Open Access. Powered by Scholars. Published by Universities.®

Computer Sciences Commons™

Open Access. Powered by Scholars. Published by Universities.®

2025

Discipline
Institution
Keyword
Publication
Publication Type
File Type

Articles 2101 - 2130 of 3497

Full-Text Articles in Computer Sciences

Grp-010 Autonomous Agents In The Loop: Strengthening Educational Recommenders With Computer Use Agents, Mourya Teja Kunuku Apr 2025

Grp-010 Autonomous Agents In The Loop: Strengthening Educational Recommenders With Computer Use Agents, Mourya Teja Kunuku

C-Day Computing Showcase

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—using browser automation, mouse control, …

Go to article

Grp-053 Expand: Explainable Ai Integrated Deep Learning-Based Reconstruction Of The Lost Packets​, Nasim Ahmed, A E M Ridwan Apr 2025

Grp-053 Expand: Explainable Ai Integrated Deep Learning-Based Reconstruction Of The Lost Packets​, Nasim Ahmed, A E M Ridwan

C-Day Computing Showcase

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 …

Go to article

Grp-071 Next-Generation Dapps Development With Self-Service Ai Agents, Viraaji Mothukuri Apr 2025

Grp-071 Next-Generation Dapps Development With Self-Service Ai Agents, Viraaji Mothukuri

C-Day Computing Showcase

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’ve 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, …

Go to article

Grp-090 A Novel Superpixel–Rag–Transformer Approach For Three-Class Melanoma Segmentation, Pablo Ordonez Apr 2025

Grp-090 A Novel Superpixel–Rag–Transformer Approach For Three-Class Melanoma Segmentation, Pablo Ordonez

C-Day Computing Showcase

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 …

Go to article

Grp-105 Prediction Of Greenhouse Gas Emissions From Electric Vehicle Charging And Road Traffic In The United States, Faysal Chowdhoury, Sai Nikhila Kanigiri Apr 2025

Grp-105 Prediction Of Greenhouse Gas Emissions From Electric Vehicle Charging And Road Traffic In The United States, Faysal Chowdhoury, Sai Nikhila Kanigiri

C-Day Computing Showcase

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 …

Go to article

Grp-134 Characterizing And Understanding The Performance Of Small Language Models On Edge Devices, Md Romyull Islam Apr 2025

Grp-134 Characterizing And Understanding The Performance Of Small Language Models On Edge Devices, Md Romyull Islam

C-Day Computing Showcase

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, …

Go to article

Uc-016 Multifamily Loan Performance, Jonathan Bell Apr 2025

Uc-016 Multifamily Loan Performance, Jonathan Bell

C-Day Computing Showcase

This study explores the performance of multifamily loans using a logistic regression model to predict loan outcomes as either “closed” or “current”. Utilizing a dataset of over one million observations and 54,771 unique loan observations, we classify loan status based on Freddie Mac’s 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 …

Go to article

Uc-019 Gwinnett County Public Schools - Data Masking Tool, Emmett Peters, Diego Frausto Ramirez, Abhay Tompkins Silas Talele, Jesus Trejo-Tamayo Apr 2025

Uc-019 Gwinnett County Public Schools - Data Masking Tool, Emmett Peters, Diego Frausto Ramirez, Abhay Tompkins Silas Talele, Jesus Trejo-Tamayo

C-Day Computing Showcase

In today’s 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 …

Go to article

Uc-020 Indy Micro - Virtual 8-Bit Computer, Matthew Watson, Elizabeth King, Junhyeok Lee, Gavin Frey Apr 2025

Uc-020 Indy Micro - Virtual 8-Bit Computer, Matthew Watson, Elizabeth King, Junhyeok Lee, Gavin Frey

C-Day Computing Showcase

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 …

Go to article

Uc-027 Ksublocks Tower Defense, Matthew Elledge, Annagrace Gwee, Bryan Nguyen, Logan Slicker, Ashley Ahn Apr 2025

Uc-027 Ksublocks Tower Defense, Matthew Elledge, Annagrace Gwee, Bryan Nguyen, Logan Slicker, Ashley Ahn

C-Day Computing Showcase

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.

