C-Day Program

Undergraduate and graduate projects: scale 0- 10 with 0 representing "Poor" and 10 representing "Exceeds Expectations"

• Successfully completed stated project goals and reported deliverables (0-10)
• Methodology/Approach: All required elements are clearly visible, organized, and articulated (0-10)
• Effective verbal presentation (0-10)
• Evidence of Rigor (0-10)
• Merit and Broader impact (0-10)

Games: scale 0 - 10 with 0 representing "Poor" and 10 representing "Awesome"

• TECHNICAL: Technically sound with appropriate visual & audio fidelity(0-10)
• GAMEPLAY: Engaging & Fun, with an intuitive UI. Rules of play are clear. Includes a win/lose state(0-10)
• ORIGINALITY: Sound, Art, Design, or Code(0-10)
• Evidence of Rigor (0-10)
• Merit and Broader impact (0-10)

* Project will be featured during the Flash Session

• UC-016 Multifamily Loan Performance (Undergraduate Project) by ,
  Abstract: This study explores the performance of multifamily loans using a logistic regression model to predict loan outcomes as either 鈥渃losed鈥� or 鈥渃urrent鈥�. Utilizing a dataset of over one million observations and 54,771 unique loan observations, we classify loan status based on Freddie Mac鈥檚 mortgage performance codes, with closed loans including modification with a loss, foreclosures, real estate owned, and fully closed loans. Through explanatory analysis, it reveals a nearly balanced distribution between the binary variables. This dataset supports the use of a logistic regression to model the probability of loan default or completion. The findings have implications for risk mitigation in underwriting practices, helping lenders avoid loans with characteristics like those that historically defaulted.
  Department: Data Science and Analytics
  Supervisor: Prof. Michael Frankel & Dr. Jiajing (Horatio) Huang
   | 

• UC-019 Gwinnett County Public Schools - Data Masking Tool (Undergraduate Project) by , , , ,
  Abstract: In today鈥檚 data-driven world, organizations handle vast amounts of sensitive information, including personally identifiable information (PII), health records, and financial data. For institutions like schools, this data often includes sensitive details about students, parents, and staff, making data protection not just important, but critical. With increasing privacy regulations such as GDPR and HIPAA, organizations must implement robust measures to protect this information while still enabling its use for legitimate purposes like testing, analytics, and development. Our web-based data masking tool addresses this need by allowing organizations to protect sensitive data without compromising its usability. By applying dynamic masking rules to relational databases and generating masked data extracts, the tool ensures compliance with privacy laws while improving operational efficiency. It automates data protection processes and generates realistic, anonymized data, allowing organizations to securely manage and share sensitive information for non-production purposes. Designed for scalability and ease of use, the tool helps organizations streamline their data protection workflows while maintaining the integrity of their testing and analytical environments.
  Department: Software Engineering or Game Design and Development
  Supervisor: Dr. Yan Huang
   | 

• * UC-020 Indy Micro - Virtual 8-Bit Computer (Undergraduate Project) by , , , ,
  Abstract: The Indy Micro is a desktop application which simulates the functionality of an eight-bit personal computer. Its aim is to mimic the feel of owning one such computer in that era, as well as provide an engaging way to learn about low-level computing concepts. The Micro consists of two components: the virtual machine, which is based on the Von Neumann architecture, and the code editor, which allows users to write assembly code and execute it on the virtual machine. The aim is for the Indy Micro to serve as an educational jumping-off point, a step between the casual programmer and the dedicated hobbyist, developing software for real eight-bit systems. One of the ways students get started with programming is with Scratch (scratch.mit.edu), a visual drag-and-drop programming experience. The project鈥檚 goal is to create something like Scratch, but for assembly language, allowing students and hobbyists to learn about low level programming.
  Department: Computer Science
  Supervisor: Prof. Sharon Perry
   | 

• UC-023 Bathtub Racing Game (Undergraduate Project) by , , , , ,
  Abstract: The Virtual Bathtub Racing Game is a capstone project that uses an interactive 3D digital experience to preserve and modernize the long-standing bathtub racing tradition at Southern Polytechnic State University (SPSU). With real-world physics, adjustable features, and multiplayer capabilities, the Unity-developed game recreates the famous event where students raced imaginatively designed bathtub carts. Since stakeholder input influences the creation of tracks, sound profiles, and gameplay elements that replicate the original races, alumni involvement is crucial in determining the authenticity of the game. In order to provide a captivating user experience for both new players and past SPSU students, the project prioritizes historical accuracy while utilizing contemporary gaming technologies. Custom bathtub models, gender-selectable racers, several engine types, and meticulously re-created tracks from SPSU's 1980s layouts and the Marietta campus are among of the key features. The game will be submitted for possible presentation at the C-Day Computing Showcase, setting the stage for upcoming improvements like AI-powered opponents and VR integration. The Virtual Bathtub Racing Game gives a nostalgic yet entertaining digital tribute to SPSU's engineering ethos while building alumni ties by fusing tradition and innovation.
  Department: Information Technology
  Supervisor: Prof. Donald Privitera, Project Sponsor: Joesph Locker
   | 

• * UC-026 Pet Matchmaker (Undergraduate Project) by , , ,
  Abstract: Angel Among Us is a non-profit organization that saves animals from high-killing rate shelters in Georgia. Their goal is to find homes for homeless pets. To increase their efforts, they are developing a web-based application to help improve adoption processes. The goal is to increase adoption rates and be able to provide adopters with information about pets and overall reduce the number of pet returns. This will be accomplished by using adopters鈥� information and preferences from the web-based application to find long-term compatibility with their recommended pets. The objective is to create a web application interface that includes many core components. First, the web application will use PetFinder API to collect pet data and use a catching mechanism to increase data storage by updating pet data daily and ensuring enhanced performance. The mechanism will also decrease the frequency of calls leading to good API accuracy. Adopters will then complete questionnaires to gather information about their preferences, lifestyle, allergies, and experiences. The system will use ChatGPT to analyze the adopter's results and send out recommended pet suggestions based on the given data. Also, pet profiles will be created to get information for database analysts. This will ensure that PetFinder鈥檚 animal description template will be consistent with the saved information from pet profiles to have proper pet descriptions. Overall, this approach encourages responsible pet ownership, reduces pet returns, and increases successful adoption rates.
  Department: Computer Science
  Supervisor: Prof. Sharon Perry
   |  | 

• * UC-027 KSUBlocks Tower Defense (Undergraduate Project) by , , , , ,
  Abstract: Our project, KSUBlocks Tower Defense, is a Minecraft Plugin designed to create a game mode in the Tower Defense genre. We aim to create a unique and fun game for the students in the KSU Minecraft server. Our project is entirely configurable allowing for easy maintenance and room for future expansions, while ensuring the server performance remains steady alongside KSU's other game modes. It is developed in Java, utilizing IntelliJ and Paper API. We plan to deploy it on the KSU Minecraft Server upon finalization.
  Department: Computer Science
  Supervisor: Prof. Sharon Perry
   |  | 

• UC-028 Intelligent Arm Meets Machine Vision (Undergraduate Project) by , ,
  Abstract: Most AI and robots have been used to make mundane tasks easier for humans however intricate tasks, such as monitoring have not been tackled. Using the OpenMANIPULATOR-X, Robot Operating System (ROS2), and machine vision, we planned on having AI tracking monitor with 4 degrees of freedom.
  Department: Computer Science
  Supervisor: Prof. Waqas Majeed & Prof. Sharon Perry
   | 

• * UC-029 GraphBat: Subterranean Data Visualizer (Undergraduate Project) by , , , ,
  Abstract: GraphBat is a desktop data visualization application designed for speleology and similar fields that bundles common graph types with a unique heatmap tool which few comparable apps provide. It was developed in Python and is intended as an open-source tool available for use and extension by the scientific community. The heatmaps offer two data interpolation methods鈥攊nverse distance weighting and linear interpolation鈥攖o visualize the spread of data across a space using a real-world map and sensor data relative to the space. GraphBat aims to expediate scientific analysis and facilitate the presentation of results across many fields of subterranean study.
  Department: Computer Science
  Supervisor: Prof. Sharon Perry
   |  | 

• * UC-030 HeartSpeak AI (Undergraduate Project) by , , , , ,
  Abstract: This project is Sentiment Analysis AI for comprehensive text review analysis and more. The system leverages a fine-tuned BERT-based models to classify overall sentiment, detect emotions, identify sarcasm, and extract aspect-level opinions. Evaluations show robust performance across tasks, with sentiment accuracy around 69%, aspect analysis. Emotion and sarcasm. The pipeline provides actionable insights, empowering businesses to refine products and improve customer satisfaction with OpenAI Integration.
  Department: Computer Science
  Supervisor: Prof. Sharon Perry
   |  | 

• * UC-037 Dynamic Requirements for a Software Training Environment (Undergraduate Project) by , , ,
  Abstract: STEDR outlines the development of a software training environment for Warner Robins Air Base (Robins) to enhance employee coding skills to foster innovative solutions for Air Force projects. STEDR is a proof of concept serving as a dynamic requirements document represented by a user interface, to be delivered to a development team. STEDR involved two phases: (1) requirements gathering via interviews; and (2) interactive user interface development for feedback. The resulting proof of concept, includes an interactive UI and refined requirements, and serves as the foundation for a collaborative project with KSU, enabling the computing colleges to contribute to military computing power.
  Department: Software Engineering or Game Design and Development
  Supervisor: Prof. Sharon Perry
   |  | 

