivity Disorder (ADHD), tend to face challenges achieving their goals due to struggling with breaking goals into tasks, procrastination, and planning and remembering important deadlines.<\/p>\n\n\n\n
This project explores a novel approach for motivating people with attention difficulties to set and achieve goals using two motivating factors identified in literature: social accountability and gamification. Social accountability occurs in the form of a body doubling simulator, while a visual journey map of the user\u2019s goals and milestones helps make their progress more meaningful with gamification.<\/p>\n\n\n\n
In the first stage of a user-centered design process, a prototype of the app was designed using the design tool, Figma. Usability testing and interviews with four users who self-identified as having attention difficulties revealed key findings regarding engagement and ease of use. The findings revealed the importance of gamification and novelty in users with attention difficulties, the strength of peer accountability, and the importance of a clear, distraction-free UI. Building on these insights, an Android app, “”Sail the Ship””, was developed. This second stage, where six participants used the app, showed that peer accountability through body doubling increased task initiation and feelings of social pressure for most users. Additionally, the journey map supported engagement through milestone completion and gamification, with participants requesting even more gamified features. This work contributes a novel app design that helps people with attention difficulties venture forward and achieve their goals.<\/p>\n\n\n\n
Wednesday, December 3<\/h4>\n\n\n\nMARINA ROSENWALD<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Michael Stiber
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
8:45 A.M.
Thesis:<\/strong> Application of Graph Neural Networks in Spike-Time-Dependent-Plasticity Graphitti Simulations<\/p>\n\n\n\nSpike-timing-dependent plasticity (STDP) is a core mechanism that drives how biological neural networks’ connectivity changes over time, but understanding the structural changes for large networks is still difficult. Graphitti, the simulator used in the Intelligent Networks Laboratory, can generate large, densely-interconnected STDP networks, yet interpreting what emerges from those simulations is not straightforward. This thesis uses Graph Neural Networks (GNNs) to analyze these networks and to uncover patterns that are hard to see through traditional tools.<\/p>\n\n\n\n
The main goal of this work is to apply different GNN architectures, most notably Graph Attention Networks (GATs), to Graphitti-generated networks in order to identify key structural and behavioral changes. The focus of this thesis is to identify key neurons as STDP unfolds. By treating each simulation as a series of graphs over time, we can apply Graph Neural Networks to explore how connectivity shifts over time, which nodes receive higher attention weights in a Graph Neural Network, and whether these \u201ckey neurons\u201d lead to the expected graph structure. Alongside this analysis, an equally important objective is to improve our understanding and implementation of GNNs for large, fully connected neural graphs. We document the full workflow, challenges, and model choices, and contribute these findings for future work.<\/p>\n\n\n\n
Overall, this thesis links large scale STDP simulation with modern graph learning methods, using attention mechanisms to highlight influential nodes and offering new tools and insights for analyzing complex neural systems.<\/p>\n\n\n\n
\n\n\n\nCOLLEEN LEMAK<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Geethapriya Thamilarasu
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
11:00 A.M.; DISC (Discovery Hall) Room 464
Thesis:<\/strong> Hybrid Static-Dynamic Feature-Weighted Analysis for IoT Botnet Malware Detection<\/p>\n\n\n\nAs the Internet of Things (IoT) domain continues to evolve, IoT devices face escalating security challenges. Recent waves of IoT botnets have exploited device vulnerabilities to launch dangerous large-scale Distributed Denial of Service (DDoS) attacks from compromised, resource-constrained devices. These networks of infected devices pose a unique threat to modern infrastructure, homes, schools, medical facilities, and transportation systems at heightened risk of malicious exploitation.<\/p>\n\n\n\n
This paper proposes a novel hybrid framework that combines static and dynamic analysis techniques for IoT botnet malware detection without relying on complex Machine Learning (ML) models. By extracting and weighing the importance of key features from malware binaries based on their relevance to DDoS behavior, the framework maintains statistical adaptability to observed data while avoiding large memory usage and opaque black-box decision processes common in ML. Designed for interpretability and efficiency, this malware detection framework bridges code-level structure and runtime behavior, offering a transparent and practical botnet detection strategy for diverse resource-constrained IoT ecosystems.<\/p>\n\n\n\n
\n\n\n\nLOGAN CHOI<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Wooyoung Kim
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
1:15 P.M.
Thesis:<\/strong> Benchmarking TenSEAL\u2019s Homomorphic Encryption Through Predicting Encrypted RNA Sequencing Data<\/p>\n\n\n\nThis study addresses the growing need to protect sensitive healthcare data as digital technologies and cloud-based analytics become integral to modern medical research and care delivery. Healthcare data, such as clinical or genomic information, holds immense potential to enhance disease understanding and improve diagnostics through machine learning models; however, adopting third-party cloud technologies increases the risks of data breaches and noncompliance with regulations such as the Health Insurance Portability and Accountability Act (HIPAA). To address these concerns, this research investigates homomorphic encryption, a cryptographic method that allows computations on encrypted data without exposing sensitive information. The study benchmarks the TenSEAL library to evaluate its performance in encrypting healthcare test datasets and executing predictions through a pre-trained machine learning model, while also evaluating memory utilization and encryption time. The findings show that TenSEAL\u2019s CKKS encryption scheme effectively enables data encryption and secure machine learning inference on genomic datasets for breast, lung, and prostate cancers, achieving an average accuracy of 90% across all datasets. On the other hand, our results also highlight a key trade-off: as encryption strength and dataset size increase, computational overhead rises sharply. Thus, medical professionals and data scientists must carefully balance the need for security with the practical deployment in real-world healthcare systems.<\/p>\n\n\n\n
\n\n\n\nKAMERON VUONG<\/h5>\n\n\n\n
Chair<\/strong>: Dr. David Socha
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
3:30 P.M.
Project:<\/strong> Understanding Software Engineering Principles by Evolving Luna mHealth\u2019s Support for PowerPoint<\/p>\n\n\n\nLuna mHealth is a mobile health education platform designed to deliver culturally relevant, offline-accessible health education content to resource-constrained communities. Luna mHealth is composed of two parts: Luna authoring system and Luna mobile. Luna authoring system is a desktop application that converts PowerPoint-based health content into structured, interactive learning modules called Luna content modules. Luna mobile is a mobile application that readers use to view Luna content modules on their mobile phone without the need for an active Internet connection.<\/p>\n\n\n\n
In this project, I evolved Luna mHealth\u2019s support for the core PowerPoint features needed for Luna, such as connection shapes (aka lines), textbox shapes, picture shapes, hyperlinks, and placeholder shapes from slide layout templates. For Luna mobile, I evolved Luna mobile\u2019s rendering logic to render the extracted text and images, ensuring that visuals appear consistently and accurately. I also evolved Luna mobile\u2019s navigation system to support inter-page hyperlinks, allowing readers to navigate to different pages of a Luna content module. I also did enough work to demonstrate that we could extend this system to support LibreOffice Impress presentation files.<\/p>\n\n\n\n
Through practices like Test-Driven Development and principles like Single Responsibility and You Aren\u2019t Gonna Need It, I ensured the code I contributed to the Luna mHealth\u2019s codebase was clean, simple, and maintainable.<\/p>\n\n\n\n
\n\n\n\nWENTAO GAO<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Erika Parsons
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
5:45 P.M.
Project:<\/strong> Virtual Reality and sEMG Integration for Phantom Limb Pain Therapy<\/p>\n\n\n\nPhantom limb pain (PLP) affects a substantial proportion of amputees and remains challenging to treat despite decades of research. Virtual reality (VR) has emerged as a promising non-pharmacological intervention, offering immersive visual feedback and controlled movement experiences. However, existing VR systems face significant barriers to widespread adoption to treat PLP: high costs, limited adaptability to individual differences due to reliance on pre-trained models and fixed control schemes, and restriction to specific upper or lower limb applications. Systematic reviews highlight the need for more affordable, customizable, and versatile rehabilitation technologies.<\/p>\n\n\n\n
This project presents the design and preliminary evaluation of a novel, low-cost VR system capable of supporting both below-elbow and below-knee amputation scenarios through a self-calibrating machine learning pipeline. The system integrates consumer-grade VR hardware (Meta Quest 2), surface electromyography (sEMG) armband (8 channels total), and AprilTag visual markers to provide real-time, personalized limb control in immersive game environments. For limb control, the system classifies three discrete gestures (clench, extension, rest) combined with positional tracking and rotation state, which drive a rigged virtual limb model through animation blending to achieve naturalistic movement suitable for amputee rehabilitation. The system features an automated guided data collection and training workflow that allows users to record EMG signatures and train custom Random Forest gesture recognition models in approximately 5 minutes, without requiring technical expertise or pre-trained models with great accuracy.<\/p>\n\n\n\n
System feasibility is validated through a pilot study with able-bodied participants who complete both upper and lower limb training sessions and interact with purpose-designed VR games. Preliminary results demonstrate successful gesture classification (mean accuracy >85%), acceptable system latency (<50ms end-to-end), and positive user experience metrics. The system addresses a critical gap in accessible rehabilitation technology by providing a single platform adaptable to multiple amputation types while maintaining low implementation costs.<\/p>\n\n\n\n
Thursday, December 4<\/h4>\n\n\n\nYU WEN<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Annuska Zolyomi
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
11:00 A.M.
