24067 research outputs found
Sort by
GRP-0197 Evaluating the Ability of LLMs to Interpret, Optimize and Translate LLVM IR
No full textThis study investigates whether modern state-of-the-art Large Language Models (LLMs) can interpret, optimize, and translate low-level intermediate representations (IR) used in compilers and binary translation software. We evaluate LLM performance on LLVM IR across three tasks: 1) interpreting the underlying algorithmic behavior, 2) identifying missed optimization opportunities, and generating improved IR variants, 3) Translating IR between AArch64 and x86-64 targets. To ensure correctness, all LLM generated IR is checked using a validation pipeline that verifies its syntax, structural correctness, compiles it into machine code, and executes it on randomized test data. Early results show that LLMs can perform non-trivial IR transformations on top of existing LLVM-O3 optimizations, highlighting their potential role in future compiler and binary translation workflows
GRP-0165 Reinforcement Learning for Latency-Aware Priority Boosting in Linux Completely Fair Scheduler
No full textTail latency remains a persistent challenge in Linux’s Completely Fair Scheduler (CFS), particularly when short, latency-sensitive jobs compete with longer ones. Traditional boosting heuristics improve tail latency but require manual tuning and generalize poorly across workloads. This project evaluates whether a reinforcement-learning (RL) controller can dynamically apply priority boosts more effectively than fixed heuristics. Using a discrete-event Python simulator modeled after CFS, this project compares baseline CFS, heuristic boosting, and PPO-based learned boosting under mixed workloads. Results show that while heuristics achieve the lowest absolute latency, RL achieves competitive tail-latency reduction with significantly better fairness and adaptability. A state-space study further analyzes how observation richness affects RL performance
GRM-1252 How Humans Perceive Mobile Robots: Anxiety, Environment, And Behavior Analysis
No full textWe conducted 6,400 physics-based simulations to examine how environmental and behavioral factors shape human anxiety during interactions with mobile robots. The model incorporated robot behavior, environmental density, visibility, indoor/outdoor settings, and human age. Anxiety was driven primarily by context: levels were highest indoors, during daytime, and in sparse environments, while nighttime and outdoor interactions consistently reduced anxiety. Robot behavior produced smaller effects, with erratic and non-avoidance strategies yielding slightly higher responses. Older adults showed marginally greater anxiety across all conditions. These findings suggest that environmental design and deployment context matter more than avoidance strategy, offering guidance for improving the safety and acceptance of autonomous mobile robots in public spaces
GRM-1245 A Synthetic Data Engine for Explainable Injection-Area Perception
No full textVision-Language-Action (VLA) systems are beginning to support everyday clinical workflows. Deltoid intramuscular injection is a representative task, but progress is limited by data scarcity, privacy constraints, and the cost of expert annotation. Recent text-to-image (T2I) models make large-scale data synthesis possible, yet ensuring anatomical correctness, diversity, and label quality remains difficult. To address this gap, we propose a Synthetic Data Engine tailored for medical perception, integrating cold-start filtering, controlled T2I generation, CLIP-based quality checks, and iterative segmentation training. We further introduce an anthropometry-grounded formulation of injection safety that produces interpretable safe-zone guidance. Experiments show that synthetic data can effectively bootstrap deltoid-segmentation performance and support reliable injection-area perception
GC-1198 Onboarding Tool for New Smartphone Users
No full textThe Smartphone Onboarding Tool is an interactive web platform created to help seniors and new smartphone users become comfortable with mobile technology. It offers a realistic, simulated smartphone interface, guided walkthroughs, and an easy-to-use design that builds confidence in performing everyday tasks. Caregivers can monitor user progress, while learners can practice safely without affecting an actual device. The solution is developed with a React frontend, a Node.js/Express backend, and an SQLite database, all built with a strong focus on mobile-first accessibility
GC-1154 Peer Evaluation Automation and Feedback System
No full textA web-based platform to streamline peer evaluations in team-based courses. Professors can securely create and manage student rosters, assign students to courses/teams, trigger email invitations, and receive structured, professor- friendly reports with both numeric and textual feedback. Optional AI features may summarize comments and flag potential concerns, depending on timeline and scope
GC-1146 Student Engagement Portal: Enhancing Student Success through Milestone Tracking
No full textThe Student Engagement Portal, also known as the Milestone Map, is a platform developed to help students within KSU’s College of Computing and Software Engineering monitor their academic and professional growth. The system enables students to log milestones, check in at events, and view progress toward personal and departmental goals. Built with a full-stack architecture using NestJS, React, and MongoDB, the portal also includes an administrative dashboard for event management and analytics. The project demonstrates how progress tracking and clear visualization of achievements can improve communication, organization, and engagement between students and the college
GC-0270 OncoBoost - Hydration Monitoring Application
No full textDehydration is a common and preventable complication for oncology patients, especially those undergoing chemotherapy and radiation. Side effects such as nausea, fatigue, and loss of appetite make it difficult for patients to maintain adequate fluid intake, contributing to avoidable discomfort and potential treatment disruptions. This capstone project presents Onco-Boost, a mobile hydration monitoring application designed to help adult oncology patients track daily fluid intake, recognize their intake patterns, and stay engaged in daily self-care between clinic visits. Built with React Native and Expo, and backed by Firebase for authentication and cloud data storage. Onco-Boost translates clinical hydration guidance and research into an intuitive, patient-friendly interface. The app features a streamlined home page for quick-tap beverage logging with 4 beverage options, a history page that displays previous entries and daily totals, and a profile page where users can set or adjust basic information. The system is designed with simple data models, role-appropriate access, and secure storage to support future integration with clinical workflows if desired. Onco-Boost demonstrates how a focused, data-driven mobile tool can support oncology patients in maintaining hydration and may reduce preventable complications related to low fluid intake
UC-1222 Active Learning System for Labeling Chest X-rays
No full textThis project aims to develop a complete Active Learning System for chest X-ray image classification, designed to automate data preparation, streamline model training, and reduce the manual effort required for medical image labeling. The system establishes a structured and scalable pipeline that moves from raw data ingestion to automated decision-making, incorporating dataset indexing, patient-aware splitting, preprocessing, configuration management, and validation to ensure data flows reliably through the system. The model component uses CNNs to generate baseline diagnostic predictions across chest pathologies. Active learning strategies are then applied to identify the most informative unlabeled images, enabling iterative retraining that improves model performance while minimizing labeling cost. Proposed strategies for training improvements include transfer learning with a pretrained model. Evaluation tools such as a dashboard to track performance would also enable a reproducible, clinically relevant image workflow. The final system would deliver a reproducible framework capable of managing large medical imaging datasets, selecting high-value samples for annotation, and continuously refining classification accuracy over time
University Band & Wind Symphony present: “Of Song and Dance”
No full textDavid Roush, Conductorhttps://digitalcommons.kennesaw.edu/musicprograms/2962/thumbnail.jp