• * UC-040 Security Lookup Interface Project (Undergraduate Project) by , , , ,
  Abstract: The "Security Lookup Interface" capstone project aims to create a streamlined tool for COX's cybersecurity team, enabling analysts to efficiently perform IP address and hostname lookups while providing actionable, data-driven insights to enhance security investigations. The project will develop a user-friendly interface that simplifies the lookup process, allowing cybersecurity analysts to quickly retrieve relevant data and make informed decisions during security investigations. One of the key features of the tool is its seamless integration with both internal APIs and external resources. This integration will ensure that analysts have quick and easy access to valuable information, minimizing manual effort and enabling faster response times. By consolidating data from various sources, the interface will empower security analysts to conduct thorough investigations with minimal friction. A core aspect of the project is its focus on data-driven insights. The system will aggregate data from multiple internal and external sources, presenting actionable conclusions to assist cybersecurity analysts. These insights will help analysts identify malicious patterns, such as frequent appearances of certain IP addresses in known malicious activities, and detect anomalous behaviors, like repeated access attempts or unusual traffic patterns. This aggregated data will streamline the threat investigation process, making it easier for analysts to prioritize threats and take immediate action.
  Department: Computer Science
  Supervisor: Prof. Sharon Perry
   | 

• UC-046 Cat Classification of 20 Distinct Breeds (Undergraduate Project) by , ,
  Abstract: Cat breed classification algorithms have been made time and time before due to cats being such a popular and beloved animal. As such, classification algorithms aim to identify their breeds for veterinary pursuits and wildlife tracking which necessitates accurate classification. Our classification algorithm identifies 20 different CFA-recognized pedigreed cat breeds utilizing TensorFlow with the MobileNetV3 Large model as the base for training. Our preliminary results over 25 initial epochs and 25 fine tuning epochs resulted in a model with a test accuracy of 65%. In the future, we plan to add more techniques to prevent overfitting and experimenting with a more robust dataset which we hope will allow us to achieve our target accuracy of 80%.
  Department: Computer Science
  Supervisor: Dr. Chen Zhao
   | 

• * UC-048 DineNGo - AI Genie (Undergraduate Project) by , , , , 
  Abstract: This projects aims to enhance the flagship product from Driven Software Solutions called DineNGo by implementing a new chatbot to help users with technical troubleshooting. This will allow for instant technical support for common issues and reduces the number of support tickets being created. It provides informed and brief response in a quick manner to walk users through whatever technical issues they are currently having with the DineNGo software. It was built with an Angular frontend and a Node.JS backend as well as a MongoDB database for querying information.
  Department: Computer Science
  Supervisor: Prof. Sharon Perry
   | 

• UC-049 From Forecast to Fortune: Portfolio Optimization and Prediction (Undergraduate Project) by 
  Abstract: This project explores the intersection of time series forecasting and portfolio optimization to support data-driven investment strategies. Historical price data from 30 individual stocks was analyzed using two forecasting models: ARIMA and Prophet. Each model鈥檚 performance was evaluated using key accuracy metrics, including Mean Absolute Percentage Error (MAPE), Root Mean Squared Error (RMSE), and Mean Directional Accuracy (MDA). Results showed that ARIMA performed better on error-based metrics, while Prophet excelled at predicting directional trends. In parallel, historical return data was used to construct optimized portfolios using Modern Portfolio Theory. Two strategies were implemented: one minimizing overall volatility and another maximizing the Sharpe ratio. The optimized asset weights were translated into a simulated $10,000 portfolio, allocating shares based on recent prices. This dual analysis highlights the strengths of different forecasting approaches and demonstrates how predictive modeling can enhance real-world investment decisions.
  Department: Data Science and Analytics
  Supervisor: Prof. Michael Frankel & Dr. Jiajing (Horatio) Huang
   | 

• * UC-061 From Frustration to Function: Enhancing Usability in Public Transportation (Undergraduate Project) by , , , 
  Abstract: Public transportation apps have recently become an essential tool for helping individuals navigate complex transit systems, however, many users still face issues with usability, accessibility, and reliability. Taking this into consideration, this project aims to evaluate the user experience of these apps and how one in particular can be improved. In doing so, our group hopes to create a more user-friendly experience that can make public transportation easier and more reliable for everyone.
  Department: Software Engineering or Game Design and Development
  Supervisor: Prof. Nick Murphy
   | 

• * UC-066 Thought-Memory Model for Multi-Agent Simulation (Undergraduate Project) by , , 
  Abstract: A 2D web-based multi-agent simulation leverages Large Language Models to model human-like interactions among generative agents. A Thought-Memory system retrieves relevant data and prior memories from a database to construct JSON-style prompts for the LLM, which outputs intended agent actions. The system allows for observable, emergent interactions between agents within the simulated space.
  Department: Computer Science
  Supervisor: Prof. Sharon Perry
   | 

• UC-072 鈥淐ommand Center, do you copy?鈥� (Undergraduate Project) by , , 
  Abstract: 鈥淐ommand Center, do you copy?鈥� is a sci-fi themed survival horror game where players must sneak around and fend off an alien like enemy using their flashlight while also trying to find the parts needed to fix their communications system to call for help.
  Department: Software Engineering or Game Design and Development
  Supervisor: Prof. Nick Murphy
   | 

• UC-085 Berry and Carrot - A 2.5D Unity Platformer Game (Undergraduate Project) by , , , 
  Abstract: In Berry and Carrot, you play as two stuffed animals, a bear and a bunny, who are trying to escape from a claw machine that they have been trapped in for years. Each character has different strengths, and the player must use these skills strategically by switching between the two characters to solve puzzles themed around the inner workings of the claw machine featuring screws, springs, levers, and more.
  Department: Software Engineering or Game Design and Development
  Supervisor: Prof. Nick Murphy
   | 

• UC-092 Cookly.io - Advanced Recipe Generator (Undergraduate Project) by , , 
  Abstract: Cookly.io was a passion project started during the AI Club Hackathon where it was awarded 3rd place. Cookly is an AI powered recipe assistant that helps users use available ingredients into delicious meals. Users can input ingredients manually or upload a photo of the pantry or fridge where Cookly will use computer vision to identify the ingredients and SBERT to match the ingredients with the perfect recipe.
  Department: Computer Science
  Supervisor: Dr. Femi Ojo & Dr. Chen Zhao
   | 

• * UC-100 Agentic AI Quiz Generation: Personalized Tutoring through Intelligent Retrieval and Adaptive Learning (Undergraduate Project) by 
  Abstract: This research presents a personalized, agentic AI-powered system for multiple-choice question (MCQ) generation tailored to college-level tutoring in machine learning and software engineering domains. The primary objective is to enhance adaptive learning through reliable, context-aware quiz generation using long-context large language models (LLMs) and modular agent workflows. Our methodology is based on an eight-stage agentic architecture that separates tasks into two main phases: vector indexing and personalized quiz generation. In the indexing phase, academic PDFs are parsed, chunked with LangChain鈥檚 RecursiveCharacterTextSplitter, embedded via Google's text-embedding-005, and indexed using FAISS. A verification agent ensures topic alignment and integrity of the vector database. Upon receiving a user query, a retriever agent performs vector search, followed by a selector agent that filters high-quality chunks. A processor agent curates the final prompt, and a response agent generates the MCQ using Gemini 1.5 Pro. The evaluator agent assesses generated questions against ground truth using metrics like ExactMatch, Faithfulness, and BERTScore. Experimental results over 150 MCQs show Gemini鈥檚 accuracy improves from 78.00% (raw) to 93.33% when enhanced with context vectors, a 100k-token cache, and a 1M-token long-context window鈥攁chieving a +15.33% overall gain. Gemini also excels in Non-Hallucination (0.9150), Certainty (0.8883), and Answer Correctness (0.9260), indicating safe and reliable generation. When supplemented with context vectors and a training cache, these results highlight Gemini鈥檚 effectiveness as a reliable and context-sensitive model for personalized, agentic quiz generation in educational settings, offering strong potential for scalable and adaptive AI tutoring systems.
  Department: Computer Science
  Supervisor: Dr. Nasrin Dehbozorgi
   | 

• UC-101 Sight-Singing Feedback (Undergraduate Project) by , , , 
  Abstract: This project creates an engaging and interactive music-learning experience. Users start by selecting a tempo and melody number. The app then displays sheet music to guide them through the exercise. While singing, performers receive real-time visual feedback on pitch accuracy and tempo progression, allowing for dynamic adjustments and improved performance precision. The system continuously updates the music staff based on user performance, ensuring seamless interaction. This approach integrates technology with musical education, enhancing skill development through intuitive, data-driven feedback. By combining user-driven selections, interactive visualization, and real-time analysis, the application provides a structured, engaging platform for improving musical skills.
  Department: Information Technology
  Supervisor: Prof. Donald Privitera
   |  | 

• * UC-107 Draw The Night Sky (Undergraduate Project) by , , , , 
  Abstract: Draw The Night Sky is a game project made in collaboration with Carter鈥檚 Lake to make their constellation viewing program more accessible. The stars in the sky are quite difficult to see without the perfect conditions, so an alternative would assist with this greatly. By creating a fun and interactive experience through a game, it should teach the visitors of the nature center to be able to search for stars even outside of the game. Utilizing an accurate star map based on the Yale Bright Star catalogue, we have an accurate star map that mirrors the real world which adds to the immersion of the players. In addition to this, very simple real-world tools are provided to the player to find the constellations. Exploring space is quite lonely, so a fun companion in the form of Stella is there to give the player all the right tools to get the job done.
  Department: Software Engineering or Game Design and Development
  Supervisor: Dr. Lei Zhang
   | 