Project:<\/strong> ConverSwim: A Visual and Emotion-Aware Platform to Support Reflective Conversation for Neurodivergent Users<\/p>\n\n\n\nNeurodivergent individuals\u2014including those with autism and ADHD\u2014often face challenges in conversations, such as interpreting social cues, recognizing emotional signals, and maintaining mutual understanding. These difficulties can hinder effective communication and self-reflection. To address these issues, we developed ConverSwim, a web-based platform that supports users in reflecting on conversational dynamics through an integrated and interactive canvas. This canvas combines real-time chat, NLP-driven emotion detection, and a structured swimlane diagram that chronologically maps each user\u2019s internal emotional state and external verbal expression. Swimlanes within the canvas visualize the conversation across five columns: each participant has two parallel lanes\u2014Inside (representing emotions and internal states) and Outside (representing spoken or written expressions)\u2014while the central Common Ground lane captures moments of shared understanding as they emerge. By unifying interaction, emotional analysis, and visual structure in a single space, ConverSwim is designed to enable users to explore communication patterns and observe emotional shifts as conversations unfold. The platform includes a step-by-step tutorial to guide users through its features and reduce cognitive load. Through iterative, user-centered design, ConverSwim promotes emotional awareness, improves conversational reflection, and facilitates more empathetic, collaborative dialogue. Preliminary user sessions with neurodivergent participants evaluate usability, emotional awareness, and collaborative reflection. Early feedback indicates that participants found the system intuitive to use, helpful for noticing emotional cues, and supportive of reflecting on conversational flow.<\/p>\n\n\n\n
\n\n\n\nBREANNA POWELL<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Afra Mashhadi
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
1:15 P.M.; DISC (Discovery Hall) Room 464
Project:<\/strong> Accelerating Climate Forecasts: Vision Transformer Models for Arctic Sea Ice Concentration Prediction with Novel Tokenization Techniques<\/p>\n\n\n\nReliable forecasting of Arctic Sea Ice Concentration (SIC) is critical for safe shipping, sustainable settlement planning for indigenous communities, and effective wildlife conservation. Yet, accurate SIC prediction remains difficult: observational data are sparse, models often fail at extremes, and climate science depends heavily on computationally expensive physics-based simulations.<\/p>\n\n\n\n
This work introduces IceForecastTransformer, a Vision Transformer\u2013based architecture trained on synthetic data from the U.S. Department of Energy\u2019s Energy Exascale Earth System Model (E3SM) Version 3. By incorporating Freeboard\u2014a satellite-observable metric derived from ice and snow thickness\u2014alongside flexible tokenization and positional encoding strategies, the model provides a possible bridge between simulation outputs and real-world observations. IceForecastTransformer is trained on 163 years of monthly SIC and Freeboard data, achieving 0.16 RMSE and 0.32 JSD for 12-month SIC forecasts, and 0.84 F1 Score with 0.98 AUC for Sea Ice Extent (SIE), all with training times under 15 minutes.<\/p>\n\n\n\n
These results demonstrate that transformer-based models can efficiently scale down complex Earth system data while maintaining high accuracy, offering a practical complement to physics-based simulations. The approach provides a pathway toward operational Arctic forecasting, with potential for fine-tuning on satellite data and broader applications in climate research and planning.<\/p>\n\n\n\n
Friday, December 5<\/h4>\n\n\n\nJONATHAN CHO<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Wooyoung Kim
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
11:00 A.M.
Project:<\/strong> Interpretable Machine Learning Studio Platform Web Application<\/p>\n\n\n\nMachine learning models are often treated as black boxes, creating challenges in domains such as healthcare where interpretability is critical. This project enhanced a Machine Learning Studio Platform (MLSP) originally developed at the University of Washington by improving reliability, expanding supervised and unsupervised model support, and redesigning the interface for usability. Interpretability methods including SHapley Additive exPlanations (SHAP), Local Interpretable Model-Agnostic Explanations (LIME), Partial Dependence Plot (PDP), Permutation Feature Importance (PFI), and Exploratory Data Analysis (EDA) were integrated and tested on benchmark datasets such as the Breast Cancer Wisconsin dataset and others of varying size and complexity. The enhanced system produced consistent metrics and visual explanations, enabling users to evaluate models more effectively and approach results with greater trust and understanding. These contributions demonstrate how interpretability features can make machine learning workflows more transparent and accessible for students, researchers, and healthcare professionals.<\/p>\n\n\n\n
\n\n\n\nXIU ZHOU<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Geethapriya Thamilarasu
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
1:15 P.M.; DISC (Discovery Hall) Room 464
Project:<\/strong> Privacy Information Detection Using Imagined Speech EEG-to-Text Analysis<\/p>\n\n\n\nBrain\u2013Computer Interface (BCI) devices are rapidly entering consumer markets and offer several benefits, such as hands-free interaction and improved accessibility. However, their passive recording capability introduces a severe privacy threat, as malicious actors could intercept neural signals and apply machine learning techniques to infer sensitive information. Among neural modalities, imagined speech\u2014the internal process of thinking words or characters without vocalization\u2014is especially concerning, as it may reveal private content such as passwords or Personal Identification Numbers (PINs).<\/p>\n\n\n\n
To assess this risk, we investigate whether passwords can be decoded purely from imagined-speech EEG without prior knowledge of candidates. Moving beyond prior work limited to classifying among known passwords, we propose a novel hierarchical Binary-Digit-Letter (BDL) system for character-level reconstruction. Using a commercial headset, our method achieves an unprecedented end-to-end accuracy exceeding 60\\% on the full alphanumeric set. This performance markedly reduces the effective entropy of real-world passwords (e.g., from $62^{8}$ to roughly $2^{8}$ candidates for an 8-character password).<\/p>\n\n\n\n
We further present a systematic comparison of two deep neural architectures (CNN+BiLSTM and CNN+Transformer) across four experimental conditions to evaluate their password decoding performance and robustness. The results offer strong evidence that sensitive information can be extracted from passively collected EEG and establish key architectural baselines for future security analyses and privacy-preserving BCI design.<\/p>\n\n\n\n
\n\n\n\nNEELIMA CHOWDHURY<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Annuska Zolyomi
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
3:30 P.M.
Project:<\/strong> HealthMate: Designing a Senior Centered Digital Twin App for Health Monitoring and Predictive Insights<\/p>\n\n\n\nThis project introduces HealthMate, a senior centered mobile health app that uses the concept of a digital twin to present personal health information and future health predictions in a simple format. Current commercially available fitness apps often depict numbers in complex data dashboards designed for mainstream users, not older adults. HealthMate is designed from the beginning to be accessible for older adults and to provide meaningful data visualization and health predictions for health areas that are a priority for them. Due to the popularity of the Apple ecosystem with older adults, HealthMate was chosen to be developed as an iOS app that is compatible with the Apple Watch.<\/p>\n\n\n\n
The key features of HealthMate are to provide seniors with a Digital Twin interface to access their daily health data by tapping on body parts of the Digital Twin or avatar, provide predictive insights based on past week statistics, enable customized notifications for daily health alerts, and provide an intuitive and clear interface with minimal visual clutter and focused views. This is achieved by the app reading user data from wearable devices like Apple Watch, analyzing the past week\u2019s data and statistics, and displaying likely trends or potential issues in areas the user wants to focus on. Currently the data shown reflects a user\u2019s most consistent heart rate, step count consistency, and sleep habit consistency. Future enhancements aim to show anxiety or high stress propensity, fall detection, cardiovascular anomaly, dizziness propensity, and irregular heart rhythm. The predictive insights module analyzes last week\u2019s data and forecasts the current week\u2019s activity, heart rate, and sleep patterns, and may indicate possible incidents such as anxiety, falls, or heart palpitations linked to inadequate sleep. The prework for this module uses generated synthetic datasets that are similar to typical wearable device data. This supports model development and safe investigation of uncommon problems without exposing actual users. The objective is to create a trustworthy and understandable companion that assists senior citizens in identifying changes and helps inspire small improvements to daily routines for better and healthier living.<\/p>\n\n\n\n
SUMMER 2025<\/h3>\n\n\n\nWednesday, June 25<\/h4>\n\n\n\nARJUN TANEJA<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Michael Stiber
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
3:30 P.M.
