Computer Engineering Commons^™

Ai-Driven Biomarker Discovery & Progression Modeling For Precision Diagnosis Of Glaucoma, Cheng Huang

Computer Science and Engineering Theses and Dissertations

This dissertation presents a comprehensive study on the integration of artificial intelligence (AI) for glaucoma diagnosis and retinal image analysis. Leveraging multimodal imaging data including fundus photography, Optical Coherence Tomography Optical Coherence Tomography (OCT) and Optical Coherence Tomography Angiography (OCTA), the research develops a suite of deep learning frameworks designed to detect early glaucomatous changes with high precision, robustness, and interpretability. A series of novel architectures are introduced, spanning vessel segmentation networks, biomarker discovery pipelines, and multimodal fusion models, all designed to enhance diagnostic accuracy and generalizability across diverse populations. To facilitate reproducible and scalable ophthalmic AI research, this work …

En-Feat: An Effective Feature Selection Method Using Ensemble Approach, Sasank Nath, Dhruba Kumar Bhattacharyya

Mansoura Engineering Journal

Feature selection is a crucial step in machine learning and data preprocessing, significantly influencing model performance and interpretability. This paper presents a comprehensive study and contributions in the domain of feature selection by integrating traditional learning techniques with ensemble-based, proposing an effective approach. We propose a Mutual Information-based feature aggregation approach applied to union sets of features, aiming to derive an optimal subset of features that maximizes accuracy. Then, we employ an ensemble method that utilizes forward selection over union sets to identify the optimal feature subsets through sequential feature selection. Our ensemble-based feature selection method called En-feat, is evaluated …

Predictive Natural Language Metrics Of Alzheimer's Disease And Cognitive Decline Trend Analysis, Zerui Ma

Computer Science and Engineering Theses and Dissertations

Inspired by Dr. David Snowden's Nun Study, which linked early-life Propositional Idea Density (PID) to later-life Alzheimer's disease, this thesis investigates two questions: whether fine-tuned Transformer-based large language models (LLM) can detect cognitive decline from patient speech transcripts with meaningful feature attribution, and whether longitudinal PID trends are observable across large-scale internet and academic text corpora. We evaluate dementia prediction on the DementiaBank Pitt Corpus and conduct an exploratory longitudinal PID analysis across seven diverse datasets spanning up to 29 years and over 12.6 million documents. This work suggests that linguistic ability metrics, traditional PID metrics and novel LLM-based analysis, …

Machine Learning And Formal Methods In Quantum Chemistry: Theory And Application, Ishna Satyarth

Computer Science and Engineering Theses and Dissertations

In recent years, the progress in inter-disciplinary application of machine learning and artificial intelligence (ML/AI) have truly transformed various fields, from weather forecasting and drug development to medical diagnostics, energy, and sustainability. Computational chemistry uses computational tools to model, predict, analyze, and explain chemical phenomena, while the Quantum chemistry specifically uses techniques based on quantum mechanics (as opposed to classical mechanics or empirical models). Quantum chemistry or Computational chemistry has also observed a momentum in application of ML techniques over the past decade significantly accelerating results and providing valuable insights into vast datasets, often surpassing traditional methods.

This dissertation explores …

Green Technology: A Systematic Review Of Ai And Iot Solutions For A Sustainable Future, Nesma Abd El-Mawla, Mohamed A. Berbar, Nawal A. El-Fishawy, Mohamed A. El-Rashidy, Mahmoud Badawy

Mansoura Engineering Journal

Green technology offers a solution to the pressing environmental crisis. It can change the structure and generation of waste so as not to harm the earth, and people can become environmentally friendly. To address complex environmental challenges like climate change and pollution, innovative Artificial Intelligence (A.I.) and Internet of Things (IoT) solutions are essential. These technologies can help optimize resource use, reduce waste, and promote sustainable development. However, it's crucial to balance economic growth, social equity, and environmental protection when implementing green technologies. This survey paper systematically examines the landscape of Green Technology, focusing on its pivotal components: Measures of …

Mapping Leo Satellite Internet Performance Using Mobile Starlink Deployment, Annika Govil, Jacob Gray

Undergraduate Research Posters

High-speed, low-latency internet connectivity on the move is a critical challenge for applications in connected vehicles, disaster response, and remote education. While terrestrial networks such as 4G or 5G are widespread, they lack coverage in remote or rural geographic areas.

