Computer Engineering Commons^™

Culturally Inclusive Usability Engineering (Ciue): A Framework For Evaluating Cultural Inclusion In Study Abroad Platforms, Louis Muhammad

Master's Theses

Study abroad programs provide students with opportunities to develop global com- petence and intercultural understanding. However, the digital platforms that support these programs often fail to communicate information that reflects the cultural and faith-based needs of diverse student populations. While prior research in Human-Computer Interac- tion (HCI) and usability engineering has explored cross-cultural and accessible design, cultural and religious considerations are rarely integrated into usability engineering pro- cesses. This thesis introduces the Culturally Inclusive Usability Engineering (CIUE) frame- work, which extends traditional usability engineering by integrating culturally informed con- siderations into the system development lifecycle. CIUE begins by identifying the …

An Investigation Of Data Granularity In Rag Pipelines For Personalized Medicine, Paritosh Pandey

Master's Theses

Generative AI, exemplified by large language models like the OpenAI GPT and Meta LLaMA families, can produce diverse content in response to prompts. This capability offers a promising solution to challenges in precision medicine, which seeks to tailor treatments to individual clinical profiles but often struggles with data collection, cost, and privacy concerns. By generating realistic, privacy-preserving patient data, generative AI has the potential to transform patient-centric healthcare. With such motivation, this research develops a comprehensive Generative AI pipeline emphasizing data granularity for accurate prediction of personalized treatments. The pipeline features a central Large Language Model interacting with a Machine …

Demystifying Hardware Formal Verification For Undergraduate Education: A Risc-V Processor Case Study With Coursework Implementation, Riley A. Peters

Master's Theses

Hardware verification engineers apply formal methods to prove that a digital device always behaves according to its specification. This differs from traditional functional verification, in which engineers establish correctness by repeatedly sending test inputs to the device and comparing the outputs against a reference model. With the growing complexity of integrated circuits, the demand for digital verification engineers with formal methods experience has continued to increase. However, California Polytechnic State University: San Luis Obispo's current curriculum lacks dedicated material to prepare students for these roles.

This thesis seeks to address the lack of formal methods material through two efforts. First, …

Exploring Math Word Problem Generation With Llms, Trung Hieu Vuong

Master's Theses

Math Word Problem (MWP) is an important building block for learning math. This type of problem is particularly useful for younger audiences because solving it involves two simultaneous skill sets: reading comprehension and mathematical reasoning. With publicly available large language models (LLMs), generating additional MWPs is readily achievable. While researchers have started using LLMs as MWP facilitators, there still exists a gap in studies about the diversity of MWPs generated by unmodified, publicly accessible LLMs. For that reason, our study focused on two goals: (1) to evaluate the diversity of MWPs generated by publicly available LLMs when provided with examples …

Developing A Vietnamese Text Summarization Large Language Model On Limited Hardware, Tin Pho

Master's Theses

Text summarization models have achieved significant growth during the last few years because of major Large Language Model (LLM) technological advancements. The application of LLMs are widely used in news distribution (TL;DR news), translation tools (DeepL Translate), or virtual assistants (ChatGPT, DeepSeek, Claude, etc.). However, the progress has not yet reached all languages equally. The Vietnamese language is used by more than 90 million people, but the language is not as highly developed for LLM as it has many homophones, five different tones that affect meaning of words, and irregular grammar compared to other languages (e.g. English, Spanish, etc.). Also, …

Automatic Guitar Transcription Of Polyphonic Music, Ritwik Patil

Master's Theses

Transcribing guitar music automatically is a complex task due to polyphonic overlap, tuning variations, and diverse playing techniques. Current transcription systems focus on identifying note pitches and timing while ignoring performance techniques that describe how the notes are played, treating guitar recordings as generic polyphonic audio and producing MIDI-like outputs that lose important information about articulation and style. To address these challenges, we propose an end-to-end transformer model for automatic guitar transcription. The system uses a T5-based encoder-decoder architecture that processes the Constant-Q Transform (CQT) of stereo audio input. The stereo representation helps separate individual guitar parts within a mix …

Latent Action Trajectory Optimization, Rahul Milind Kandekar

Master's Theses

Learning from demonstrations offers a path to bypass the sample inefficiency of reinforcement learning, but obtaining action-labeled expert demonstrations remains expensive and often impractical. Learning from Observations (LFO) addresses this by learning policies from observation-only demonstrations. Recent LFO work relies heavily on behavior cloning: VPT and LAPO use observation-only data combined with limited action labels to train BC policies, while AIME offers an alternative policy inference approach but requires the majority of its training data to have action labels. Through systematic experiments in the Lunar Lander environment, we investigate whether latent action methods can function when state and action dimensionalities …

Parameter-Efficient Multimodal Adaptation: Ocr-Integrated Lora For Textvqa And Captioning, Karthik Ganesh Malini