Project:<\/strong> Spatiotemporal Analysis of Neuronal Avalanches in Self-Organized Criticality<\/p>\n\n\n\nNeural spike activity forms the fundamental basis of information processing and communication in the brain, exhibiting complex spatiotemporal dynamics. Recent advances in computational neuroscience have enabled high-resolution numerical simulations of large-scale neural networks, generating spike-time and location data that capture emergent phenomena such as neuronal avalanches and whole-network bursting. In the case of neuronal avalanches, prior analysis has almost exclusively focused on temporal information, ignoring the spatial information associated with spike data. This project presents an efficient algorithm that incorporates both spatial and temporal constraints for avalanche classification, moving beyond conventional spike-train analyses. Through systematic comparison of temporal-only and spatiotemporal methods, we investigate how the inclusion of spatial information affects avalanche identification and characterization. Finally, we examine how varying spatiotemporal constraints influence the detection and properties of whole-network bursting events, providing insight into the large-scale organization of neural activity.<\/p>\n\n\n\n
Thursday, August 7<\/h4>\n\n\n\nPANKTI BHASKAR SHAH<\/h5>\n\n\n\n
Chair<\/strong>: Professor Mark Kochanski
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
3:30 P.M.
Project:<\/strong> Design and Implementation of a Unified Appointment Scheduling System for Canvas LMS<\/p>\n\n\n\nThis project explores how instructor-student appointment scheduling can be integrated into a university\u2019s Learning Management System (LMS), specifically Canvas, while enabling synchronization with external calendars such as Google and Outlook. Although Canvas centralizes academic workflows, it lacks built-in support for course-aware, role-sensitive scheduling. As a result, instructors rely on disjoint third-party tools that lack integration with course data and often lead to scheduling conflicts and increased administrative burden. This project implements a unified scheduling system that uses OAuth authentication, Canvas API integration and supports external calendar providers for event synchronization. Users can create and manage availability programs, manage appointments, and seamlessly sync events across their preferred calendars. The system was developed using agile practices, offering practical experience in scalable software design, modular architecture, and API integration. This solution addresses a clear gap in academic infrastructure and offers a scalable approach for improved user-experience and efficient scheduling for students, instructors, and administrators.<\/p>\n\n\n\n
Friday, August 8<\/h4>\n\n\n\nAATMAN RAJESHKUMAR PRAJAPATI<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Munehiro Fukuda
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
3:30 P.M.
Project:<\/strong> An Enhancement of Distributed Graph Queries in an Agent-Based Graph Database<\/p>\n\n\n\nGraph databases are increasingly employed in domains that demand efficient data-modelling and querying of highly interconnected data. This project enhances an agent-based graph database system built on the MASS (Multi-Agent Spatial Simulation) Java library, which adopts the property graph model to support dynamic schemas and rich relationship semantics in distributed environments. Unlike traditional graph systems, this implementation leverages autonomous agents to navigate and manipulate in-memory data across multiple computing nodes, enabling scalable and parallel graph operations.<\/p>\n\n\n\n
This work focuses on extending the system’s query capabilities by integrating support for the Cypher WHERE clause, a critical feature for filtering and refining data retrieval. A modular approach was adopted to implement this functionality\u2014beginning with an abstract syntax tree (AST) for parsing, followed by a dynamic evaluation mechanism for efficient constraint resolution. The enhancements improve the expressiveness and performance of read operations, while preserving the system’s core agent-based execution model. This development not only broadens the practical utility of the system but also establishes a foundation for future support of more complex query patterns and mutation operations.<\/p>\n\n\n\n
Monday, August 11<\/h4>\n\n\n\nRAVITEJA TANIKELLA<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Dong Si
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
8:45 A.M.
Thesis:<\/strong> Emotionally Intelligent Voice Language Models For Mental Health Therapy<\/p>\n\n\n\nConversational assistants for mental health therapy primarily rely on text-based models or cascaded architectures that first transcribe speech to text, a process that discards crucial paralinguistic information. This information bottleneck limits the AI’s ability to perceive critical emotional cues and broader psychological states, hindering its capacity to reason over human voice and provide effective therapy. This thesis details the development of an emotionally intelligent voice language model designed to overcome these limitations. The process began with a systematic evaluation of heuristic-based approaches and end-to-end model architectures, where automated benchmarking and a double-blind human study confirmed that large audio language models provided the most effective foundation for voice understanding and reasoning. Building on these findings, we propose and implement policy optimization methods to fine-tune this base voice language model on therapy data. The resulting aligned model demonstrated improved performance in benchmark evaluations, exhibiting emotional intelligence and generating therapeutically relevant responses. By presenting a complete development framework, from architectural validation to targeted alignment, this research establishes a clear and proven roadmap for creating the next generation of efficient and adaptive voice language models for mental health therapy and multimodal conversational AI.<\/p>\n\n\n\n
\n\n\n\nKAUSHIK REDDY MITTA<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Marc Dupuis
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
11:00 A.M.
Project:<\/strong> Demographic Disparities in Cybersecurity: Analyzing Victimization, Tool Adoption, and Awareness Across Ethnicity and Region<\/p>\n\n\n\nCybersecurity disparities across different demographic groups and regions have become a growing concern, as certain populations face disproportionate risks in terms of cyber victimization, tool adoption, and security awareness. These disparities are often overlooked in conventional cybersecurity research, which tends to simplify the relationship between demographic factors and cyber preparedness. This study analyzes these disparities by examining how ethnicity and geographic region influence cybersecurity outcomes in the United States. Using survey data from 470 participants collected by Professor Marc Dupuis, the research employs comprehensive statistical methods including correlation analysis, ANOVA, Kruskal-Wallis tests, and logistic regression to explore patterns across ethnic groups and urban, suburban, and rural settings. The study focuses on four key areas: cyber victimization, cybersecurity tool adoption, cyber incident experiences, and levels of cybersecurity awareness. The findings reveal significant disparities that challenge conventional assumptions about cybersecurity preparedness and risk exposure, highlighting the complex interactions between technology adoption, socioeconomic factors, and the strategies of threat actors targeting specific groups. The study calls for culturally competent interventions to address the unique vulnerability profiles of different demographic groups.<\/p>\n\n\n\n
\n\n\n\nMANPREET KAUR<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Dong Si
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
3:30 P.M.
Project:<\/strong> A Multimodal AI System for Mental Health Support Using Emotion Recognition<\/p>\n\n\n\nEmotion\u2011aware conversational agents have the potential to transform mental health support by providing accessible, empathetic and context\u2011aware interactions. This project enhances the iCare Carebot by expanding it beyond a text\u2011only interface to a multimodal system that integrates text, audio and visual cues to better infer and respond to user emotions. We develop dedicated pipelines for each modality and benchmark both unimodal and multimodal emotion\u2011recognition models, fusing their outputs through cutting\u2011edge foundation models. For visual emotion recognition, we evaluate several vision\u2011language models including Phi\u20114, Qwen2.5\u2011VL\u20117B and Qwen2.5\u2011Omni\u20117B and find that Qwen2.5\u2011VL\u20117B offers the best balance of accuracy and inference efficiency. Across multimodal tasks, instruction\u2011tuned models consistently outperform zero\u2011shot and supervised\u2011fine\u2011tuned baselines. The system\u2019s performance is assessed via the Multimodal Language Analysis benchmark and complemented by a user survey, which indicates broad receptiveness to a multimodal chatbot provided privacy concerns are addressed. Comparative evaluation using the Multimodal Language Analysis benchmark reveals that, among these models, Phi-4-multimodal-instruct provides the most balanced and effective performance across tasks. This work delivers a complete end-to-end framework from emotion signal extraction to context-aware response generation and marks a significant step toward making iCare a more empathetic and emotionally perceptive mental health support tool.<\/p>\n\n\n\n
Tuesday, August 12<\/h4>\n\n\n\nSWANAND WAGH<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Min Chen
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
11:00 A.M.
Project:<\/strong> Designing a User-Centered Digital Experience for the Entomon Institute<\/p>\n\n\n\nIn the digital era, a strong online presence is essential for educational institutions to effectively engage with their audiences. However, typical institute websites for Zoology often suffer from poor accessibility, suboptimal UI\/UX design, excess resources cluttered altogether, and inadequate adaptability, hindering their effectiveness. This project asks: how can I design a digital platform that meaningfully enhances public interaction with zoology & entomology?<\/p>\n\n\n\n
To answer this, I partnered with the Entomon Institute to develop a cutting-edge, user-centric web platform to address these issues. The envisioned website serves as a dynamic resource hub, offering personalized materials on zoology and entomology, facilitating collaboration among institutes and research labs, and managing events such as insect hunts and ant colony explorations. Key features include interactive blogs, streamlined event management & proper user & admin dashboards. By leveraging modern web technologies, the platform aims to enhance educational outreach for zoology & entomology, fostering an engaging community, and provide up-to-date information in an accessible and interactive format.<\/p>\n\n\n\n
At its core, this project is about making an institutional website feel less like a brochure and more like a space people actually want to visit. The result is a fast, accessible, and visually engaging platform that finally brings the Entomon Institute into the modern digital era.<\/p>\n\n\n\n
\n\n\n\nHARI PRIYA DHANASEKARAN<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Michael Stiber
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
1:15 P.M.