Low Earth Orbit (LEO) satellite constellations, such as SpaceX's Starlink, promise global high-bandwidth, low-latency internet. However, their performance is well-documented in stationary scenarios, while data for *mobile* applications is scarce. This project explores the feasibility and real-world performance of LEO satellite internet while in motion.

Current research documents the performance of LEO satellite constellations in stationary settings. However, data concerning …

Edge Guided Channel Attention In Fsrcnn: A Novel Approach For Depth Super Resolution, Yagneshkumar Jayantilal Parmar, Paresh M. Dholakia

Mansoura Engineering Journal

Depth images from low-cost sensors often suffer from blurred edges and structural distortions when processed with standard super-resolution models. While FSRCNN is efficient for RGB images, it struggles to handle the unique geometric requirements of depth maps. To solve this, we propose the Edge Guided Channel Attention FSRCNN (EGCA FSRCNN). This method incorporates an edge-guided modulation mechanism to preserve object boundaries and a Squeeze and Excitation (SE) block to focus on critical structural features. A major benefit of this framework is the use of frozen, pretrained FSRCNN weights, which bypasses the requirement for retraining. Our evaluation on the UTKinect, Middlebury, …

Green Ai-Enhanced Deep Learning Model For Breast Cancer Detection And Classification In Mammography Images: Bc-Net-512, Nesma Abd El-Mawla, Mohamed A. Berbar, Nawal A. El-Fishawy, Mohamed A. El-Rashidy

Mansoura Engineering Journal

This study champions a sustainable approach for developing a Deep Learning (DL) model for medical image analysis, specifically focusing on breast cancer (BC) detection in mammograms. By prioritizing low-computing algorithms to achieve high diagnostic accuracy while minimizing the model's environmental footprint, that aligns with the principles of Green AI. In this paper, an innovative architecture called BC-Net-512 was constructed for the classification of BC mammography. It is composed of lightweight Convolutional Neural Network (CNN) blocks for texture, density, and structure feature extraction and detection, a thin, fully connected layer for learning complex patterns and correlations in the extracted features, and …

Ai-Powered Accessibility Tracker For Inclusive Public Spaces, Yenny Ma, Kevin Beltran

College of Engineering Summer Undergraduate Research Program

This research project will develop and evaluate a smartphone-based, AI-powered system to crowdsource and analyze accessibility features and barriers in public spaces. Using computer vision and geospatial mapping, the system will identify and categorize issues such as uneven sidewalks, missing or inadequate curb ramps, damaged tactile paving, obstructive overhangs, and the absence of visual or auditory wayfinding cues. The overarching goal is to generate a dynamic, real-time accessibility map that empowers individuals with diverse mobility, sensory, and cognitive needs to navigate public spaces more safely and confidently. The project will integrate technologies and methods from applied machine learning, mobile computer …

Csc36000 - Modern Distributed Computing Assignment, Saptarashmi Bandyopadhyay

Open Educational Resources

This assignment covers standard performance metrics for Distributed Systems and the basics of Multiprocessing for CSC36000 - Modern Distributed Computing at the City College of New York CUNY. It is an interactive coding assignment intended to be executed in a Python notebook.

Sbv_Dps: A Stacking-Bagging-Voting Nested Ensemble Based Diabetes Prediction System Using K-Fold Cross Validation, Sourabh Shastri, Sachin Kumar, Paramjit Kour, Vibhakar Mansotra

Mansoura Engineering Journal

With the advancement of machine learning techniques, the introduction of the most accurate model has become a necessity. In real-world scenarios, every model has some constraints and assimilates errors, so their performance is not always highly efficient; this sparked the development of ensemble learning. The ensemble approach aims to consolidate the strengths of existing approaches and minimize their weaknesses or decision-making risks. The proposed diabetes prediction system encases a resampling filter, applied to balance the dataset and model builder method, i.e., without the SBV ensemble and with the SBV ensemble method. The model is initially built without using the SBV …

Meshlet Rendering Using D3d12 Mesh Shading Pipeline, Anishva Bardhan

Programming Theses and Dissertations

Modern video games must render scenes with increasingly complex geometry. Technologies like Nanite in Unreal Engine 5 enable the handling of scenes with significantly higher object and triangle counts than ever before. This project draws inspiration from Nanite by operating on triangle clusters, allowing artists to focus solely on creating high-poly meshes. The primary objective is to implement fine-grained culling techniques on meshlets, combined with efficient meshlet instancing, to reduce render time and memory usage.