• * UC-111 Accessible Interactive Map (Undergraduate Project) by , , , , 
  Abstract: Finding that walking campus gets you out of breath? We did too! Using React and Flask, we are building a web application that directs KSU students to the path with the lowest elevation and shows the shifts in between. It also displays accessible doors. The purpose of this app is to develop a more inclusive application so people with asthma, cardiovascular issues, and wheelchairs at KSU can safely traverse campus.
  Department: Computer Science
  Supervisor: Prof. Sharon Perry
   |  | 

• UC-116 Robot Tactics (Undergraduate Project) by 
  Abstract: Robot Tactics is a first person strategic shooter made in Unity where the player takes control of an agent who fights off bodyguards who are chasing him while using Robots to detour them
  Department: Software Engineering or Game Design and Development
  Supervisor: Prof. Nick Murphy
   | 

• UC-125 Database Masking Tool - Project 04 - Team 1 (Undergraduate Project) by , , , , 
  Abstract: The Database Masking Tool for Gwinnett County Public Schools secures sensitive data while preserving its analytical utility. Developed alongside an in-depth research paper, this web-based solution enables real-time masking of information in SQL Server and MySQL databases. Utilizing automated field recognition, it applies three masking techniques鈥擣aker-based masking, hash masking, and pseudonymization through generalized masking鈥攖o protect personally identifiable information. Key features include an intuitive interface for configuring masking rules, real-time data previews, and an export function for generating masked datasets in multiple formats. Built with a React-Flask stack and containerized for consistency, the system supports compliance with GDPR, HIPAA, and FERPA. Guided by feedback from sponsor Ed Van Ness and academic advisors, this project establishes a robust framework for scalable data anonymization, enhancing operational efficiency and regulatory compliance.
  Department: Information Technology
  Supervisor: Prof. Donald Privitera
   |  | 

• * UC-130 Some Kind of Tundra Escape (Undergraduate Project) by , , , 
  Abstract: Some Kind of Tundra Escape is an action-adventure game where you play as a penguin trying to escape a vast, walled-in tundra. As you explore, you鈥檒l rescue other penguins scattered throughout the area. Your goal is to gather all the penguins and escape together, but the journey is far from easy. The tundra is full of danger, including snowmen who ambush you and ice golems who patrol certain areas. You'll need to avoid these monsters and set traps to stay safe while rescuing your fellow penguins. As your group grows, so does the challenge鈥攎ore penguins means more obstacles to overcome. Strategize carefully, outsmart the monsters, and lead your group to the exit. Some Kind of Tundra Escape is a thrilling test of wit, stealth, and survival as you race against time to escape with your fellow penguins.
  Department: Software Engineering or Game Design and Development
  Supervisor: Prof. Murphy Nick
   | 

• * UC-136 Foster AI Interview and Biography Generation (Undergraduate Project) by , , , , 
  Abstract: This capstone project presents a proof of concept for a mobile and web-based application designed to streamline communication between foster caregivers and the Angels Among Us Pet Rescue team. The application addresses critical inefficiencies in generating pet biographies and coordinating photography efforts, which are essential components in increasing adoption rates. Leveraging cutting-edge technologies such as Twilio, Retell AI, and OpenAI, the app implements a bio generation workflow that conducts foster interviews via phone calls, transcribes responses using AI-powered voice-to-text, analyzes sentiment, and produces structured, engaging pet bios for platforms like Petfinder. Additionally, the system automates email workflows to coordinate photography sessions, minimizing manual coordination. Built using React for the front end, PostgreSQL for the database, and Python for backend automation, the application emphasizes both usability and data security, ensuring sensitive foster and pet information is handled responsibly. This project demonstrates the practical application of AI and full-stack development to solve real-world challenges in animal rescue and adoption operations.
  Department: Information Technology
  Supervisor: Prof. Donald Privitera
   |  | 

* Project will be featured during the Flash Session

• * GC-008 MediVault: An AI-Powered Secure Medical Image Sharing Platform (Graduate Project) by , , 
  Abstract: MediVault is a secure, cloud-native platform that empowers patients and healthcare providers to upload, view, and share medical images like X-rays and MRIs with confidence. Built using Next.js and Node.js, and deployed on AWS Free Tier (S3, RDS, KMS), the system implements role-based access, two-factor authentication, and end-to-end encryption to ensure privacy and HIPAA compliance. MediVault features an intuitive interface and integrates AI-enhanced automation to streamline metadata tagging and detect potential anomalies in scans. Designed for scalability, usability, and compliance, this project showcases real-world expertise in full-stack cloud development and healthcare cybersecurity.
  Department: Information Technology
  Supervisor: Dr. Ying Xie; Project Sponsors: Gennadiy Kemelmakher, Arpna Aggrawal , Richard Windland
   |  | 

• * GC-025 Secure Medical Image Sharing Platform 鈥� MedShare (Graduate Project) by Adenuga, Toyese, 
  Abstract: The Secure Medical Image Sharing Platform is a cloud-based solution that ensures secure upload, management, and sharing of medical images, adhering to HIPAA and GDPR. It utilizes advanced encryption protocols, role-based access control (RBAC), and audit trails to safeguard patient data. The platform's user-friendly interface facilitates seamless interaction between patients and healthcare providers, enabling the use of AI tools for diagnostic support.
  Department: Information Technology
  Supervisor: Dr. Ying Xie; Sponsors: Mrs. Arpna Aggarwal, Gennadiy Kemelmakher Advisors: Richard Windland
   | 

• * GC-033 OncoClarify 鈥� AI Powered Cancer Report Simplifier (Graduate Project) by 
  Abstract: Cancer pathology reports are important for diagnosis and treatment planning, yet their complex language poses a significant challenge for patients and nurses to understand. This communication barrier often results in confusion, anxiety, delayed decisions, and reduced care quality. To address this, OncoClarify, an AI-powered tool, has been developed to simplify cancer pathology reports and provide role-specific explanations tailored to doctors, nurses, and patients. By leveraging the Gemini API, the system extracts critical details from pathology reports and generates customized summaries. Doctors receive technical insights highlighting biomarkers, mutations, and prognostic indicators. Nurses are provided with actionable guidance for managing side effects and care protocols, supporting effective treatment implementation. Patients benefit from plain-language summaries, enabling a better understanding of diagnoses, treatment options, and potential side effects. This clarity fosters informed decision-making and enhances communication among all parties involved. The tool integrates Tally Forms for report submission and role identification, PDF.co for text extraction from various file formats, and Make.com for seamless automation. The user interface, designed with Gamma.app, ensures ease of access and usability for non-technical users. Output is delivered via email or downloadable PDF. Validation was performed using real-world cancer pathology reports from The Cancer Genome Atlas (TCGA). Accuracy, usability, and processing speed were measured, and feedback from healthcare professionals confirmed its effectiveness. The tool supports scanned images and complies with HIPAA and GDPR standards. Future plans include support for multiple languages, tumor visualization using medical imaging AI, EHR integration, and application to other medical fields such as radiology and cardiology.
  Department: Computer Science
  Supervisor: Dr. Sanghoon Lee
   | 

• GC-039 ClinicPix: Secure Medical Image Sharing Web Application (Graduate Project) by , , , 
  Abstract: ClinicPix is a cloud-based system designed to streamline the management of medical images such as X-rays and MRIs. It offers healthcare providers and patients a secure, intuitive interface to upload, view, and share medical images across institutions and devices. The platform ensures full compliance with HIPAA through robust security measures, including role-based access control, end-to-end encryption, and comprehensive audit trails. Its scalable architecture supports growing data needs while maintaining high performance and reliability. By enhancing accessibility and safeguarding sensitive health information, the platform aims to improve clinical workflows, patient engagement, and collaborative care.
  Department: Information Technology
  Supervisor: Dr. Ying Xie
   | 

• * GC-058 Personalized Wellness Recommendations (Graduate Project) by Yaganti, Varshini
  Abstract: A health recommendation system using machine learning, built on Hadoop, Spark, and HDFS, represents a significant advancement in personalized healthcare. This project aims to leverage big data technologies to process and analyze vast amounts of medical data across a distributed computing environ- ment, utilizing at least three virtual machines.The background of this project lies in the increasing prevalence of chronic diseases and the growing volume of health-related data collected by healthcare providers. The motivation for this project stems from several key factors. Firstly, traditional healthcare systems often struggle to provide personalized recommendations due to the sheer volume and complexity of medical data. By utilizing Hadoop and Spark鈥檚 distributed processing capabilities, this system can efficiently analyze large-scale health data, enabling more accurate and timely recommendations. Secondly, the integration of machine learning algorithms with big data technologies allows for the identification of subtle patterns and correlations in patient data that may not be apparent through conventional analysis methods. This can lead to more precise diagnoses and treatment plans tailored to individual patients.The expected results of this project include a robust health recommendation system capable of processing and analyzing large volumes of medical data in a distributed environment.
  Department: Computer Science
  Supervisor: Dr. Chen Zhao
   | 

• GC-059 Large-Scale Cybersecurity Threat Detection (Graduate Project) by CHILUKURI, PAVAN CHOWDARY, Thiriveedhi, Mohan Krishna Kandimalla, Triveni, Sammeta, Raghava, challapalli, venkata Basanth
  Abstract: Cybersecurity threats are becoming more sophisticated, posing serious risks to critical systems. Traditional intrusion detection systems often fail to manage the scale and complexity of network traffic. This study investigates large-scale threat detection using machine learning in PySpark, utilizing the UNSW-NB15 dataset. It focuses on building scalable models through preprocessing, feature selection, and implementing algorithms like Decision Trees, Na茂ve Bayes, Random Forest, and Gradient Boosting. Evaluation metrics include accuracy, precision, recall, F1-score, and ROC-AUC, with emphasis on hyperparameter tuning and minimizing false positives. Leveraging PySpark鈥檚 distributed computing, the system ensures efficient real-time analysis of vast network data. The research supports modern cybersecurity strategies by enhancing detection reliability and reducing risks from emerging cyber threats.
  Department: Computer Science
  Supervisor: Dr. Dan Lo
   | 