Go to article

Uc-030 Heartspeak​ Ai, Nabeel Faridi, Shammah Charles, Isha Minocha, Ethan Barnes, Aravind Iyer Apr 2025

Uc-030 Heartspeak​ Ai, Nabeel Faridi, Shammah Charles, Isha Minocha, Ethan Barnes, Aravind Iyer

C-Day Computing Showcase

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.

Go to article

Uc-037 Dynamic Requirements For A Software Training Environment​, Cassidie Grogan, Jesus Valdez, Dawson White Apr 2025

Uc-037 Dynamic Requirements For A Software Training Environment​, Cassidie Grogan, Jesus Valdez, Dawson White

C-Day Computing Showcase

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 …

Go to article

Uc-048 Dinengo - Ai Genie, John Sheffield, Juwon Atunnise, Jaz Ankrah, Alex Colas, Kayla Grant Apr 2025

Uc-048 Dinengo - Ai Genie, John Sheffield, Juwon Atunnise, Jaz Ankrah, Alex Colas, Kayla Grant

C-Day Computing Showcase

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.

Go to article

Uc-101 Sight-Singing Feedback, Terah Dann, Sandy Giroux, Nathanael Johnson, Amber Scarbro Apr 2025

Uc-101 Sight-Singing Feedback, Terah Dann, Sandy Giroux, Nathanael Johnson, Amber Scarbro

C-Day Computing Showcase

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.

Go to article

Uc-111 Accessible Interactive Map​, Justin Connick, Megan Ingram, Derrick Novak, Spencer Williams, Emily Zhu Apr 2025

Uc-111 Accessible Interactive Map​, Justin Connick, Megan Ingram, Derrick Novak, Spencer Williams, Emily Zhu

C-Day Computing Showcase

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.

Go to article

Uc-136 Foster Ai Interview And Biography Generation, Joshua Crawford, Katlin Lahr, Mikayla Haigh, Brittany Payne, Daria Morhun Apr 2025

Uc-136 Foster Ai Interview And Biography Generation, Joshua Crawford, Katlin Lahr, Mikayla Haigh, Brittany Payne, Daria Morhun

C-Day Computing Showcase

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 …

Go to article

Uc-125 Database Masking Tool - Project 04 - Team 1, Lleyton Callison, Josh Tettey-Enyo, Reda Salimi, Stephen Sigmon, Alec Quillen Apr 2025

Uc-125 Database Masking Tool - Project 04 - Team 1, Lleyton Callison, Josh Tettey-Enyo, Reda Salimi, Stephen Sigmon, Alec Quillen

C-Day Computing Showcase

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—Faker-based masking, hash masking, and pseudonymization through generalized masking—to 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. …

Go to article

Ur-002 Fedda-Tsformer: Federated Domain Adaptation With Vision Timesformer For Left Ventricle Segmentation On Gated Myocardial Perfusion Spect Image, Yehong Huang Apr 2025

Ur-002 Fedda-Tsformer: Federated Domain Adaptation With Vision Timesformer For Left Ventricle Segmentation On Gated Myocardial Perfusion Spect Image, Yehong Huang

C-Day Computing Showcase

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 …

Go to article

Ur-044 Quantum Machine Learning For Science And Engineering, Caleb Dow, Daniel Tebor, Dante Lewis Apr 2025

Ur-044 Quantum Machine Learning For Science And Engineering, Caleb Dow, Daniel Tebor, Dante Lewis

C-Day Computing Showcase

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.

Go to article

Ur-031 Impact Of Motor Skill On Learning Experiences And Outcomes Using Note-Taking In Vr, Sawyer Strickland Apr 2025

Ur-031 Impact Of Motor Skill On Learning Experiences And Outcomes Using Note-Taking In Vr, Sawyer Strickland

C-Day Computing Showcase

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.