Meshlet instancing plays a crucial role in optimizing rendering performance by allowing multiple objects sharing the same geometry to be rendered efficiently. Instead of duplicating …

3d Multi-Threaded Ai Navigation With Pathfinding And Obstacle Avoidance, Jabari Belgrave

Programming Theses and Dissertations

In this thesis, I developed a 3D multi-threaded AI navigation system using my own custom-built C++ game engine. The system combines triangle-based A* pathfinding with real-time obstacle avoidance using a set of velocity-obstacle algorithms. It is designed to support large numbers of agents navigating complex environments while avoiding collisions. I created two main simulation modes: Navigation Mode, which integrates A* with ORCA to handle large-scale pathfinding and movement, and Obstacle Avoidance Mode, which allows direct comparison between VO, RVO, HRVO, and ORCA in a controlled test setting.

The terrain is procedurally generated using Perlin noise, and this terrain data is …

Gpu-Based Visual Effects System, Matthew Jaffe

Programming Theses and Dissertations

The objective of my thesis is to create a robust and efficient VFX system that can be used to edit and add particle effects to games. This system utilizes a compute shading pipeline to simulate millions of particles in real time. The behavior of particles is widely customizable through many different properties which can be manipulated changed over the lifetime of particles and introduce procedural randomness. There are many ways to customize the motion of the particles with various forces and collision. Additionally, particles can be rendered as billboarded quads, full meshes or partial meshes with different settings to further …

Efficacy Of Immersive Virtual Reality Gameplay In Environmental Attitude Change: The Case Of Abandoned Offshore Oil Platforms In Santa Barbara, Arun Prasad Srinivasan Manoharan

Master's Theses

Public perception plays an important role in shaping conservation policies and decisions, especially in contested environmental spaces. Offshore oil platforms, historically viewed as environmental hazards, have been found to serve as marine habitats that support diverse marine life. However, public perception remains largely negative, influenced by concerns over pollution from past oil spill accidents. Traditional environmental education methods, such as lectures and documentaries, often fail to engage audiences effectively or shift entrenched opinions. This study explores the efficacy of immersive Virtual Reality (VR) gameplay in changing environmental attitudes, specifically in the context of abandoned offshore oil platforms in Santa Barbara, …

Development And Evaluation Of Machine Learning Models For Early Pediatric Sepsis Prediction, Ancita M. Andrade

Browse all Theses and Dissertations

Sepsis is a leading cause of pediatric mortality, claiming more lives in the United States annually than all childhood cancers combined. Early identification in Emergency Departments (EDs) remains challenging, as the current Phoenix criteria establishes an updated international consensus definition for sepsis, however is not designed for use as a screening tool. This study aimed to develop predictive models identifying pediatric patients at risk of sepsis within 24 hours of admission. Multiple tree-based and deep learning models were trained utilizing clinical and laboratory data from the initial four hours of presentation. Both the LightGBM and LSTM architectures demonstrated superior performance, …

Wearable Sensor Data Analysis For Machine Learning-Based Detection Of Posture And Autonomic Responses, Chaitanya Vardhini Anumula

Browse all Theses and Dissertations

This study investigates how Iyengar yoga postures influence autonomic nervous system (ANS) activity by analyzing multimodal physiological signals collected via wearable sensors. The physiological mechanisms underlying Iyengar yoga’s therapeutic effects remain under-explored at the granular, pose-level. Using data collected from 16 participants, this research evaluates whether machine learning models can distinguish between baseline, parasympathetic-dominant, and sympathetic-dominant states based on wrist-worn sensor data. The goals were to explore whether subtle postural variations elicit measurable autonomic responses and to identify which sensor features most effectively capture these changes. Participants performed a sequence of yoga poses while wearing synchronized sensors measuring electrodermal activity …

Impact Of Graph Structures For Rag Outcomes In Llms, Chris Davis Jaldi

Browse all Theses and Dissertations

Explainability, interpretability and adaptability (EIA) remain three central motivations for next-generation Artificial Intelligence (AI), especially as Large Language Models (LLMs) continue to engage with ever-increasing knowledge bodies. As the landscape pushes toward controllable agentic Retrieval-Augmented Generation (RAG) systems where AI agents engage in iterative, guided reasoning, a critical question arises as to the extent to which the knowledge design itself shapes these models' reasoning behavior. This work conducts a systematic evaluation of how different conceptualizations and representation of the identical knowledge affect an LLM's path-based reasoning capabilities. Through the introduction of controlled variations along graph structural complexity, linguistic and semantic …