Master's Theses

Vision-Language Models (VLMs) have emerged as transformative technologies for multimodal AI, yet they face significant hurdles in processing text-rich images required for enterprise applications like document understanding, medical imaging, and industrial inspection. Current VLMs struggle with accurate text extraction and reasoning, often exhibiting high hallucination rates and poor Optical Character Recognition (OCR) token utilization. To address these limitations, this research presents a comprehensive framework for optimizing parameter-efficient Low-Rank Adaptation (LoRA) fine-tuning strategies on state-of-the-art architectures, including LLaVA-1.5 and BLIVA-FlanT5. Our methodology integrates enhanced OCR token utilization, faithful caption generation, and specific hallucination mitigation techniques. We employ a multi-dimensional evaluation protocol …

Building A Data-Driven Security Ai Framework Using Machine Learning Models, Christopher Chan Vi

Master's Theses

This research project explores a modern approach to Intrusion Detection System (IDS) anomaly detection by leveraging Artificial Intelligence (AI), Machine Learning (ML), Large Language Models (LLM), and Explainable AI (XAI). The purpose is to assess the effectiveness of these technologies in enhancing the understanding of intrusion events. This research study addresses the challenges of threshold determination and the interpretability of anomaly detection results. The proposed solution involves an LLM-based framework with XAI capabilities, integrated with a Retrieval Augmented Generation (RAG) architecture, to provide clear explanations for detected anomalies, utilizing both custom and pre-trained datasets. The study navigated inherent challenges, including …

Hallucination Techniques For Self-Supervised Synthetic Datasets For Mobile Robots, Wyatt D. Colburn

Master's Theses

Classical techniques in autonomous navigation struggle in tightly constrained spaces. Machine learning has been shown to perform better in these difficult environments but most techniques require large amounts of navigation experience for training. Using a new machine learning paradigm learning from hallucination (LfH), training data can be collected in a safe environment and not require supervision. Data is collected in real time while an agent performs a random walk in free space, supervision is not required as there are no obstacles for the robot to run into. After a random walk a post processing pipeline will hallucinate a safety corridor …

Enhancing Ad/Adrd Management Through Ihelpcare: A Compliant And Culturally Sensitive Ai-Driven Digital Healthcare Platform, Trisha Bhowmick

Master's Theses

The digital healthcare field is expanding fast, and now it requires platforms that use advanced technology and maintain robust data security and compliance practices. In the present paper, we present the main structure, key methods, and compliance strategies of the digital healthcare system iHelpCare, which, while fully meeting the HIPAA/GDPR requirements, provides health services more accessible, efficient, and inclusive. The proposed platform is powered by AI for personalized care solutions, with the main emphasis on preventive health management and providing tools for people with disabilities.

iHelpCare achieves real-time patient monitoring while securing medical data management and easy communication between patients, …

Llm-Powered Question Answering For Object States In Virtual Reality, Shiyi Ding

Master's Theses

Recent advances in large language models (LLMs) and multimodal large language models (MLLMs) enable natural language–based querying in virtual reality (VR). However, VR environments are highly localized, personalized, and dynamic, making it challenging for general-purpose models to answer environment-specific queries or reason about subtle object state changes. To address these challenges, this thesis develops two systems for 3D question answering in VR.

First, we present RAG-VR, the first retrieval-augmented 3D question-answering system designed for VR. RAG-VR augments an LLM with external knowledge retrieved from a localized knowledge database and includes a pipeline for extracting environmental and user-related information. To improve …

Retrieval Augmented Framework For Deepfake Audio Detection, Avinash Saxena

Master's Theses

The widespread use of AI-based audio deepfakes threatens severely to undermine media integrity and public trust. Speech synthesis techniques have improved dramatically in voice conversion (VC) and text-to-speech (TTS) in recent years, making forgeries sound highly realistic, and concerns are raised about possible malevolent uses. Existing state-of-the-art techniques for identifying fake speech have proven to be effective in some cases but are still limited in application and robustness when faced with novel attacking strategies, different acoustic conditions, or alternative linguistic domains. To address some of these limitations, the current research presents a novel deepfake audio detection system based on personalized …

Towards A Generalized And Optimized Apriori Approach, Artem Abdikov

Master's Theses

Apriori is a machine learning algorithm developed in 1994 by R. Agrawal and R. Srikant for association rule mining purposes. This family of algorithms takes transactional data and analyzes relationships between variables in large datasets. The typical output of such algorithms is a prediction that if users choose item X, it is highly likely that they will also choose item Y. Apriori is known to be a robust algorithm and is used by many large companies in order to analyze user tendencies and even make recommendations. Although Apriori is a powerful algorithm, its original implementation is known to have limitations, …