Go to article

Ur-115 Mobinav: Accessible Campus Navigation, Eric Legostaev, Damien Castro, Dom Evans Apr 2025

Ur-115 Mobinav: Accessible Campus Navigation, Eric Legostaev, Damien Castro, Dom Evans

C-Day Computing Showcase

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 Kennesaw State University's Marietta campus, it is designed for scalability to other locations.

Go to article

Ur-126 Multimodal Neuroimaging Meets Ai: Enhancing Alzheimer's Diagnosis With Pyradiomics, Dina Xu Callaway, Maya Castillo, Richard Haynes Apr 2025

Ur-126 Multimodal Neuroimaging Meets Ai: Enhancing Alzheimer's Diagnosis With Pyradiomics, Dina Xu Callaway, Maya Castillo, Richard Haynes

C-Day Computing Showcase

Alzheimer’s 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’s 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 …

Go to article

Uc-107 Draw The Night Sky, Dominic Ho, Richard Deas, Monica Phillips, Gabe Strong, Conner Hartsfield Apr 2025

Uc-107 Draw The Night Sky, Dominic Ho, Richard Deas, Monica Phillips, Gabe Strong, Conner Hartsfield

C-Day Computing Showcase

Draw The Night Sky is a game project made in collaboration with Carter’s 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 …

Go to article

Gc-033 Oncoclarify – Ai Powered Cancer Report Simplifier, Sai Chandana Koganti Apr 2025

Gc-033 Oncoclarify – Ai Powered Cancer Report Simplifier, Sai Chandana Koganti

C-Day Computing Showcase

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.

Go to article

Gc-039 Clinicpix: Secure Medical Image Sharing Web Application, Michael Harris, Selorm Kumi, Zhi Ern Tan, James Hodgson Apr 2025

Gc-039 Clinicpix: Secure Medical Image Sharing Web Application, Michael Harris, Selorm Kumi, Zhi Ern Tan, James Hodgson

C-Day Computing Showcase

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.

Go to article

Gc-059 Large-Scale Cybersecurity Threat Detection, Pavan Chowdary Chilukuri, Mohan Krishna Kandimalla Triveni Thiriveedhi, Raghava Sammeta, Venkata Basanth Challapalli, Triveni Kandimalla Apr 2025

Gc-059 Large-Scale Cybersecurity Threat Detection, Pavan Chowdary Chilukuri, Mohan Krishna Kandimalla Triveni Thiriveedhi, Raghava Sammeta, Venkata Basanth Challapalli, Triveni Kandimalla

C-Day Computing Showcase

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, Naïve 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’s distributed computing, the system ensures efficient real-time analysis of vast network data. The research supports …

Go to article

Gc-089 Safecircle:​ Ai And Micro-Radar-Based Remote Monitoring For Patients With Ad/Adrd​, Awan-Ur Rahman, Shakib Quddus, Soarov Chakra Borty Apr 2025

Gc-089 Safecircle:​ Ai And Micro-Radar-Based Remote Monitoring For Patients With Ad/Adrd​, Awan-Ur Rahman, Shakib Quddus, Soarov Chakra Borty

C-Day Computing Showcase

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.

Go to article

Gc-128 Multi-Label Commit Message Classification Using P-Tuning, Tanvi Mistry Apr 2025

Gc-128 Multi-Label Commit Message Classification Using P-Tuning, Tanvi Mistry

C-Day Computing Showcase

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.

Go to article

Grm-012 (Tcc) Transformer Embedded Synthetic Source Code Multiclass Classification, Rene Lisasi, Patrick Wu Apr 2025

Grm-012 (Tcc) Transformer Embedded Synthetic Source Code Multiclass Classification, Rene Lisasi, Patrick Wu

C-Day Computing Showcase

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 …

Go to article

Grm-038 Optimizing Prompts For Alzheimer's Speech Classification Using Llm, Imaan Shahid Apr 2025

Grm-038 Optimizing Prompts For Alzheimer's Speech Classification Using Llm, Imaan Shahid

C-Day Computing Showcase

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’s 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 …

Go to article

First Prev Next Last