Re-Parameterizing Adversarial Reprogramming In Low-Dimensional Subspace For Efficient Software Vulnerability Detection, Hootan Alavizadeh

Browse all Theses and Dissertations

Software vulnerabilities are a major cause of security breaches, making effective detection critical. Traditional learning-based methods require large datasets and significant computational resources, which are often impractical due to high annotation costs and data scarcity. To address this, we propose an innovative system, RearVul, which Re-parameterizes adversarial reprogramming in a low-dimensional subspace for software vulnerability detection. Unlike conventional approaches, RearVul repurposes a pre-trained classification model using adversarial reprogramming, enabling detection with minimal modifications. It learns a universal perturbation applied to program representations, preserving the original model’s feature extraction capabilities while adapting it to a new domain. Furthermore, we introduce a …

Subjective Readiness Forecasting Using Supervised Machine Learning And Wearable Device Data, Nathaniel Michael Weiland

Browse all Theses and Dissertations

Recent advances in wearable technology allow continuous monitoring of physiological and behavioral data, opening new opportunities for real-time assessments of readiness and well-being. However, creating predictive models that generalize across diverse users remains challenging, especially in high-stakes settings like the military, where preventable injuries, illnesses, and stress-related performance declines are frequent. This research assesses the feasibility of using supervised machine learning models trained on wearable device data to predict subjective readiness indicators—recovery, stress, injury, and illness. Data from over 10,000 users in the OHWS (Optimizing the Human Weapons System) program combined daily check ins with physiological metrics from Garmin, Polar, …

Isometric Centroid Encoder (Ice) And Synthetic Data Generation Approaches For Biological Datasets, Prathyusha Kanakamalla

Browse all Theses and Dissertations

This thesis addressed two main challenges in biological data analysis: structure-preserving dimensionality reduction and synthetic data generation for small sample datasets. I proposed the Isometric Centroid Encoder (ICE), a supervised dimensionality reduction method that preserves pairwise distances between class centroids during dimension reduction. Unlike existing methods like Centroid Encoder and Super Encoder, ICE explicitly maintains geometric relationships between biological classes, achieving nearly perfect structure preservation at C dimensions (where C equals the number of classes) with strong performance even in 2D and 3D spaces. Additionally, I compared three generative models (VAE, LSH-GAN, and scDiffusion) for synthetic data generation on small …

Leveraging Counterfactuals For Enhanced Natural Language Inference: A Multitask Knowledge Distillation Approach, Brian J. Davis

Browse all Theses and Dissertations

Natural-language inference (NLI) asks whether a hypothesis is entailed by, contradicts, or is neutral with respect to a premise. Modern transformers reach high raw accuracy on benchmarks such as SNLI, MNLI, and ANLI, yet they often rely on brittle lexical shortcuts and provide little insight into their decision process. This thesis shows that counterfactual-augmented knowledge distillation can simultaneously boost robustness and supply faithful, token-level explanations—without scaling model size. Four T5-v1_1 students (60M, 220M, 770M, 3B parameters) are trained under four curricula: (1) standard fine-tuning, (2) fine-tuning with free-text rationales, (3) multi-task distillation with naive counterfactuals, and (4) multi-task distillation with …

Patient Subset Classification Using Encoded Embeddings And Knowledge Graph Retrieval-Augmented Generation, Benjamin A. Holmes

Browse all Theses and Dissertations

The widespread adoption of electronic medical records has created a vast reservoir of clinical data that can be leveraged to better understand how interventions relate to patient outcomes. Much of this information, however, exists as unstructured free-text, posing significant challenges for traditional statistical and machine-learning methods. Solving these challenges would allow the extraction of specific patient subpopulations (clinically relevant cohorts of individuals who share overlapping symptoms, risk factors, or diagnostic criteria), which could be used in precision medicine. Despite this promise, extracting these subpopulations from unstructured medical notes is an ongoing challenge due to the variability of clinical language and …

Generative Adversarial Networks (Gans) For High-Dimensional Biological Data, Harigovind Harikumar