• * GC-074 Real-Time Object Detection (Graduate Project) by Malik, Rohit,
  Abstract: This project explores the implementation of real-time object detection using the You Only Look Once (YOLO) architecture. Leveraging its speed and accuracy, we developed a system capable of identifying and localizing multiple objects within live video streams. Our implementation focused on optimizing YOLO's performance for real-time applications, specifically addressing the trade-off between speed and accuracy. We employed a pre-trained YOLO model and fine-tuned it on a custom dataset tailored to specific object classes. This fine-tuning process aimed to enhance the model's ability to recognize objects in our target environment. The system was implemented using Python and the OpenCV library, enabling seamless integration with camera input and real-time video processing. Performance was evaluated based on frames per second (FPS), mean Average Precision (mAP), and detection latency. Results demonstrate the system's capability to achieve high FPS, facilitating real-time object detection, while maintaining acceptable mAP for accurate object recognition. This project showcases the practicality of YOLO for applications requiring fast and reliable object detection, such as surveillance, autonomous driving, and robotics.
  Department: Computer Science
  Supervisor: Dr. Sanghoon Lee
   | 

• * GC-079 NibbleAI (Graduate Project) by , , 
  Abstract: Ever looked into your fridge or pantry and wondered, 鈥淲hat can I make with this?鈥� NibbleAI is a mobile app designed to solve exactly that. Using artificial intelligence, the app identifies ingredients from user-uploaded images and suggests recipes based on what鈥檚 available. Built with React Native and powered by a DenseNet169 model for image recognition, NibbleAI seamlessly analyzes photos and returns curated recipe ideas 鈥� all within a few taps. This intuitive approach helps users reduce food waste, save time, and get creative with the ingredients they already have.
  Department: Computer Science
  Supervisor: Dr. Arthur Choi
   | 

• * GC-089 SafeCircle: AI and Micro-radar-based Remote Monitoring for Patients with AD/ADRD (Graduate Project) by , , 
  Abstract: Alzheimer's disease and related dementias (AD/ADRD) is an irreversible and degenerative neurological condition that severely impacts neurons, resulting in cognitive decline and memory loss. This study explores a mHealth system, including a SafeCircle iOS prototype, a novel solution that combines artificial intelligence with cutting-edge micro-radar technology. The platform offers a variety of features, including management of patient and caregiver profiles, real-time alerts in case of emergencies, emergency contact lists, one-touch SOS support, sharing of live locations, and recording of unusual events in video. It is a responsive and reliable care assistant that optimizes patient safety while reducing caregiver burden.
  Department: Information Technology
  Supervisor: Dr. Nazmus Sakib & Dr. Sumit Chakravarty
   |  | 

• * GC-123 Deep Learning-Based Skin Cancer Detection (Graduate Project) by , , ,
  Abstract: Skin cancer is increasingly becoming a severe health problem globally today, but early detection is essential to enhance survival rates. Nonetheless, conventional diagnosis relies largely on visual examinations by dermatologists, which can be subjective and time-consuming. This research examines the application of deep learning for the automation of skin cancer detection based on dermoscopic images from the HAM10000 dataset. The models VGG19, DenseNet121 and ResNet152 will be trained and evaluated, with class mbalance addressed using data augmentation strategies. The outputs will demonstrate the applicability of deep learning to improve skin cancer diagnosis. Classification optimization using ensemble modeling and its improved architecture with an attention U-Net to ffer segmentation integration for improved lesion localization and explainability will be future research.
  Department: Computer Science
  Supervisor: Dr. Sanghoon Lee
   | 

• GC-128 Multi-label commit message classification using p-tuning (Graduate Project) by Mistry , Tanvi
  Abstract: Version control systems (VCS) play a crucial role by enabling developers to record changes, revert to previous versions, and coordinate work across distributed teams. In version control systems (e.g., GitHub), commit message serves as concise descriptions of code changes made during development. In our project, we propose to evaluate the performance of multi-label commit message classification using p-tuning (learnable prompt templates) through pre-trained models such as BERT and DistilBERT. The initial results show that p-tuning can provide similar results by designing various flexible templates that are not restricted by fixed templates.
  Department: Software Engineering or Game Design and Development
  Supervisor: Dr. Xia Li
   | 

* Project will be featured during the Flash Session

• UR-001 Large language model enabled mental health app recommendations using structured datasets (Undergraduate Research) by 
  Abstract: The increasing use of large language models (LLMs) in mental health support neces-sitates detailed evaluation of their recommendation capabilities. This study compares four modern LLMs鈥擥PT-4o, Claude 3.5 Sonnet, dataset-enhanced Gemma 2, and dataset-enhanced GPT-3.5-Turbo鈥攊n recommending mental health applications. We constructed a structured dataset of 55 mental health apps using RoBERTa-based sentiment analysis and keyword similarity scoring, focusing on depression, anxiety, ADHD, and insomnia. Standard LLMs emonstrated inconsistent accuracy and often relied on outdated or generic information. In contrast, our retrieval-augmented generation (RAG) pipeline enabled lower-cost models to achieve 100% accuracy, compared to baseline models (GPT-4o at 45% and Claude at 70%), while maintaining good diversity and recommending apps with significantly better user ratings. These findings demonstrate that dataset-enhanced, cost-effective LLMs can outperform expensive proprietary models in domain-specific applications like mental health resource recommendations, potentially improving accessibility to quality mental health support tools.
  Department: Computer Science
  Supervisor: Dr. Md Abdullah Al Hafiz Khan
   | 

• UR-002 FedDA-TSformer: Federated Domain Adaptation with Vision TimeSformer for Left Ventricle Segmentation on Gated Myocardial Perfusion SPECT Image (Undergraduate Research) by Huang, YeHong
  Abstract: This study presents FedDA-TSformer, an approach for accurate left ventricle segmentation in gated myocardial perfusion single-photon emission computed tomography (MPS) images, designed to ensure both high segmentation quality and patient data privacy. By integrating federated learning with domain adaptation techniques, the proposed model leverages a novel Divide-Space-Time-Attention mechanism that effectively captures spatio-temporal correlations inherent in multi-centered MPS datasets. Domain discrepancies among data from three different hospitals are mitigated using a local maximum mean discrepancy (LMMD) loss, enabling robust performance across various clinical settings. Evaluated on a dataset comprising 150 subjects with eight distinct cardiac cycle phases, FedDA-TSformer achieved Dice Similarity Coefficients of 0.842 and 0.907 for the segmentation of the left ventricular endocardium and epicardium, respectively. These results demonstrate the model's potential to improve the functional assessment of the left ventricle while upholding stringent data security standards.
  Department: Computer Science
  Supervisor: Dr. Chen Zhao
   | 

• * UR-017 SHERLOCK: Self-supervised Histopathological Evaluation for Recognition of Lymphocytes and Other Cancerous Kinds (Undergraduate Research) by 
  Abstract: Whole Slide Images (WSI) are gigantic images (e.g. 100k x 100k pixels) of tissue samples. The goal of SHERLOCK is to detect cancer cells in those tissue samples. We do this by using a pretrained Masked Autoencoder (MAE), from Facebook鈥檚 research lab, that we finetune on the PanNuke dataset. The benefit of using an MAE is that unlike supervised learning the WSI鈥檚 don鈥檛 need to be labeled. This is important because it will save a lot of time and money that would be spent on labeling WSI鈥檚.
  Department: Computer Science
  Supervisor: Dr. Sanghoon Lee
   |  | 

• UR-018 Towards Bounding the Behavior of Deep Neural Networks (Undergraduate Research) by 
  Abstract: Recent advances in Artificial Intelligence (AI) have unlocked many new possibilities but have also brought with it many new challenges. While modern AI systems have been continuously exceeding expectations, our ability to interpret and understand their behavior lags behind. For example, an AI model trained to detect pneumonia from X-rays may fail in new hospitals because it learned to recognize hospital logos instead of medical patterns. Why do some succeed while others fail? Do they truly understand their tasks, or are they relying on patterns that may not always hold? To enumerate the most informative explanations of a neuron鈥檚 behavior, we developed an improved approach to bounding the behavior of individual neurons within artificial neural networks. In this research we demonstrate, both theoretically and empirically, the utility of our approach.
  Department: Computer Science
  Supervisor: Dr. Arthur Choi
   | 

• * UR-031 Impact of Motor Skill on Learning Experiences and Outcomes using Note-Taking in VR (Undergraduate Research) by 
  Abstract: Immersive learning experiences have been proposed to offer rich immersion and interaction, effectively addressing the distractions and low engagement commonly found in typical online learning environments. Research in neuroscience and psychology suggests that motor skills, such as note-taking, help students improve their learning by enhancing cognitive abilities and decision-making, ultimately leading to better performance. This study aims to investigate the impact of motor skills, specifically note-taking with a physical VR stylus, on learning experiences, outcomes, and retention in our VR classroom environment.
  Department: Software Engineering or Game Design and Development
  Supervisor: Dr. Sungchul Jung
   | 

• * UR-044 Quantum Machine Learning for Science and Engineering (Undergraduate Research) by , , 
  Abstract: This research explores the comparative effectiveness of traditional machine learning algorithms and their quantum counterparts. Traditional and quantum implementations of algorithms including Support Vector Machines (SVM), logistic regression, Principal Component Analysis (PCA), random forest classifiers, neural networks, and convolutional neural networks (CNN) are evaluated and contrasted. Findings highlight that quantum algorithms can provide certain clear advantages in some models and data while exhibiting inferior performance in others. By assessing these nuances, this research helps contribute to the understanding of quantum machine learning algorithms and their potential applications for science, engineering, and industrial tasks.
  Department: Computer Science
  Supervisor: Prof. Sharon Perry & Dr. Yong Shi
   | 