Thesis:<\/strong> Understanding Burst Patterns Through Graph Neural Network Explainability in Simulated Neuronal Networks<\/p>\n\n\n\nSpontaneous bursting activity in neural networks represents a fundamental mode of information processing in the brain, yet the mechanisms triggering these synchronized events remain poorly understood. While graph-based representations of neural networks are established, discovering the specific connectivity and activity patterns that predict burst initiation remains a significant challenge. This work leverages graph neural networks (GNNs) to classify and explain burst initiation in simulated cortical networks generated by the Graphitti simulator, advancing beyond conventional spike-train analysis to incorporate the rich relational structure of neural connectivity. By representing neural populations as graphs where nodes encode individual neurons with their temporal firing statistics and edges capture the synaptic connections, this work naturally integrates both activity patterns and network architecture. However, accurate prediction alone provides limited scientific insight. To move beyond black-box classification, we applied GNNExplainer to identify the minimal neural connectivity patterns driving model predictions. This explainability analysis revealed which specific neurons and synaptic connections the model deemed most critical for each prediction. This work demonstrates how explainable AI can transform our understanding of complex neural dynamics, providing insights that pure predictive modeling cannot offer. By combining the representation power of graph neural networks with explainability techniques, we bridge the gap between prediction and understanding. Our findings challenge prevailing views of burst initiation as a localized phenomenon, instead revealing the role of distributed precursor patterns in driving network-wide synchronization. This methodology opens new avenues for investigating emergent behaviors in complex networks.<\/p>\n\n\n\n
\n\n\n\nVICTOR LONG<\/h5>\n\n\n\n
Chair<\/strong>: Dr. David Socha
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
3:30 P.M.
Project:<\/strong> Building Continuous Integration Tools for Luna mHealth<\/p>\n\n\n\nThe Luna mHealth group is a graduate research group building a medical education application for use in low-connectivity areas of the world. Developing good quality code in a larger code base with other developers is a difficult and complex task. Industry professionals turn towards toolsets and processes, like continuous integration, to provide a robust framework for teams to develop quality code. This project implements a continuous integration and continuous delivery toolset to automate a variety of quality gates that help propel Luna developers toward best practices in every part of the development cycle. By using automated testing and linting tools, developers\u2019 code gets checked for testing issues and best-practice deviations before getting merged with the whole project. This project enhanced the student development experience and team consistency by providing a containerized environment. Such an environment allows students to develop on the same version of tools and dependencies. By developing with these continuous integration and continuous delivery tools, students are introduced to the tooling used in many industry settings and experience best practices for building good quality code, and the code quality has risen dramatically.<\/p>\n\n\n\n
Friday, August 15<\/h4>\n\n\n\nRUOXI SU<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Dong Si
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
11:00 A.M.
Project:<\/strong> Emotion-Adaptive Retrieval-Augmented Chatbot: Elevating Mindfulness-Based Mental-Health Conversations with Lightweight LLM Reasoning<\/p>\n\n\n\nLarge-language models (LLMs) have begun to transform mental-health support, yet most systems still ignore the user\u2019s moment-to-moment affect. We present an emotion-adaptive retrieval-augmented generation (RAG) pipeline that grounds counseling responses in a curated mindfulness corpus while dynamically tailoring both retrieval and language style to the client\u2019s emotional state. At each dialogue turn, a few-shot light-weighted LLM classifier assigns one of 13 clinically salient emotions and produces a \u2264 25-word situation summary; this lightweight front-end attains 83\\% accuracy on a balanced 100-scenario test set. The two signals seed a MiniLM-L6 vector index built over 600-token chunks from 10 expert-reviewed articles, and the top-4 passages are merged with the chat history before generation. This design contracts the average context to $\\sim 800$ tokens and trims retrieval latency to $\\approx$ 1800 ms per turn while preserving semantic completeness.<\/p>\n\n\n\n
We benchmark the system against an otherwise identical, emotion-agnostic RAG baseline. On a 100-dialogue test set, the emotion-aware front-end consistently improves quality over an otherwise identical RAG baseline. With GPT-4o-mini, human ratings rise for Empathy (+0.61 on a 0\u20134 scale; 2.35\u21922.96), Topical Relevance (+0.40 on 0\u20133; 1.98\u21922.38), and Helpfulness (+0.19 on 1\u20133; 2.07\u21922.26), while Safety is unchanged (2.70\u21922.73; 0\/100 unsafe). With Llama-3 8B, gains are smaller but consistent\u2014Empathy +0.06 (3.53\u21923.59), Relevance +0.17 (2.39\u21922.56), Helpfulness +0.22 (2.38\u21922.60)\u2014with Safety stable at 2.95. The emotion classifier itself achieves Macro-F1 = 0.81 (AUROC \u2248 0.91) without fine-tuning. Together, these results indicate that a lightweight emotion-classification and situation-summary layer can sharpen topical fit and increase empathic\/helpful tone without compromising safety.<\/p>\n\n\n\n
\n\n\n\nSHRISTI SRIVASTAVA<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Dong Si
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
1:15 P.M.
Thesis:<\/strong> AutismCarebot: An Empathy-Enhanced and Retrieval-Augmented Chatbot for Autistic Users<\/p>\n\n\n\nAutism Spectrum Disorder (ASD) typically requires specialized communication strategies, tailored emotional understanding, and personalized resources – areas commonly overlooked by general mental health chatbots. This thesis introduces AutismCarebot, an empathy-enhanced, retrieval-augmented conversational AI built upon the LLaMA-3.2 (3B) Instruct model, designed explicitly for autistic users. Initially fine-tuned on empathetic dialogue datasets (Empathetic Dialogues, Amod Mental Health Counseling Conversations, Psych8k, ExTES), AutismCarebot was further specialized through fine-tuning on autism-specific data (TASD), enabling nuanced recognition and appropriate responses to autistic communication styles. The chatbot employs keyword-based heuristics to detect emotional cues such as anxiety, sadness, and overwhelm, responding with clear, empathetic validation and tailored reassurance. Retrieval-Augmented Generation (RAG) ensures accuracy by embedding evidence-based advice from credible autism-related resources, transparently cited within responses. Additionally, proactive crisis detection identifies self-harm indicators, immediately connecting users to global crisis support resources. Usability testing and qualitative feedback from autistic individuals, caregivers, and educators demonstrated favorable usability, enhanced empathy, emotional responsiveness, and increased user trust. AutismCarebot provides a replicable, ethical, and effective approach to enhancing emotional support and crisis safety for autistic users, highlighting its practical value in therapeutic, educational, and peer-support environments.<\/p>\n\n\n\n
Tuesday, August 19<\/h4>\n\n\n\nJANYA BYSANI<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Geethapriya Thamilarasu
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
12:00 P.M.
Project:<\/strong> Low-Latency Packet-Processing Pipeline for Anomaly Detection in IoMT Networks Using<\/p>\n\n\n\nThe adoption of Internet-of-Medical-Things (IoMT) devices has turned IoMT networks into
dense, latency sensitive fabrics that relay vast quantities of protected health information. To
safeguard this traffic, this project develops an edge based intrusion prevention system that
combines a Data Plane Development Kit (DPDK) fast path with a machine learning inference
engine. Packets are captured in zero copy, kernel bypass mode and streamed through a
feature extraction stage before being classified by a LightGBM model that delivers sub-500
\u00b5s inference latency.<\/p>\n\n\n\n
This pipeline further introduces a privacy preserving adaptation loop that uses federated
learning. Each gateway retrains the classifier on its local traffic and transmits the model to a
custom aggregator. The aggregator evaluates incoming models and selects the candidate
with the lowest loss as the new global model, which is then redistributed a<\/p>\n\n\n\n
SPRING 2025<\/h3>\n\n\n\nMonday, May 19<\/h4>\n\n\n\nSAVANNA (sAV) WHEELER<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Marc Dupuis
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
1:15 P.M.; UW2 (Commons Hall) 327
Thesis:<\/strong> Investigating The Relationships Between SNS Usage, Personal Information Disclosure, And Cybersecurity<\/p>\n\n\n\nAs social networking sites (SNSs) become ubiquitous for daily activities, users and regulators have raised concerns with the state of digital privacy and security. Cyberattacks on SNSs have exposed private data of millions of users, and cybersecurity threats propagate through social engineering over SNSs. To determine whether frequent SNS users who disclose personal information are at greater risk of cyberattacks, I conducted a hybrid survey-interview study measuring the correlations between personal information disclosure, SNS usage, cybersecurity practices, and past experiences with cybersecurity threats. The survey findings (n = 276) suggest that SNS usage frequency and usage for MNPS (meeting new people and socializing) or MEPO (make, express, or present more popular oneself) purposes have positive correlations to personal information disclosure. SNS usage frequency and personal information disclosure also had positive correlations with experienced cybersecurity threats, but little correlation with cybersecurity behavior. Interview responses highlighted how subjects experienced cybersecurity threats within SNSs.<\/p>\n\n\n\n
Thursday, May 22<\/h4>\n\n\n\nELIAS MARTIN<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Afra Mashhadi
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
1:15 P.M.