Browse all Theses and Dissertations

This thesis investigates the application of Generative AI models, mainly Generative Adversarial Network (GAN) models to high dimensional and low sample size biological datasets like Motion Sickness, Breast Cancer, Crohn, and Melanoma. We utilized and compared three generative AI frameworks: Vanilla GAN, Wasserstein GAN (WGAN), Locality-Sensitive Hashing GAN (LSH-GAN) and Omics GAN. To address the challenges associated with high-dimensionality and low sample size, which was leading to very poor outputs of biological synthetic samples, we came up with an approach to stop the model when it reaches its saturation level. That is, we printed the loss plots to see where …

Learning Under Data Scarcity: Reasoning And Negative Distillation For Texts And Graphs, Calvin T. Greenewald

Browse all Theses and Dissertations

Modern machine learning (ML) models rely on large amounts of high-quality labeled data to achieve optimal performance. However, in many real-world domains, such as cyber security, acquiring sufficient labeled data is often infeasible due to cost, privacy concerns, and the rapid evolution of underlying phenomena. This challenge underscores the importance of learning under data scarcity. This thesis addresses this challenge by proposing distinct, modality-specific techniques for text and graph domains, which allow models to generalize effectively with minimal data. For text classification task, we incorporate distilled rationales from large language models and adversarial perturbations into the input space to improve …

Synthetic Network Creation And Visualization, Kaylee Sloat, Jeremy Evert

Student Research

The code in the repository, “synthetic_network_creation_and_visualization” is inspired by the foundational work presented in "NetProbe: A Fast and Scalable System for Fraud Detection in Online Auction Networks" by Shashank Pandit, Duen Horng Chau, Samuel Wang, and Christos Faloutsos.

Link to the original paper:https://kilthub.cmu.edu...

Real Time Pii Scanning, John David

Electronic Theses and Dissertations

The increased amount of web applications and internet software solutions utilizing cloud frameworks has contributed to large data sets of system log messages being generated constantly. These messages may contain sensitive data, creating an additional security risk for the systems and contributing to the need for analysis of such large volumes of data in real time. Large commercial data monitoring systems can solve for these analysis requirements, but they can be costly. We present a solution to analyzing web application log data which ingests it, processes it and visualizes sensitive data found within in real time. Our solution utilizes an …

Heterogeneous Multi-Robot Person-Following In Constrained Environments, Ori A. Miller

Electronic Theses and Dissertations

Maintaining visibility of a person requires effective systems. Security cameras or ground robots might be ideal, but they often fail in uncontrolled or unknown environments. A single ground robot struggles to navigate and track an agent at the same time. This work addresses the challenge by developing a multi-robot system with a slow ground robot and an agile aerial robot. Three methods are evaluated: FORWARD-PF, RL-Person Following (RL), and a baseline closed-loop method. FORWARD-PF proved the most reliable, completing all nine paths and reaching targets nearly twice as fast as RL. Despite completing seven paths, RL faltered on complex tasks. …

Integrating Authentication Schemes In Augmented And Virtual Reality Classrooms, Naheem Noah

Electronic Theses and Dissertations

Augmented Reality and Virtual Reality (AR/VR) technologies are revolutionizing educational experiences, but their widespread adoption hinges on addressing critical security and usability challenges, particularly in the domain of user authentication. This research presents an investigation into the security landscape of AR/VR and explores a graphical authentication scheme called “Things” that enhances both security and usability in immersive learning environments. Through a systematic evaluation of popular AR/VR devices and applications, potential vulnerabilities and limitations were identified, such as high usage of pin/passwords which are susceptible to shoulder-surfing attacks, lack of multi-factor authentication, and unclear data-sharing practices. A review of existing knowledge-based …

Activity Map Generation And Event-Based Sensor Processing With Spiking Autoencoders And Sparse Dictionary Learning, Jack Easton

Computer Science and Engineering Theses and Dissertations

This thesis explores the potential of Spiking Neural Networks (SNNs) in processing event sensor data and generating high-fidelity activity maps. Event sensors capture asynchronous binary events with high dynamic range, but traditional processing methods often fail to leverage their advantages fully. SNNs, with their asynchronous, event-driven nature, offer a promising alternative.

A Spiking Autoencoder (SAE) was employed in this thesis to handle the stochastic and sparse event data, integrating deep dictionary learning to enhance the feature space and improve activity map quality. The encoder, modeled after the VGG network, extracts features from event streams generated by speckle patterns, which are …