• UR-047 Empathetic VR Classroom (Undergraduate Research) by Brice, Seth
  Abstract: We used a virtual reality classroom setting to investigate how accurately humans can determine the emotional state of NPC avatars based on nonverbal body language. Volunteers presented a topic in front of several virtual agents, who would respond with a gesture that reflected their current emotional state, giving the presenter the opportunity to physically and emotionally respond to these changes.
  Department: Software Engineering or Game Design and Development
  Supervisor: Dr. Sungchul Jung
   | 

• UR-063 K86: 16b Computer and Assembler Design and Implementation (Undergraduate Research) by , 
  Abstract: With this project, we designed a general-purpose 16-bit RISC+CISC computer architecture, alongside an assembler, instruction embedder, and preliminary compiler. Our computer architecture, K86 (最色导航 86), is inspired by the Intel x86 and ARM architectures that have enabled computing systems to perform many of the modern functionalities we rely on today. To allow for fluid programming and processing, KASM (最色导航 Assembler) translates assembly code into machine instructions which will be stored in the computer memory by the embedder. With the addition of a preliminary compiler to produce assembly from high-level source code, our project defines much of the foundation of a sophisticated computing system.
  Department: Computer Science
  Supervisor: Prof. Waqas Majeed
   |  | 

• UR-086 Whole Slide Image Analysis (Undergraduate Research) by , 
  Abstract: Whole Slide Images are used to capture details of patient cells. Hospitals and clinics have different processes and methods to create WSIs resulting in WSIs not being standardized. Different file formats are used and different colors are used to represent different features. The normalization process helps set up the WSI into a format that the current model can easily process.
  Department: Computer Science
  Supervisor: Dr. Sanghoon Lee
   | 

• * UR-094 AIStudy: Using AI to Study AI (Undergraduate Research) by 
  Abstract: Interactive AI studying tool or AIStudy is a flask-based web-app which enables users to quickly search, save, and study scientific papers. AIStudy streamlines the literature review process by utilizing large language models (LLMs) allowing for users to engage with research in a creative and interactive way. To begin with a user searches up papers using the arXiv API and PyMuPDF for scraping the contents. These are saved to a user database managed by SQL Alchemy. The user can then ask a chatbot about one or more papers at a time through Ollama鈥檚 API in order to produce Retrieval-Augmented Generated (RAG) responses. Using AIStudy, we investigated the current research in human-AI collaboration.
  Department: Computer Science
  Supervisor: Dr. Md. Abdullah Al Hafiz Khan & Prof. Sharon Perry
   | 

• UR-099 Empowering Mental Wellness: A Comprehensive Study and Design of a Predictive System for Early Mental Health Intervention (Undergraduate Research) by 
  Abstract: Mental health is an essential part of living a balanced and fulfilling life, but it is often overlooked compared to physical health. While physical health is important for performing daily activities, mental health plays a crucial role in how we manage stress, build connections, and make decisions. Previous research studies have shown that nearly 60 million Americans experienced a mental illness in 2024, yet there were only 340 people for every one mental health provider in the U.S. Furthermore, young adults aged 18鈥�25鈥攚ho are the most digitally connected generation鈥攕uffer from the highest rates of severe mental illness yet are the least likely to seek or receive treatment. These findings highlight a growing crisis where more people are struggling with mental health issues, but the resources available to help them remain insufficient. This study presents a comprehensive investigation into predictive models and datasets for early mental health intervention, combining a systematic literature review with empirical research. We examine a range of existing machine learning algorithms and datasets that focus on behavioral and physiological indicators, including heart rate variability, sleep patterns, device usage, and social interaction metrics. Through critical evaluation of these models, we identify key features and data types most effective for predicting early signs of mental health conditions. Based on these insights, we design a predictive system architecture, including form-matching tables that align symptom inputs with appropriate risk levels and recommended actions. To translate the system into an accessible user experience, we develop mobile application wireframes and conduct usability research on features that support early detection and intervention. This work aims to bridge the gap between technical innovation and user-centered design, offering a holistic and proactive approach to empowering mental wellness through early intervention.
  Department: Computer Science
  Supervisor: Dr. Maria Valero
   |  | 

• * UR-112 Monarch: A Privacy-focused NLP Model for Emotional Pattern Detection (Undergraduate Research) by , 
  Abstract: Introducing: Monarch 鈥� a privacy-focused deep learning model that interprets emotional patterns in text. Monarch is trained on large, lexicon-based datasets and uses fine-tuned NLP models (BERT) to identify patterns associated with sadness, worry, anger, and distress. It runs entirely offline with no data collection, making it ideal for private use. Monarch evaluates text and returns clear, readable probability scores across emotional categories, giving users insight into emotional trends. Monarch is interpretive, not diagnostic, displaying results based on scientifically backed linguistic patterns. Its potential use in schools could help flag early signs of distress, giving educators a chance to support those in need. Monarch is also suitable for research in linguistics, mental health, and ethical AI implementations.
  Department: Computer Science
  Supervisor: Dr. Jeff Adkisson
   |  | 

• UR-114 K86: 16-Bit Computer Design, Optimization, and Implementation (Undergraduate Research) by , , , , 
  Abstract: This research focused on the implementation of modern computing systems by designing and simulating a 16-bit RISC-based ISA computer. The computer is built on a Von Neumann memory architecture with 1024脳16-bit word-addressable space and a 6-bit ISA with 36 implemented instructions. The central processing unit (CPU) includes a control unit (CU) that automatically drives the fetch-decode-execute (FDE) cycle, four addressable general-purpose registers (GPRs), and an Arithmetic Logic Unit (ALU) comprising 21 operations and producing four flags. We validated the system by executing Euclid's GCD algorithm, generating the binaries with a custom assembler written in Python.
  Department: Computer Science
  Supervisor: Prof. Waqas Majeed
   | 

• UR-115 MobiNav: Accessible Campus Navigation (Undergraduate Research) by , , 
  Abstract: MobiNav addresses the gap in campus navigation by providing personalized route planning for individuals with diverse mobility requirements. The system uses dual-layer routing (Google Maps API and custom OSRM routing), real-time obstacle reporting, and detailed accessibility feature mapping. It creates custom routes considering wheelchair access, elevation changes, building entrances, and temporary obstacles. Initially scoped for 最色导航's Marietta campus, it is designed for scalability to other locations.
  Department: Software Engineering or Game Design and Development
  Supervisor: Dr. Yan Huang
   |  | 

• * UR-126 Multimodal Neuroimaging Meets AI: Enhancing Alzheimer's Diagnosis with PyRadiomics (Undergraduate Research) by Callaway, Dina Xu, Castillo, Maya, Haynes, Richard
  Abstract: Alzheimer鈥檚 disease (AD) is a progressive neurodegenerative disorder that requires early and accurate diagnosis for effective intervention. This research explores how multi-modal data integration can enhance Alzheimer鈥檚 disease staging prediction by developing an AI model that classifies patients into normal, mild cognitive impairment (MCI), or AD stages. Unlike traditional methods that rely on clinical assessment to make diagnoses, this study develops an AI-driven approach that integrates clinical and imaging data to improve classification accuracy. The research utilizes the Australian Imaging, Biomarkers & Lifestyle (AIBL) dataset, importing patient clinical data along with PET and MRI scans. First, image features were extracted from 1,312 MRI scans (705 patients) and 1,566 PET scans (829 patients) using PyRadiomics. Each scan yielded 112 features. Then, these extracted features were combined with 46 clinical variables to create a multi-modal dataset. To ensure consistency, data selection was performed by including only patients with both MRI and PET scans and a recorded CDR score, while non-numerical features were removed. This resulted in 270 multi-modal features used to train a machine learning model on 681 patients (1,448 scans). The model demonstrated strong performance, with an overall accuracy of 94% in distinguishing between normal control (NC), mild cognitive impairment (MCI), and Alzheimer鈥檚 disease (AD). Binary classification models further highlight the model鈥檚 effectiveness, achieving 100% accuracy (AUC = 1.000) in AD vs. NC classification, 93% accuracy (AUC = 0.972) in AD vs. MCI, and 97% accuracy (AUC = 0.949) in MCI vs. NC. This research contributes to the field by proposing a data-driven AI framework for precise AD diagnosis, potentially aiding clinicians in early intervention decisions and improving patient outcomes. Future work will validate the model on larger, diverse cohorts to ensure generalizability.
  Department: Computer Science
  Supervisor: Dr. Chen Zhao
   | 

* Project will be featured during the Flash Session

• GRM-011 Non-Invasive Convolution-Based Coronary Artery Blood Pressure Prediction (Master's Research) by 
  Abstract: The primary objective of this proposal is to develop an innovative technique for determining the functional significance of coronary artery lesions in patients with coronary artery disease (CAD) and evaluate its utility for clinical decision-making using coronary computed tomography angiography (CCTA).
  Department: Computer Science
  Supervisor: Dr. Chen Zhao
   | 

• GRM-012 (TCC) Transformer Embedded Synthetic Source Code Multiclass Classification (Master's Research) by , 
  Abstract: Recent advances in large language models have significantly increased their capability to write code. While tools such as ChatGPT are useful and represent increased efficiency for many programmers, they represent a major issue when used in academically dishonest ways. To solve the problem of identifying code written by language models, we offer a novel, light-weight classification solution based on a transformer architecture. We compare the performance of three separate transformer models (GraphCodeBERT, PLBART, and CodeBERT) for tokenization and processing and then perform classification using a random forest classifier. Preliminary results indicate that the GraphCodeBERT-based model has a 100% test and train accuracy on detecting human or AI generated code and PLBART has 100% train with 95% test F1-score on categories of AI generators like chatbot, model, IDE extension, or human
  Department: Computer Science
  Supervisor: Dr. Md Abdullah Al Hafiz Khan
   | 