Project<\/strong>: AuditAI – An Interactive Platform to Audit Question-Answer (QA) Pairs from AI and LLM Agents<\/p>\n\n\n\nAuditAI is an interactive platform to help users audit a question-answer pair that they received from conversations with an artificial intelligence (AI) agent.<\/p>\n\n\n\n
This tool is especially relevant to current happenings in the world, as large language models (LLMs) and other AI agents become more prevalent and used in many contexts globally. In my capstone, I built out a framework for users to be able to gain insight into whether their question-answer pair is accurate and appropriate, taking into account machine learning (ML) models combined with other sources of content from the internet as a ground truth (using Google Search to fact check answer).<\/p>\n\n\n\n
The goal of AuditAI is not to definitively say whether a question-answer pair being audited is correct or incorrect, but to give the user unprecedented access to information at their fingertips to come to their own conclusion about the information they are trying to audit.<\/p>\n\n\n\n
\n\n\n\nPRATHAMESH PRAKASH BHALANGE<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Min Chen
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
3:30 P.M.
Project<\/strong>: Peer-to-Peer Experience Synchronization System: Enhancing Communication and Control through Chat, Video Streaming, and Admin Features<\/p>\n\n\n\nAs the demand for real-time collaboration and media sharing tools has increased in the past decade, the development of real-time innovative communication platforms capable of delivering a seamless communication experience has risen rapidly. Traditional systems rely heavily on centralized infrastructures to maintain coordination across distributed clients. While centralized architecture is effective for smaller-scale deployments, it poses significant challenges in scalability, performance overhead, higher operational costs, and infrastructure complexity.<\/p>\n\n\n\n
This project presents a Peer-to-Peer Experience Synchronization System, implemented as a browser extension, that leverages decentralized communication using WebRTC to enable scalable, real-time synchronization without depending on persistent server connections. This project introduces session creation and connection establishment through a lightweight signaling server, which minimizes asset exchange cycles and reduces the server communication burden.<\/p>\n\n\n\n
Beyond synchronization, the system integrates a robust communication framework designed to improve user interaction and control. Core features include real-time communication chat functionality, a two-person video conferencing enabling face-to-face video with synchronized audio collaboration, and admin access control mechanisms embedded within the browser extension UI.
By extending the communication mechanisms leveraging WebRTC and unifying these features within a single browser extension, this project demonstrates a holistic approach to decentralized streaming and communication. The system not only reduces infrastructure dependency but also enriches user experience through integrated tools for interaction and control.<\/p>\n\n\n\n
\n\n\n\nVIVEKANANDA REDDY LENKALA<\/h5>\n\n\n\n
Chair<\/strong>: Dr. William Erdly
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
5:00 P.M.
Project<\/strong>: BuzzMagnet: A Bidding Ecosystem for Democratizing Access Between Social Media Promoters and Marketing Opportunities<\/p>\n\n\n\nIn today’s digital landscape, social media usage has grown exponentially, creating opportunities for influencers to reach massive audiences through innovative approaches. However, many influencers struggle to establish sustainable partnerships with businesses that would allow them to monetize their content creation efforts. Simultaneously, businesses face challenges identifying appropriate influencers whose audience and style align with their brand values and marketing objectives. BuzzMagnet addresses this market inefficiency by creating a transparent bidding marketplace where businesses post advertising opportunities and influencers can bid to secure these campaigns. This reverse-auction approach empowers businesses to evaluate multiple potential influencers based on their proposals, while giving influencers of all sizes equal opportunity to showcase their value beyond mere follower counts. The platform provides comprehensive analytics to business owners, enabling data-driven decisions when selecting influencers based on relevant metrics such as engagement rates, and previous campaign performance. By democratizing access to advertising opportunities, BuzzMagnet particularly benefits emerging influencers who might otherwise be overlooked in traditional influencer marketing approaches. For businesses, the platform offers the advantage of discovering high-potential content creators at earlier stages of their careers, potentially securing more authentic promotion at competitive rates. Through this innovative bidding system, BuzzMagnet aims to transform the influencer marketing ecosystem by facilitating more transparent, merit-based connections between businesses seeking promotion and the diverse community of digital content creators.<\/p>\n\n\n\n
Friday, May 23<\/h4>\n\n\n\nKEVIN JOSEPH GRAFF<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Wooyoung Kim
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
11:00 A.M.
Project<\/strong>: Dual-Modality Alzheimer\u2019s Detection with Interpretable Models<\/p>\n\n\n\nEarly diagnosis of Alzheimer\u2019s disease (AD) is critical for enabling timely intervention and improving patient outcomes. However, traditional diagnostic methods often face limitations in accessibility, accuracy, efficiency, and specifically interpretability. This capstone project explores the application of both Machine Learning (ML) models and Deep Learning (DL) techniques by addressing the limitations but focusing on interpretability for detecting Alzheimer\u2019s disease using demographic tabular data and 3D structural brain MRI scans from the OASIS-2 (Open Access Series of Imaging Studies) dataset. It investigates classical iML models including Decision Trees, Random Forests, and Logistic Regression, applied on demographic tabular data, with an emphasis on interpretability and transparency in addition to predictive performance. Evaluation metrics such as accuracy, precision, recall, and F1-score are used to assess model quality. To further interpret model behavior, global and local agnostic explanation techniques are applied, including SHAP (SHapley Additive exPlanations), Permutation Feature Importance (PFI), and Partial Dependence Plots (PDPs). In parallel, a 3D Convolutional Neural Network (CNN), as one of DL models, is trained to classify patients as either \u201cDemented\u201d or \u201cNon-Demented\u201d based on medical imaging of entire brain volumes. Explainability techniques such as Grad-CAM and saliency maps are employed to generate spatial visualizations that highlight key regions of the brain contributing to each prediction. These visual explanations help bridge the gap between model output and clinical insight, potentially increasing trust in ML-assisted diagnostics. This dual approach provides both structured insights from tabular data and spatial insights from medical imaging, illustrating how iML and DL can complement one another in the context of Alzheimer\u2019s detection.<\/p>\n\n\n\n
\n\n\n\nWonWhoo (Andrew) nah<\/h5>\n\n\n\n
Chair<\/strong>: Dr. David Socha
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
3:30 P.M.
Project<\/strong>: Understanding Software Engineering Principles: Developing a PowerPoint Builder Framework for Luna mHealth\u2019s Authoring System<\/p>\n\n\n\nLuna mHealth has evolved over time, but its architecture required refactoring to improve extensibility and maintainability. This project explores how software engineering principles can transform PowerPoint based educational content into a structured format in the Luna authoring system. At its core, PptxTreeBuilder is a modular framework built upon test-driven development that extracts and organizes essential elements from .pptx files into Luna content modules, ensuring efficiency, readability, maintainability, and scalability. Through this project, I learned how to build a clean architecture, and use lean development principles. Beyond its technical contributions, this system establishes a sustainable, extensible authoring platform, empowering future developers to refine and expand its capabilities.<\/p>\n\n\n\n
Wednesday, May 28<\/h4>\n\n\n\nPRAGATI AMOL DODE<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Erika Parsons
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
11:00 A.M.; UW1 (Founders Hall) 370
Thesis<\/strong>: Exploring Classification Methods for Motor Imagery and Execution EEG Signal Fluctuations<\/p>\n\n\n\nBrain-Computer Interfaces (BCIs) offer promising applications in neurological rehabilitation through motor imagery (MI)-based training, which is the “intent” of performing an action. This research addresses the challenge of accurately classifying MI and motor execution (ME) based on Electroencephalography (EEG) signals. This kind of data is often limited by subject variability, non-stationarity, environmental noise during data collection, EEG device quality, and small dataset sizes. For our study, we propose to make use of an external large dataset, including data from 103 subjects (compared to the 9\u201312 subject datasets used in prior work). One of the main goals of this research is to integrate multiple feature extraction techniques spanning time, frequency, and spatial domains. Effective EEG channel selection was guided by fMRI studies identifying MI- and ME-relevant Brodmann areas, combined with EEG-based statistical analysis, resulting in a refined set of 12 informative electrodes. Several machine learning models (SVM, RF, KNN, XGBoost, MLP) are evaluated, achieving up to 80\\% accuracy with improved robustness across subjects. These findings demonstrate enhanced generalizability and support the development of more reliable BCI applications for real-world rehabilitation scenarios.<\/p>\n\n\n\n
\n\n\n\nYUZE WANG<\/h5>\n\n\n\n
Chair<\/strong>: Dr. William Erdly
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
1:15 P.M.
Project<\/strong>: The QuickCheck Vision Screening Application: Multi-platform Mobile Development (iOS and Android) and Preparation for Clinical Trials<\/p>\n\n\n\nAccording to the World Health Organization, roughly 2.2 billion people worldwide have some visual impairment. The key challenge in treating many forms of visual impairment is detection. Visual impairment is often gradual, meaning that people with it may not notice the changes in their perceptions. Also, children who do not initially develop normal visual abilities can lack a baseline of \u201cgood\u201d vision to compare their experiences against. And because of that, they may not realize they have vision problems. These impairments can affect individuals\u2019 independence, educational achievements, and mobility, and may also lead to physical injuries and mental health challenges. Traditionally, these vision conditions are expected to be detected in school eye exams. However, eye exams only take place a few times throughout the school year.