• * GRM-022 Exploring Coronavirus 2019 Datasets With Convolutional Neural Networks (Master's Research) by , , , , 
  Abstract: Over the past several decades, the healthcare sector has increased its data creation velocity at an astonishing rate. More doctors and patients have access to real-time imaging technology, which leads to earlier detection and diagnosis for a variety of diseases. In this project, we will explore several datasets that were gathered by various health organizations during the Coronavirus 2019 (COVID-19) pandemic. We will leverage big data analytics techniques and neural network modeling to gain deeper insights into the differentiated diagnosis of COVID-19.
  Department: Computer Science
  Supervisor: Dr. Dan Lo
   | 

• GRM-038 Optimizing Prompts for Alzheimer's Speech Classification Using LLM (Master's Research) by Shahid, Imaan
  Abstract: Large Language Models (LLMs) are widely used in Alzheimer's disease research to classify speech patterns. However, there is no standardized framework to ensure the reliability of prompts used in these classifications. This study investigates the sensitivity of Alzheimer鈥檚 disease classification prompts to small variations and finds that these prompts are indeed sensitive, leading to inconsistencies in model performance. To address this, we implement an automatic prompt optimization framework to refine the base prompt. Experimental results demonstrate that the optimized prompt improves classification accuracy by 12.83% compared to the baseline, underscoring the significance of systematic prompt engineering in enhancing the reliability of LLM-based Alzheimer鈥檚 disease detection. Although the optimized prompt remained sensitive to variations, it consistently showed improved overall accuracy.
  Department: Data Science and Analytics
  Supervisor: Dr. Seyedamin Pouriyeh
   | 

• * GRM-041 AI-Driven Analysis of OpenALG Curriculum: Mapping AI Competencies Across Georgia鈥檚 Higher Education Landscape (Master's Research) by , , 
  Abstract: This project investigates the presence of artificial intelligence (AI) competencies across Georgia鈥檚 higher education curriculum using university course catalogs as the primary data source, supplemented by OpenALG materials. We applied large language models, including OpenAI鈥檚 ChatGPT and embedding APIs, to analyze over 34,000 courses summarizing content, classifying AI relevance, and mapping to global frameworks (AI4K12 and UNESCO). Techniques such as topic clustering, semantic similarity analysis, and geographic distribution mapping were used to uncover patterns in AI integration. Findings reveal that AI content is concentrated in computing disciplines and research universities, with limited coverage in community colleges, MSIs, and non-technical fields. The results highlight the need for more equitable and interdisciplinary AI education across Georgia鈥檚 institutions.
  Department: Information Technology
  Supervisor: Dr. Ying Xie
   | 

• * GRM-042 iHelp: A Care Partner Activation Program mHealth System for AD/ADRD Caregivers (Master's Research) by Bhowmick, Trisha,
  Abstract: The iHelpCare platform is designed to offer a seamless and supportive experience for patients and caregivers through a clear and user-friendly interface. Users begin at the login page, where they can either sign in or create a new account. Once logged in, the home page provides access to essential services such as a 24/7 helpline, emergency visit coordination, emergency support, and a service directory. It also includes engagement tools like discussion forums, learning modules, and resource materials, along with community-focused features such as events, activities, and support groups. The personalized dashboard allows users to monitor health conditions, review patient history, receive notifications and alerts, and communicate directly with caregivers through live chat. Overall, iHelpCare combines health support, education, and real-time communication in one platform, making healthcare management more accessible and efficient.
  Department: Information Technology
  Supervisor: Dr. Nazmus Sakib
   | 

• GRM-043 Performance Assessment of DeepSeek versus Bard and ChatGPT in Detecting Alzheimer鈥檚 Dementia (Master's Research) by 
  Abstract: Alzheimer鈥檚 disease is a growing public health issue due to its progressive nature and increasing prevalence. Large language models (LLMs) offer promising avenues for non-invasive cognitive assessment through natural language understanding. In this study, we evaluate DeepSeek鈥檚 general-purpose model V3 and reasoning-enhanced R1 variant鈥攆or identifying Alzheimer鈥檚 dementia (AD) and Cognitively Normal (CN) individuals using transcripts derived from spontaneous speech. Two baseline prompting strategies (zero-shot, chain-of-thought ) were applied to both model types and an additional query (self-consistency prompting) was applied to assess better predictions. Accuracy was the primary performance metric. When positively identifying AD, the general-purpose DeepSeek V3 model produced the highest true positives at 88%, but tended to misclassify CN as AD. In contrast, the DeepSeek-R1 model achieved the highest true negatives at 90% for CN classification. Overall, DeepSeek models surpass chance-level classification, but further refinement is needed before clinical applicability can be ensured.
  Department: Software Engineering or Game Design and Development
  Supervisor: Dr. Seyedamin Pouriyeh
   | 

• * GRM-050 Context-Aware Misinformation Detection Using Fine-Tuned BERT and BiLSTM with Attention (Master's Research) by Gurung, Rakshak, Tkabladze, Nino
  Abstract: Misinformation spreads fast, and 60% of consumers now question media reliability (Redline Digital, 2023). Manual verification is slow, and most systems still rely on binary real/fake classification, which overlooks nuanced types of misinformation. We propose a multi-class deep learning approach using a fine-tuned BERT model and a custom BiLSTM with attention to better detect categories like satire, conspiracy, and bias. Our models were trained on a balanced subset of the Fake News Corpus using nine distinct misinformation classes. By addressing both class imbalance and linguistic ambiguity, this system enhances contextual understanding and improves detection across varied news content. Our approach demonstrates that scalable, multi-class classification provides a more accurate and insightful solution to misinformation detection.
  Department: Computer Science
  Supervisor: Dr. Md Abdullah Al Hafiz Khan
   | 

• GRM-060 Abstractive Summarization of Informal Text: Fine-Tuning Transformers on Reddit Discussions (Master's Research) by Ming, Nong, Challa, Arpana, Edward John, Sharon
  Abstract: In recent years, the rapid growth of social media platforms has led to an information overload, as a result, the ability to compress long and complex texts into short and precise summaries is essential, especially in online discussions and comment sections. Summarizing such content is difficult due to inconsistencies in sentence structure, slang, abbreviations, and the lack of formal grammar. State-of-the-art models such as BART and PEGASUS have shown promising results, but their performance on informal datasets remains lower compared to structured text benchmark. To address these challenges, we fine-tune BART and PEGASUS on the Reddit TIFU dataset, leveraging their transformer-based architectures to improve abstractive summarization of informal text. Our contribution lies in adapting state-of-the-art summarization models specifically for informal, user-generated discussions, making summarization more effective for online platforms. Our fine-tuned model achieves a 6.6% improvement in ROGUEL compared to existing summarization model, demonstrating its effectiveness in generating concise and coherent summaries of Reddit discussions.
  Department: Computer Science
  Supervisor: Dr. Md Abdullah Al Hafiz Khan
   | 

• GRM-076 Assessing the Performance of Intelligent Agents in Visual Food Recognition Relative to Manual Data Entry (Master's Research) by , 
  Abstract: Accurate dietary assessment remains a critical yet time-consuming task in health and nutrition monitoring. This study benchmarks the macronutrient estimation capabilities of three intelligent vision agents: GPT Vision, Claude, and Gemini against manually logged food data. We unify two distinct datasets: MenuMatch, annotated by a professional nutritionist, and CGMacros, populated through user entries on MyFitnessPal. After flattening and cleaning both datasets, we first assess each model鈥檚 performance in calorie estimation. GPT Vision outperforms the others with the lowest percentage error 13.83% and is subsequently used to benchmark the macro estimations of Claude and Gemini. While Claude shows higher carbohydrate and fat estimation errors, Gemini yields the most balanced results across protein 12.55%, carbohydrates 19.57%, and fats 17.07%. These findings reveal strengths and trade-offs in current intelligent agents for visual food recognition, informing the development of more accurate, user-friendly, AI-powered nutrition tracking systems.
  Department: Computer Science
  Supervisor: Dr. Maria Valero
   |  | 

• GRM-080 Leveraging Data Science for Resilience: Improving Trauma-Informed Care Practice for Adverse Childhood Experience with AI & Data Science Application (Master's Research) by 
  Abstract: Adverse Childhood Experiences (ACEs) have long-lasting effects on physical health, mental well-being, education, and socioeconomic outcomes. Resilient Georgia (RG), a statewide initiative, seeks to address ACEs through trauma-informed care and data-driven strategies. However, challenges in data collection, analysis, and tracking hinder the effectiveness of these efforts. This study explores the role of data science and interactive visualization tools in improving outcomes for individuals and communities affected by ACEs. A key focus of this research is the development of a data science management application designed to enhance data collection and facilitate real-time decision-making. The application features interactive dashboards that allow stakeholders such as policymakers, healthcare providers, and community leaders to visualize program outcomes and track trends. Users can input data manually, upload files, and generate custom visualizations, making data more accessible and actionable for informed decision-making. By integrating these data-driven tools, RG can improve the monitoring and evaluation of its programs, optimize resource allocation, and ensure that vulnerable populations receive timely support. This study contributes to the growing body of literature on leveraging technology and data science to mitigate the impact of ACEs and promote long-term well-being.
  Department: Data Science and Analytics
  Supervisor: Dr. Nazmus Sakib
   |  | 