QuickCheck is a quick and easy-to-use vision screening tool for mobile devices that can perform vision tests to determine if a patient is likely to have any specific visual conditions as the first step toward diagnosis and treatment.
Throughout the project, functions were restored in QuickCheck. By adding functions for generating and sending result reports, QuickCheck became a standalone application. Also, support for IOS devices was added to QuickCheck during this project, bringing it to a wider user group.<\/p>\n\n\n\n
Keywords: Vision Screening, Mobile Application Development, Unity, Xcode<\/p>\n\n\n\n
\n\n\n\nMICHAEL LEE<\/h5>\n\n\n\n
Chair<\/strong>: Dr. David Socha
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
3:30 P.M.
Project<\/strong>: Learning Software Engineering Principles: Developing a Navigation System for Luna mHealth<\/p>\n\n\n\nLuna mHealth is a mobile health education platform designed to deliver culturally relevant, offline-accessible content to underserved and low-connectivity communities. Luna transforms PowerPoint-based health materials into structured, interactive learning modules. The system supports both guided and exploratory navigation through sequences of content, using a lightweight mobile interface optimized for usability in resource-constrained settings.<\/p>\n\n\n\n
This project focuses specifically on the design and implementation of Luna\u2019s navigation system. Navigation is a core component of the user experience, as it determines how users interact with and progress through health education materials. Foundational software engineering principles were applied throughout the development process, including test-driven development (TDD), the SOLID principles, and the KISS (Keep It Simple, Stupid) and YAGNI (You Aren\u2019t Gonna Need It) philosophies. These practices helped ensure a clean, maintainable architecture focused on simplicity, scalability, and long-term extensibility. The goal was to design a streamlined, intuitive application that enhances the user experience while providing a strong architectural foundation for future development.<\/p>\n\n\n\n
\n\n\n\nJASWANTH SRIVAN LAGADAPATI<\/h5>\n\n\n\n
Chair<\/strong>: Dr. David Socha
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
5:45 P.M.; Join Jaswanth Srivan Lagadapati’s Online Defense<\/a>
Project<\/strong>: Understanding Software Engineering Principles: Building and Evolving the Architecture of the Luna mHealth Project<\/p>\n\n\n\nLuna mHealth is a mobile health education platform designed to work offline in underserved communities. As the project grew, it became clear that its architecture needed to evolve. What began as a fast-moving prototype required a more deliberate structure, one that future contributors could understand, maintain, and build upon. In this project, I applied core software engineering principles to rework key areas of the Luna system: the authoring pipeline, the domain model, and the mobile rendering layer. I redesigned the Luna domain model to introduce clearer abstractions like LineComponent, Dimension, and SequenceOfPages, reflecting how real content is structured and consumed. I replaced a monolithic module generator with a modular builder-based framework, allowing content to be constructed in clean, testable steps. On the mobile side, I implemented logic to render line components directly from PowerPoint data, ensuring that visuals appeared consistently and accurately. Through practices like test-driven development, I gained firsthand experience with principles like Single Responsibility and Open\/Closed. More than just improving code, this work laid the foundation for a maintainable and extensible system, one that can support Luna\u2019s mission for years to come.<\/p>\n\n\n\n
Thursday, May 29<\/h4>\n\n\n\nSTEPHANIE ANNE MURRAY<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Erika Parsons
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
11:00 A.M.
Thesis<\/strong>: Exploring Quantum Machine Learning-Enhanced Models for EEG Data Classification<\/p>\n\n\n\nElectroencephalography (EEG) records brain activity linked to both executed and imagined movements, but separating real hand movement from the surrounding noise in high-dimensional EEG remains difficult. Reliable classifiers are therefore vital for accurately tracking patient progress over time. This work is part of a larger initiative, dubbed Smart NeuroRehab Ecosystem, which has two main goals: 1) to propose innovative physical-rehabilitation strategies for certain neurologic conditions (e.g., stroke) using cutting-edge technologies, making therapy more accessible, and 2) collecting EEG data for analysis and building machine-learning (ML) models to classify brain signals.<\/p>\n\n\n\n
EEG data are complex and difficult to analyze and classify. In this research we explore a quantum-machine-learning (QML) strategy for that purpose. Compared with traditional ML approaches, qubits and quantum circuits offer a different way to represent and process features, which could lead to more efficient classification.<\/p>\n\n\n\n
We implement and analyze a ten-qubit Variational Quantum Classifier (VQC) and compare its performance against a \u201ctraditional\u201d classification method\u2014a tuned Random Forest baseline. An existing publicly available data set is used for the experiments, which involve discriminating between specific motor tasks based on signals collected from a 64-channel EEG device.<\/p>\n\n\n\n
Across 40 preliminary runs, the VQC achieves macro-F1 \u2248 0.75, accuracy \u2248 0.76, and AUROC \u2248 0.83, outperforming the Random Forest (macro-F1 \u2248 0.71, AUROC \u2248 0.79). The majority of experiments were performed on a quantum simulator, with a subset of jobs submitted to a cloud-based quantum computer for testing.<\/p>\n\n\n\n
These findings show that hybrid quantum-classical models can match and occasionally surpass strong classical pipelines without expanding the computational footprint, offering a practical path toward longitudinal EEG monitoring within the scope of the Smart NeuroRehab project. Moreover, because quantum computing is still emerging, this work may open avenues for new advancements in the field.<\/p>\n\n\n\n
\n\n\n\nMANU HEGDE<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Erika Parsons
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
1:15 P.M.
Project<\/strong>: Project TLDR: Standalone desktop application for question answering and summarization using resource-efficient LLMs<\/p>\n\n\n\nThis project presents the design and development of a standalone desktop application for offline question answering and summarization over a user-provided document corpus, using resource-efficient large language models (LLMs). Targeted for Apple\u2019s M1\/M2 hardware, the application leverages on-device computation via the Apple Neural Engine (ANE) and Metal shaders, exploring the use of the NPU beyond traditional CoreML applications.<\/p>\n\n\n\n
The application addresses key concerns around data privacy, resource efficiency, and accessibility. Unlike cloud-based services that require constant internet access and raise privacy risks, this application offers a secure, local alternative optimized for researchers and students. It features a graphical interface and supports retrieval-augmented generation (RAG) over the user’s corpus, all while utilizing only a fraction of system resources to support seamless multitasking.<\/p>\n\n\n\n
Evaluation is conducted using both functional metrics (e.g., BERTScore against ChatGPT outputs) and non-functional metrics (e.g., memory and CPU usage). The result is a practical, efficient application that enables interaction with large academic corpora while preserving system responsiveness and data confidentiality.<\/p>\n\n\n\n
Friday, May 30<\/h4>\n\n\n\nVEDANG PARASNIS<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Geethapriya Thamilarasu
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
11:00 A.M.; Discovery Hall 464
Project<\/strong>: Linux Kernel Integrated Security Framework for Real-Time Prevention of DNS Data Exfiltration in Distributed Environments<\/p>\n\n\n\nData exfiltration remains one of the most persistent and sophisticated threats in cybersecurity, with DNS frequently exploited as a covert channel for tunneling and command-and-control (C2) operations. Advanced persistent threat (APT) malware leverages DNS not only for stealthy data exfiltration but also for remote exploitation across multiple compromised nodes using techniques such as remote code execution, port forwarding, reverse tunneling, and process side-channel attacks. These threats are especially critical for hyperscalers managing large-scale server data planes running on Linux, where data sovereignty and integrity are paramount. DNS\u2019s ubiquity, inherent security flaws, and mandated unblocked availability on firewalls make it a high-risk vector for stealthy breaches and C2 attacks\u2014particularly difficult to prevent in real time across distributed environments.<\/p>\n\n\n\n
Existing solutions for DNS exfiltration security primarily rely on passive traffic analysis from centralized locations, using anomaly detection based on statistical heuristics and machine learning models trained on DNS traffic anomalies. Others depend on domain reputation scoring and static blacklists\u2014all operating in userspace. However, these approaches are inherently reactive, slow to respond, increase latency, lack security enforcement at the endpoint, and are often ineffective against stealthy, adaptive C2 implants that utilize domain generation algorithms (DGAs) to obfuscate attacker infrastructure. As a result, they offer no guarantees of preventing data loss before detection and are inefficient at combating C2 attacks across distributed environments. By the time an alert is triggered, malicious commands may have already been executed and significant damage inflicted\u2014especially in the case of advanced C2 botnet vectors.<\/p>\n\n\n\n
This project develops a novel, scalable framework for real-time prevention of DNS-based data exfiltration across distributed environments. It uses an agent-based, endpoint-centric approach, embedding security code throughout the Linux kernel network stack, the mandatory access control layer, and other core kernel subsystems. The framework performs deep packet inspection for advanced DNS payload analysis by parsing the DNS protocol inside the Linux kernel via eBPF, supported by real-time inference from a deep learning model trained on diverse data obfuscation techniques to detect malicious exfiltrated DNS payloads. Malicious C2 implants are terminated at the source on the endpoint, with detailed observability metrics exported to characterize implant behavior\u2014minimizing dwell time, accelerating incident response, and containing threats immediately. Additionally, the framework supports event streaming, dynamic domain blacklisting on DNS servers, in-kernel network policy enforcement, and cross-endpoint security policies for scalability. Experimental results demonstrate horizontal scalability, deep learning model accuracy near 99%, and sub-second response and containment of sophisticated DNS C2 attacks\u2014all with negligible or no data loss. This approach enhances real-time visibility, accelerates threat containment, and fortifies endpoint defense against emerging threats across distributed environments.<\/p>\n\n\n\n
\n\n\n\nBRENDA SUGEY VEGA CONTRERAS<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Geethapriya Thamilarasu
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
1:15 P.M.; Discovery Hall 464
Project<\/strong>: SAFE-MD: A Secure Architecture Framework for Enhanced Mobile Development<\/p>\n\n\n\nThe use of mobile applications has increased over the years since the introduction of smartphones and the creation of centralized repositories like Google Play and App Store, where mobile health or mHealth apps have gained traction over. Their user base is expected to continue to grow, making them a target for attackers as they can carry over critical tasks and contain sensitive patient information. Different vulnerabilities have been identified, with multiple contributing factors. These often arise from a combination of insufficient knowledge of security best practices, limited expertise in implementation, and practical challenges, such as the need to meet development deadlines and manage financial limitations.<\/p>\n\n\n\n
Research has proposed conceptual and prototypic frameworks to mitigate these security risks. However, most implementations focus exclusively on a single platform, primarily Android. Therefore, it is crucial to develop solutions that enable mHealth app developers to reduce the likelihood of user data breaches across multiple mobile platforms.<\/p>\n\n\n\n
This project presents SAFE-MD, a comprehensive security framework with a platform-agnostic approach that facilitates the integration of security mechanisms into mHealth applications at any time during the app development lifecycle. To ensure multiplatform compatibility, the security framework is built using Kotlin Multiplatform, with an initial focus on Android and iOS devices. Different platform-agnostic and platform-specific security mechanisms were integrated to protect users, such as password strength validation and biometric authentication.<\/p>\n\n\n\n
To validate the effectiveness and portability of the security framework across different mobile development environments, two mock mHealth apps were built using Kotlin Multiplatform and Flutter, running on two of the most attractive platforms for attackers, Android and iOS. Preliminary findings during the implementation phase indicate that the incorporation of the platform-agnostic security mechanisms of the SAFE-MD in Android and iOS mock apps for both Kotlin Multiplatform and Flutter, can achieve smooth integration across all the instances of the mock app. However, the availability of platform-specific security mechanisms in the security framework may vary across platforms due to architectural or implementation differences.<\/p>\n\n\n\n
SAFE-MD simplifies the integration of security mechanisms in mHealth apps for multiple platforms and lays the foundation for future enhancements, including broader platform support, the improvement of current security mechanisms, and the integration of new ones to protect users.<\/p>\n\n\n\n
Monday, June 2<\/h4>\n\n\n\nATUL AHIRE<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Munehiro Fukuda
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
8:45 A.M.
Project:<\/strong> MASS Java Library Towards it\u2019s Use for a Graph Database System<\/p>\n\n\n\nDistributed graph databases are foundational to modern applications such as social networks, recommendation systems, and transportation platforms. The MASS (Multi-Agent Spatial Simulation) framework\u2014an agent-based parallel computing library for distributed systems\u2014designed to simulate spatial computations by deploying mobile agents over distributed data structures such as graphs. MASS\u2019s agent-based graph computing model enables intuitive programming and parallel traversal, making it well-suited for scalable graph database applications. This research intends to address two core limitations in its graph database model: inefficient vertex placement and high communication overhead during agent migration. To improve data locality and reduce inter-node communication during graph traversal, we explored and integrated graph partitioning strategy which minimizes edge cuts and better aligns with graph topology. This approach significantly reduced inter node communication for graph benchmarks like Triangle Counting and Articulation Points by localizing vertex adjacency across fewer computing nodes.<\/p>\n\n\n\n
The second challenge targeted the messaging inefficiencies caused by TCP-based agent migration. We redesigned the MASS communication infrastructure using Aeron, a high-performance UDP messaging library, to support low-latency, high-throughput data exchange. This refactoring introduced a dual communication pattern: a structured master-worker protocol for control synchronization and a peer-to-peer mesh for direct agent exchanges. By combining optimized graph partitioning with Aeron-based messaging, the enhanced MASS system reduced the communication latency and increased agent migration efficiency across large, distributed graphs.<\/p>\n\n\n\n
\n\n\n\nVANESSA ARNDORFER<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Michael Stiber
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
11:00 A.M.
Thesis<\/strong>: Network Behavior Analysis of Spike Timing Dependent Plasticity (STDP) in Simulated Neural Networks<\/p>\n\n\n\nThe machine learning landscape is rapidly evolving with researchers often turning toward
nature for inspiration. Understanding the development of neural networks in vivo contributes
significant transferrable insight for advancing both neuroscience and computational research.
This project applies a multiplicative Spike Timing Dependent Plasticity (STDP) model to
the weighted graph output from neural growth simulations and analyzes the resulting spike
and weight changes over time. This preliminary investigation establishes a baseline process
for understanding the effects of STDP on a neural network and provides a framework for
defining the resulting network behavior. Through rigorous data analysis, we qualify bursting
behavior during the refinement phase, analyze the progressive effects of STDP on synapse
weights, and compare how the network behavior changes between the growth and refinement
phases of neural development.<\/p>\n\n\n\n
Tuesday, June 3<\/h4>\n\n\n\nAVIKANT WADHWA<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Michael Stiber
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
8:45 A.M.
Thesis<\/strong>: Abstractions for Code Migration from CPU to GPU in Simulation Domain<\/p>\n\n\n\nLarge-scale simulations are crucial in science, enabling the modeling of complex phenomena that are difficult to study empirically. As they scale, they demand greater performance and efficiency. To meet this need, computing has shifted toward heterogeneous architectures that combine CPUs and GPUs. While effective, this shift introduces software engineering challenges, making abstraction an increasingly important solution. Abstractions hide low-level implementation details behind clean interfaces, improving clarity and reducing complexity.<\/p>\n\n\n\n
This thesis reviews existing abstractions, analyzing their integration effort, performance trade-offs, and limitations. It presents the design and implementation of DeviceVector, a lightweight abstraction that unifies host and device memory management in Graphitti, a high-performance graph-based simulation platform. DeviceVector reduces code duplication, introduces CPU-GPU data relationship, and abstracts CUDA boilerplate through an interface that closely mirrors a standard C++ container. The thesis also discusses design approaches for extending support in the future to object hierarchies and general function-level abstractions, further minimizing logic duplication between host and device code. Overall, this work highlights how thoughtful abstraction design can bridge the usability-performance gap in heterogeneous computing systems.<\/p>\n\n\n\n
\n\n\n\nWILLIAM D. SELKE<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Wooyoung Kim
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
11:00 A.M.
Thesis<\/strong>: Pairwise Normalization Reveals Predictive Immunosignatures in CAR-T Cell Therapy<\/p>\n\n\n\nImmunoprecipitation (IP) assays are widely used to quantify protein-protein interactions (PPIs), yet their susceptibility to variability in protein concentration and experimental conditions often obscures meaningful biological signals. This variability poses a significant challenge for downstream analyses, particularly machine learning (ML) models that require well-separated feature spaces. To address this, we introduce Pair Vector Centralization (PVC), a normalization strategy that eliminates concentration-dependent noise while preserving biologically relevant variation. PVC operates by pairing each altered-state sample (e.g., Cytokine Release Syndrome, Neurotoxicity) with a corresponding baseline (Optimal Response) under similar experimental conditions, then transforming these pairs into vectors of change. A midpoint centering step further removes global shifts unrelated to the underlying phenotype.<\/p>\n\n\n\n
We applied PVC to a high-dimensional, high-variability dataset comprising 121 PPIs from 98 CAR-T therapy patients across 27 independent experiments. Compared to standard normalization techniques. including Z-score, quantile normalization, and PCA; PVC substantially improved both clustering quality and predictive performance. Notably, the F1 score of a baseline Random Forest classifier improved from 75.71% to 95.24% post-normalization. Heatmap comparisons of multiple normalization methods and ML classifiers further demonstrate PVC\u2019s consistent superiority across classification and unsupervised clustering metrics.<\/p>\n\n\n\n
These results underscore PVC\u2019s potential as a robust preprocessing tool for biological datasets affected by batch effects and concentration variability. Its applicability extends to other domains involving biological state transitions, including drug perturbation studies and disease progression modeling. Future work will explore its integration with broader computational pipelines and validation across additional high-throughput datasets.<\/p>\n\n\n\n
\n\n\n\nNEERAJ BAIPUREDDY<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Arkady Retik
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
3:30 P.M.