• * GRM-081 Evaluation of hand-crafted features with mask images obtained from PanNuke dataset using Bayesian optimization and machine learning models (Master's Research) by 
  Abstract: Semantic image segmentation enables computing systems to understand the semantic patterns of image pixels by using deep learning models to classify the pixels into specific labels. The deep-learning models鈥� performance in image classification has been evaluated by comparing the predicted images using deep-learned features with human-labeled images or mask images. However, there remains a substantial need to investigate the performance of machine learning models that do not use deep learned features but use hand-crafted features. In this project, we perform a comprehensive evaluation of the performance of the eight machine learning models using 46 hand-crafted features extracted from the PanNuke dataset including 5,179 hematoxylin and eosin images with 161,739 cell nuclei, by optimizing feature selection through Bayesian optimization. The evaluation results indicate that the ensemble learning-based models achieve higher performance compared to others across precision, recall, f1-score, and accuracy.
  Department: Computer Science
  Supervisor: Dr. Sanghoon Lee
   | 

• * GRM-083 Leveraging Graph Attention Networks and BERT for Robotic Surgery Report Generation (Master's Research) by 
  Abstract: This project focuses on generating surgical reports from robotic surgery videos by leveraging graph-based representations of instrument-tissue interactions. We utilize Graph Attention Networks (GAT) to model these interactions, which are then integrated into a BERT-based language model for caption generation. Our approach enhances the accuracy of automated surgical reporting by capturing spatial and relational dependencies within surgical scenes. The model is evaluated on the Robotic Instrument Segmentation dataset from the 2018 MICCAI Endoscopic Vision Challenge(Endovis-18) and TORS surgery dataset, achieving high performance across multiple metrics, including BLEU-n, Cider, and ROUGE scores. By automating report generation, this study aims to assist healthcare professionals in improving post-surgical care, optimizing procedural efficiency, and enhancing decision-making in robotic-assisted surgeries.
  Department: Computer Science
  Supervisor: Dr. Chen Zhao
   | 

• * GRM-093 Advances in Non-Invasive Glucose Sensing: A Comprehensive In Vitro Analysis (Master's Research) by , 
  Abstract: This study explores non-invasive glucose sensing using infrared (IR) imaging and electrical measurements in an in-vitro setup. Glucose samples (70鈥�200 mg/dL) were prepared by diluting concentrated solutions (700鈥�2000 mg/dL) 1:10 in synthetic blood concentrate, with 2 mg/dL increments. A custom 3D-printed black cuvette holder ensured consistent alignment of components, including either an IR camera or a 1550 nm photodiode, light sources (850 nm LED/laser, 808 nm, 650 nm, or 1600 nm), and a 3 mm skin-mimicking silicone layer. A Region Based Convolutional Neural Network (RCNN) trained on IR images achieved the lowest RMSE of 10.98 mg/dL at 850 nm LED. A Random Forest model using the recorded-to-baseline voltage ratio yielded an R虏 of 0.786, RMSE of 17.62 mg/dL, and MAE of 14.05 mg/dL. Clarke Error Grid analysis confirmed clinical relevance.
  Department: Computer Science
  Supervisor: Dr. Maria Valero
   |  | 

• GRM-102 Analysis of Climate Change Effects on Bird Migration Patterns Using Long-Term Data (Master's Research) by Syed, Aadil, Beyioku, Mary, Aizebeokhai, Osi, Dogbe, Felix
  Abstract: To investigate how climate change has affected bird migration patterns over the past decades, focusing on changes in migration timing, routes, and population trends. This project will aim to identify correlations between climate variables and observed changes in bird behavior, contributing to conservation efforts and climate change research.
  Department: Information Technology
  Supervisor: Dr. Ying Xie
   |  | 

• * GRM-109 Quantum Machine Learning For Science And Engineering Research (Master's Research) by 
  Abstract: This research project aims to understand and explore the practical applications of Quantum Machine Learning (QML) in solving real-world challenges. By comparing classical machine learning models such as Support Vector Machines (SVM), Neural Networks, Logistic Regression, and Naive Bayes, with their quantum counterparts. Quantum Support Vector Machines (QSVM), Quantum Neural Networks (QNN), Quantum Logistic Regression (QLR), Quantum Deep Neural Networks (QDNN), and Hybrid Quantum Models, we gain hands-on experience in advanced machine learning techniques. The project cover diverse domains including cybersecurity, healthcare, industrial engineering, energy management, and supply chain optimization. Each part of project involves working with real-world datasets, preprocessing, parameter tuning (like qubit settings), and performance evaluation using platforms such as PennyLane and Qiskit. Through this project, we not only learn about the theoretical foundations of QML but also develop practical skills in applying quantum models to high-dimensional and complex data for tasks like fraud detection, quality prediction, patient flow analysis, energy efficiency estimation, and predictive maintenance.
  Department: Computer Science
  Supervisor: Dr. Yong Shi
   | 

• * GRM-118 Analysis of Climate Change Effects on Bird Migration Patterns Using Long-Term Data (Master's Research) by Banneni, Saikiran, Arabu, Naga Sreeja, , , Devisetty, Leelakarthik
  Abstract: This project examines the impact of climate change on bird migration patterns by integrating bird observation data from eBird with climate data from the NOAA Global Historical Climate Network (GHCN). Focusing on species such as the Arctic Tern, the study analyzes changes in migration timing, routes, and population trends over recent decades. Migration paths were visualized using QGIS, while MODIS land cover data helped assess habitat changes along these routes. Temporal analysis revealed noticeable shifts in migration timing, with earlier arrivals in some regions correlating with rising temperatures and changing precipitation patterns. For future predictions, CHELSA climate data was combined with machine learning models, including Gradient Boosting and Random Forest, to forecast migration behavior under different climate scenarios. The results highlight critical stopover sites increasingly threatened by habitat loss, emphasizing the need for targeted conservation efforts to support migratory species adapting to a changing climate.
  Department: Information Technology
  Supervisor: Dr. Ying Xie
   | 

• * GRM-120 Coding Neurodivergent (Master's Research) by 
  Abstract: This literary research provides a look at neurodivergent individuals learning coding, specifically Python, and the struggles and benefits that come from the way their brains are wired. This literature research was conducted to look at the benefits and struggles of learning to code as a neurodivergent individual. One area that has not been studied extensively is the learning of Python, or coding in general, by neurodivergent populations. This includes the benefits a neurodivergent learner may glean from learning Python as well as the challenges they may encounter - anything which makes their experience different from that of a neurotypical Python beginner. Outcomes for which data was sought included cognitive challenges neurodivergent individuals may face (such as difficulty understanding abstract concepts or issues with motivation), ways in which the neurodivergent brain may be well-suited to learning Python, and ways in which neurodivergent populations may benefit from learning Python.
  Department: Information Technology
  Supervisor: Dr. Zhigang Li
   | 

• GRM-131 XR Agent (A MLLM powered XR system) (Master's Research) by Shen, Yukang
  Abstract: This project proposes 鈥淴R Agent鈥�, a uncoupled and efficient framework for developing AI-powered extended reality (XR) applications on head-mounted displays (HMDs). Leveraging multimodal artificial intelligence鈥攊ncluding MediaPipe(Google open-source CV Model) for computer vision (object segmentation, recognition, pose estimation), multimodal large language models (MLLMs) like Gemini, and Unity鈥檚 cross-platform XR development ecosystem鈥攖he framework aims to create an extensible base system that enables rapid prototyping and deployment of intelligent XR applications. Currently, it was deployed on the Meta Quest 3 platform, XR Agent explores novel HCI(Human Computer Interaction) paradigms, combining real-time sensor data processing, immersive visualization, and adaptive AI-driven logic. This work addresses challenges modular integration of various different kinds of devices AI models. The framework also will be valuable through use cases in collaborative remote control, immersive training scenarios, and data collection for embodied AI.
  Department: Software Engineering or Game Design and Development
  Supervisor: Dr. Yan Huang
   |  | 

* Project will be featured during the Flash Session

• GRP-010 Autonomous Agents in the Loop: Strengthening Educational Recommenders with Computer Use Agents (PhD Research) by 
  Abstract: Pedagogical Design Patterns (PDPs) serve as reusable, research-informed strategies that support effective teaching, yet their discoverability remains a major hurdle for educators. In this work, we extend the PDPR (Personalized Dynamic Practice and Reflection) system with a Retrieval-Augmented Generation (RAG) framework powered by a fine-tuned large language model (LLM) to deliver context-aware PDP recommendations. A key innovation in our proposed system is the integration of a Computer-Using Agent (CUA), which acts as a fallback mechanism when the internal knowledge base lacks sufficient coverage or yields low-confidence responses. This agent autonomously interacts with a live desktop environment鈥攗sing browser automation, mouse control, keyboard input, and screenshots鈥攖o search the web, download educational resources, and summarize external content relevant to the user's input query. Through a dynamic agentic loop, the CUA bridges the gap between static knowledge and real-time discovery, expanding the system's ability to provide grounded, practical instructional support. We evaluate our framework using the RAG Triad methodology and qualitative feedback from educators, demonstrating both high relevance and adaptability of the recommendations. This hybrid architecture not only strengthens pedagogical support for educators but also pushes the boundary of AI-driven educational tools by blending structured retrieval with autonomous information-seeking Agents.
  Department: Computer Science
  Supervisor: Dr. Nasrin Dehbozorgi
   | 

• GRP-021 SHAP-Explainable Image-to-Topology Regression (PhD Research) by Fanning, Charles
  Abstract: We evaluated whether deep regression models predicting vectorized topological features (in the form of persistence landscapes) actually learn the underlying persistent homology of the image. A DenseNet-121 is trained to regress 300-dimensional persistence landscapes from grayscale scene images. Using SHAP, we evaluate the contribution of pixels in the original images to the persistence landscapes. Across all six classes, SHAP-feature overlap is consistently lower than the baseline, implying that DenseNet may not be truly learning the underlying persistent homology.
  Department: Data Science and Analytics
  Supervisor: Dr. Bin Luo
   | 