Project<\/strong>: Smart Umpire: AI-Powered Decision Support and Analytics for Grassroots Cricket<\/p>\n\n\n\nWhile professional cricket benefits from advanced technologies like DRS, local and amateur-level matches often lack reliable tools for decision-making. The few existing solutions for grassroots cricket are typically limited to LBW detection and do not support other important decisions such as run-outs and no-balls. They are also costly and not accessible to most local teams.<\/p>\n\n\n\n
Smart Umpire aims to provide a more complete and affordable alternative by using computer vision and machine learning to bring real-time decision support to local matches. The system analyzes video footage to detect LBWs, run-outs, and no-balls, as well as delivery patterns such as bounce location, deviation after pitching, and length classification. All models are developed in Python using custom datasets trained and tested on the Roboflow platform.<\/p>\n\n\n\n
This project establishes the foundation for a mobile application that will offer decision support, player analytics, and a subscription-based access model\u2014making professional-level insights available to the everyday game.<\/p>\n\n\n\n
Wednesday, June 4<\/h4>\n\n\n\nPHAT TIEN TRAN<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Wooyoung Kim
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
11:00 A.M.
Thesis<\/strong>: StackBERT-Enhancer: A Dual-Layer BERT-Based Framework for Enhancer Identification and Strength Classification in Genomic Data<\/p>\n\n\n\nEnhancers are crucial regulatory DNA sequences that control gene expression and play essential roles in cellular differentiation, development, and disease. Despite their significance, accurately identifying and classifying enhancers remains challenging due to their context-dependent activity, complex sequence features, and lack of consistent structural patterns. Traditional computational approaches, such as sequence heuristics and shallow machine learning models, often fail to capture the long-range dependencies and multi-scale contextual information embedded in genomic sequences. Moreover, they typically lack interpretability, making it difficult to extract relevant biological understanding.<\/p>\n\n\n\n
This thesis introduces StackBERT-Enhancer, a novel deep learning framework designed to overcome these limitations. It addresses two core tasks: distinguishing enhancer sequences from non-enhancer sequences and classifying identified enhancers based on their functional strength. The framework leverages multiple transformer-based language models, each trained independently on DNA sequences tokenized with different k-mer sizes. This multi-k-mer strategy enables the capture of sequence dependencies and context across various scales. These individual models then serve as base learners within a stacking ensemble architecture. This ensemble approach significantly enhances classification accuracy, robustness, and generalization performance across both tasks, achieving state-of-the-art results.<\/p>\n\n\n\n
To handle the computational demands of large genomic datasets, the framework employs distributed multi-GPU systems for efficient model training and hyperparameter optimization. In addition, to bridge the gap between model predictions and biological insight, interpretability techniques are incorporated. SHapley Additive exPlanations (SHAP) are used to evaluate feature importance, while attention score analysis supports sequence motif discovery.<\/p>\n\n\n\n
Overall, StackBERT-Enhancer combines advanced machine learning with biological insight to create a robust and interpretable framework for enhancer identification and strength classification. By uncovering complex sequence patterns, it holds strong potential for applications in disease modeling and broader biomedical research.<\/p>\n\n\n\n
\n\n\n\nHOLT OGDEN<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Munehiro Fukuda
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
1:15 P.M.
Project<\/strong>: Multi-GPU Parallelized Agent Based Modeling<\/p>\n\n\n\nAgent based modeling (ABM) is a method for simulating emergent behaviors in complex systems by modeling the interactions of individual agents. These simulations often require substantial computational resources, necessitating increased parallelization and spatial scalability to produce usable results. One approach is to use a computer\u2019s Graphics Processing Unit (GPU) to allow increased parallelization by utilizing the greater number of threads available on the GPU compared to the Central Processing Unit (CPU). The Multi-Agent Spatial Simulation (MASS) library for NVIDIA\u2019s Compute Unified Device Architecture (CUDA) provides a platform that allows users to write ABM programs to run on the GPU. However, these programs are limited in their spatial scalability by the memory available on a single graphics card. In this project, we improved the MASS CUDA library\u2019s Place object implementation by extending it to function over multiple GPUs connected via NVIDIA NVLink, increasing the potential simulation size of MASS CUDA programs. This required splitting ABM data between multiple graphics cards, ensuring memory synchronization between the cards, and recombining result data at the end of the simulation. These changes improved the runtime of ABM simulations by 35% and increased the maximum simulation size by 75%. In addition, these changes were designed to be abstracted from the user so that minimal changes are required by ABM programs written for previous versions of MASS CUDA. Overall, this project significantly expands the computational resources available to the MASS CUDA library, allowing the running of larger and more complex ABM programs.<\/p>\n\n\n\n
\n\n\n\nAZMEEN KAUSAR MOHAMMAD<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Dong Si
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
3:30 P.M.
Thesis<\/strong>: Enhancing Coach Bot Features through Adaptive Learning using NLP and Fine Tuning<\/p>\n\n\n\nArtificial Intelligence (AI) is revolutionizing mental health care by providing innovative, scalable, and adaptive training solutions for caregivers. Caregivers often face challenges in adapting to diverse mental health scenarios and require tools that support personalized and effective training. To address this, Coach Bot was developed to train caregivers in empathetic communication through scenario-based learning. However, to meet the diverse needs of caregivers and make learning more effective, it is essential for the chatbot to adapt dynamically to individual user preferences and provide real-time, actionable feedback. This project focuses on developing an advanced AI-powered Coach Bot that incorporates adaptive learning paths and personalized feedback mechanisms. A detailed review of existing literature on conversational agents and applications of NLP in mental health was conducted, along with an analysis of data classification to difficulty levels and caregiver response evaluation. The overarching goal of this work is to enhance the design and functionality of adaptive chatbots to improve caregiver training and ultimately support better mental health outcomes and include detailed feedback to the user along with the EPITOME framework scores.<\/p>\n\n\n\n
\n\n\n\nAMALAYE OYAKE<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Erika Parsons
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
5:45 P.M.; Discovery Hall 464
Thesis<\/strong>: Resilient State Convergence in Interplanetary and Edge Networks Using Conflict-Free Replicated Data Types (CRDTs)<\/p>\n\n\n\nEmbedded devices operating in harsh environments, such as outer space or remote locations, present significant challenges for system design, mainly due to their functionality under conditions that disrupt traditional assumptions. Advancements in computing power and the miniaturization of modern electronics have enabled the implementation of distributed computing systems in isolated terrestrial locations as well as in space. The most extreme environments, including polar regions, outer space, and places characterized by resource limitations such as power or network connectivity, stand to benefit from these advancements in networking, computing, and Artificial Intelligence (AI).<\/p>\n\n\n\n
Space presents particular challenges, including high communication latency, intermittent connectivity, obscured signals, and bodies in motion. Deploying enterprise software or conventional IT approaches in these extreme conditions is ineffective. Conventional enterprise software solutions adhere to design principles that are tested to their limits, often leading to fragile and unreliable systems.<\/p>\n\n\n\n
Disruption-tolerant networking (DTN) is a relatively new set of Internet standards designed to address the challenges networked systems face in highly remote locations. It enables communication despite delays, disruptions, and disconnections. One limitation of the DTN standards is the lack of a standardized mechanism for exchanging participant contact information and validating the updated contact information. This exchange is crucial for maintaining a coherent global state across a DTN network, an essential capability for real-world applications such as Edge Networks and satellite communications.<\/p>\n\n\n\n
This research explores theoretical methods for exchanging contact information among participants and applies these theories to DTN. Specifically, it examines Eventual Consistency Theory in conjunction with established Distributed Snapshot algorithms such as Chandy-Lamport, Lai-Yang, Spezialetti-Kearns, and Epidemic Model Gossip Protocols, assessing how prior research in distributed systems can be applied to extremely remote systems. The study proposes suitable algorithms and combined approaches, evaluates their advantages and disadvantages, and presents quantitative results demonstrating system performance.<\/p>\n\n\n\n
Thursday, June 5<\/h4>\n\n\n\nSUMIT HOTCHANDANI<\/h5>\n\n\n\n
Chair<\/strong>: Dr. Munehiro Fukuda
Candidate<\/strong>: Master of Science in Computer Science & Software Engineering
8:45 A.M.
Project<\/strong>: Link Prediction in Agent-Based Graph Database System<\/p>\n\n\n\nThis work introduces a comprehensive link prediction framework within an agent-based graph database system built on the Multi-Agent Spatial Simulation (MASS) library. Extending prior enhancements to the MASS framework\u2014including support for property graphs and distributed shared graph infrastructure\u2014the system integrates both topological and embedding-based approaches to infer potential links in complex networks. Classical topological methods such as Adamic-Adar, Resource Allocation, and Preferential Attachment are implemented alongside Fast Random Projection (FastRP), a dimensionality reduction technique used to generate node embeddings. These embeddings are subsequently leveraged in a k-Nearest Neighbors (kNN) classifier to identify probable future connections. The FastRP algorithm is realized using a novel agent-based propagation mechanism that exploits MASS\u2019s parallel and distributed architecture.<\/p>\n\n\n\n
By embedding these capabilities into the core MASS library, the proposed framework enables scalable, interpretable, and extensible link prediction for applications in domains such as social networks, biological systems, and recommendation engines.<\/p>\n\n\n\n
\n\n\n\nSURBHI GUPTA<\/h5>\n\n\n\n