• * GRP-053 EXPAND: Explainable AI Integrated Deep Learning-based Reconstruction of the Lost Packets (PhD Research) by Ahmed, Nasim, Ridwan, A E M
  Abstract: Advanced networking technology faces challenges with diverse usage, especially packet loss. Researchers tried deep learning to predict losses, but these black-box methods cannot explain the correlation between packet loss and parameters or mitigate losses. We propose a deep learning model to reconstruct lost packets in a complex networking scenario while integrating an explainable AI approach to explain the correlation between the networking parameters and the packet loss.. Integrating an elementary networking simulation designed in the ns2 platform, we collected data about networking packets and their associated parameters, based on which we trained and tested our deep learning model. Our approach was tested with 5-fold cross-validation, showing a mean accuracy of 79.09% for reconstructing the lost packets when maintaining a noticeable packet delivery fraction (PDF) rate of 98.9%, showing the effective performance of our proposed framework.
  Department: Computer Science
  Supervisor: Dr. Ahyoung Lee
   | 

• * GRP-071 Next-Generation DAPPs Development with Self-Service AI Agents (PhD Research) by 
  Abstract: Our research introduces a decentralized agent mesh architecture that transforms blockchain application development from fragmented human-driven processes to autonomous, systematized workflows through human-AI collaboration. We鈥檝e reimagined blockchain application development from the ground up by creating a decentralized agent ecosystem where humans and AI collaborate as peers rather than tools. Our innovation lies in the autonomous yet interconnected nature of specialized LLM powered agents handling contract creation, backend logic, frontend interfaces, and security auditing. Our proposed architecture distributes expertise across AI agents that operate in a peer-to-peer network. Furthermore, to address the emerging threat of systemic vulnerabilities from AI-generated code patterns, we鈥檝e integrated diverse verification methods that challenge and strengthen code before deployment.
  Department: Software Engineering or Game Design and Development
  Supervisor: Dr. Reza M. Parizi
   |  | 

• * GRP-077 MULISA: mHealth-enabled User-friendly Light-based Stroke Screening and Assessment in Pediatric Sickle Cell Disease in Uganda (PhD Research) by , 
  Abstract: This research presents a novel mHealth-enabled solution for stroke screening of sickle cell disease children in LMICs using light-based stroke screening technologies. We conducted a systematic literature review to identify key barriers and used these insights to develop a conceptual framework guiding the design of an integrated system. Our prototype includes a SWIR SCOS device and a wearable oximeter, combined with an AI-enhanced mHealth platform. The proposed mHealth framework aims to improve screening accessibility and adoption in low-resource settings.
  Department: Computer Science
  Supervisor: Dr. Nazmus Sakib, KSU Mentors: Dr. Paul Lee & Dr. Monica Swahn
   | 

• GRP-082 Ready Cluster One: Optimizing Film Success With Data Science (PhD Research) by Karim, Mohsin Md Abdul, Richardson, Joseph
  Abstract: This study presents a novel approach to predicting and optimizing screenplay investments by combining graph theory and finite mixture modeling (FMM) techniques. We construct a k-partite graph representing movies, genres, subgenres, production companies, directors, actors, and directors of photography, to explore the interconnectedness between these entities. Using FMM, we identify clusters within budget tiers, enabling a deeper understanding of how similar films perform based on their creative team and production characteristics. By balancing profit potential with risk-adjusted profit, the model suggests the most viable budget tiers for unproduced screenplays. This approach incorporates confidence intervals and evaluates the accuracy of budget tier recommendations, offering a data-driven solution for movie producers to make informed investment decisions.
  Department: Data Science and Analytics
  Supervisor: Dr. Bin Luo & Dr. Joseph DeMaio
   | 

• GRM-087 Empowering Alzheimer鈥檚 Caregivers: Designing Explainable and Personalized AI for Mental Health Support (Master's Research) by , Renduchintala, Chandra Rekha, 
  Abstract: This study presents the design of an AI-powered caregiver support app aimed at personalized mental health and burden management for individuals caring for Alzheimer鈥檚 patients. The design is grounded in insights drawn from a comprehensive analysis of 28 recent studies on AI-driven mental health interventions. These findings informed the implementation of key features, including machine learning models such as Random Forest, clustering, and supervised learning to create adaptive care plans tailored to patient and caregiver profiles. The system dynamically adjusts task schedules based on engagement data and provides interpretable recommendations through SHAP. With built-in emotional check-ins, mood tracking, and caregiver-centric interventions, the app promotes well-being while reducing the cognitive and emotional load often experienced in dementia caregiving.
  Department: Computer Science
  Supervisor: Dr. Nazmus Sakib
   | 

• * GRP-088 Nutrilyzer: A Vision-Based App for Macronutrient Estimation and Blood Glucose Response Prediction (PhD Research) by 
  Abstract: This study predicts postprandial glucose peaks and spike durations using 10-day multimodal data from 10 participants. Glucose, meals, workouts, and insulin doses were logged via the Nutrilyzer web app. Macronutrient content carbs, fats, and proteins was extracted using GPT-Vision, a highly accurate food analysis tool. These tuples were normalized to baseline glucose and aligned with a 3-hour window post-meal. Three models were tested: LSTM, Time Series Transformer, and ARIMA. LSTM performed best with 83.78% accuracy, followed by Transformer (71.43%) and ARIMA (62.41%). Results show the promise of AI-based food logging and time series modeling for personalized glucose forecasting.
  Department: Computer Science
  Supervisor: Dr. Maria Valero
   |  | 

• GRP-090 A Novel Superpixel鈥揜AG鈥揟ransformer Approach for Three-Class Melanoma Segmentation (PhD Research) by Ordonez, Pablo
  Abstract: Melanoma is one of the deadliest skin cancers, with early detection relying on accurately identifying both the lesion core and its often ambiguous border. Traditional CNN and U-Net models struggle with fuzzy transitions and irregular boundaries. We propose a three-class segmentation framework that labels regions as background, border, or lesion core. Our method over-segments images into superpixels, builds a Region Adjacency Graph (RAG) to capture spatial context, and generates embeddings using transformer-based autoencoders. This approach combines local image statistics with global semantic structure. Experiments on the HAM10000 dataset show improved precision and recall, especially in challenging border regions, outperforming CNN/U-Net baselines. Our results highlight the value of explicitly modeling boundaries for accurate and interpretable melanoma segmentation.
  Department: Computer Science
  Supervisor: Dr. Ying Xie
   | 

• * GRP-105 Prediction of Greenhouse Gas Emissions from Electric Vehicle Charging and Road Traffic in the United States (PhD Research) by , Kanigiri, Sai Nikhila
  Abstract: Electric vehicles (EVs) are widely considered a cleaner alternative to internal combustion engine vehicles. But their growing use creates indirect emissions via two main channels: more traffic congestion from more vehicle activity and more demand on power plants providing electricity for EV charging, usually depending on fossil fuels. This work offers a comprehensive, data-driven framework to forecast greenhouse gas (GHG) emissions connected to road traffic as well as EV-related power generation. Based on vehicle and speed characteristics, we forecast vehicle-level energy consumption and emission rates using a multi-model architecture that includes a Feed Forward Neural Network (FNN). While the Meta Prophet time series model is used to project power plant emissions under different energy demand conditions, macroscopic traffic flow models are used to estimate tract-level speed-density relationships. An Integrated Emission Model combines these elements to allow assessments particular to each area. Capturing vehicle mix, traffic dynamics, and energy grid variations, our study covers four major U.S. states鈥擟alifornia, Georgia, New York, and Washington. Results show notable spatial and temporal variations in emissions, therefore stressing the need of thorough models taking into account the intricate interactions between energy infrastructure, traffic patterns, and EV adoption.
  Department: Computer Science
  Supervisor: Dr. Mahyar Amirgholy
   | 

• GRP-134 Characterizing and Understanding the Performance of Small Language Models on Edge Devices (PhD Research) by 
  Abstract: In recent years, significant advancements in computing power, data richness, algorithmic development, and the growing demand for applications have catalyzed the rapid emergence and proliferation of large language models (LLMs) across various scenarios. Concurrently, factors such as computing resource limitations, cost considerations, real-time application requirements, task-specific customization, and privacy concerns have also driven the development and deployment of small language models (SLMs). Unlike extensively researched and widely deployed LLMs in the cloud, the performance of SLM workloads and their resource impact on edge environments remain poorly understood. More detailed studies will have to be carried out to understand the advantages, constraints, performances, and resource consumption in different settings of the edge. This project addresses this gap by comprehensively analyzing representative SLMs on edge platforms. Initially, we provide a summary of contemporary edge hardware and popular SLMs. Subsequently, we quantitatively evaluate several widely used SLMs, including TinyLlama, Phi-3, Llama-3, etc., on popular edge platforms such as Raspberry Pi, Nvidia Jetson Orin, and Mac mini. Our findings reveal that the interaction between different hardware and SLMs can significantly impact edge AI workloads while introducing non-negligible overhead. Our experiments demonstrate that variations in performance and resource usage might constrain the workload capabilities of specific models and their feasibility on edge platforms. Therefore, users must judiciously match appropriate hardware and models based on the requirements and characteristics of the edge environment to avoid performance bottlenecks and optimize the utility of edge computing capabilities.
  Department: Computer Science
  Supervisor: Dr. Kun Suo & Dr. Bobin Deng
   |  |