Towards Automated And Explainable Insider Threat Response In Electronic Health Records: A Role-Aware Machine Learning Framework, Luca Lippi Ornstil

Master's Theses

Healthcare remains a prime target for cyberattacks, with insider misuse and credential compromise posing major risks to Electronic Health Records (EHRs). This thesis introduces a role-aware, explainable anomaly detection and response framework integrated with OpenEMR to address post-authentication threats. Four models—Local Outlier Factor (LOF), Isolation Forest, Autoencoder, and Graph Neural Network (GNN)—detect behavioral deviations across temporal, device, and role-based features, with LOF serving as the primary runtime detector. A configurable policy engine maps anomaly severity to proportional actions, from email alerts to read-only restrictions or account suspension, all reversible and auditable. Evaluation on real EHR logs shows the system’s operational …

Enhancing Non-Player Character Dialogue In Video Gages: An Evaluation Of Large Language Model-Generated Responses, Lam P. Quach

Master's Theses

As video games increasingly emphasize narrative depth and player immersion, the quality of Non-Player Character (NPC) dialogue has become crucial for creating engaging gaming experiences. This thesis investigates the potential of Large Language Models (LLMs) to generate high-quality NPC dialogue by comprehensively evaluating four state-of-the-art models: Gemma 3 27B, Mistral 7B, QWEN 2.5, and LLAMA 3.1. The study employs a mixed-methods approach, combining human evaluation (N=50 participants) with AI-based assessment across five key benchmarks: coherence, personality expression, engagement, style/tone appropriateness, and overall quality. Participants evaluated 32 dialogue samples (8 per model) generated for a fantasy game context featuring two distinct …

Understanding And Evaluating Genomic Language Models, Aadit Kapoor

Master's Theses

Large Language Models (LLMs) have shown remarkable capabilities in interpreting complex patterns across various domains, yet their application to genomic data remains limited. We see great potential in leveraging LLMs for vital biological tasks, such as predicting transcription factor binding sites and identifying antibiotic-resistant genes. This emergent behavior positions LLMs as powerful tools for enhancing our understanding of intricate biological language. LLMs trained specifically on genomic data, such as DNA sequences, operate distinctly compared to those trained on natural language. This difference is evident not only in the architectural landscape of the models but also in the methodologies employed by …

Development Of A Control System For An 8-Dof Quadrupedal Robotic Research Platform, Jack Butler

Master's Theses

Quadrupedal robots offer a versatile locomotion option that can extend the operating space of a robot into uneven terrains. However, controlling these systems presents significant challenges due to nonlinearities introduced by various factors.

In this thesis, model-predictive control (MPC) is applied to an 8-DOF legged robot developed by Cal Poly’s Legged Robotics group. The MPC framework employs a lumped rigid-body model that treats the robot as a single rigid body with forces applied directly at the foot contact points. The controller is developed within the ROS2 environment, with integration of state estimation and gait-pattern generation, to provide maximum modularity and …

Cost-Effective Automated Uhi Mapping With Ai: A Case Study Of A Scalable Framework For Climate Equity In San José, California, Martin Alvarez Lopez

Master's Theses

Urban areas experience the Urban Heat Island (UHI) effect, with higher temperatures than rural areas, disproportionately impacting low-income communities. Mapping UHIs is a process that usually requires significant amount of human resources, and is not scalable. The lack of accurate and detailed UHI maps makes it difficult for decision makers to design effective mitigation strategies. In this work we introduce a cost-effective, scalable, and universally applicable UHI mapping framework that leverages open-source data and AI-driven feature extraction from remote sensing imagery. Using various causative factors such as city characteristics, anthropogenic heat, city canyons, and meteorological variables, we create UHI maps …

Quantization On Graph Neural Networks For Image Classification, Rithik Reddy Katpally

Master's Theses

Quantization has become a key approach for reducing storage and computational demands of deep neural networks while maintaining high accuracy. Although 8-bit quantization is well-established for convolutional architectures such as ResNet50 and MobileNetV2, its application to graph-based vision models remains underexplored. In this work, we extend quantization-aware training to Vision Graph Neural Networks (ViGs) and conduct comparisons with quantized CNNs on the CIFAR-100 dataset. To ensure parity, all models have same training hyperparameters such as learning rate, batch size, optimizer, number of epochs. We used numerous techniques to preserve performance for low-bit precision. First, Pauta Quantization clips activation outliers based …

Stead: Spatio-Temporal Efficient Anomaly Detection For Time And Compute Sensitive Applications, Andrew Gao

Master's Theses

This paper presents a new method for anomaly detection in automated systems with time and compute sensitive requirements, with unparalleled efficiency. As these systems become increasingly popular, ensuring their safety has become more important than ever. Therefore, this paper focuses on how to quickly and effectively detect various anomalies in the aforementioned systems, with the goal of making them safer and more effective. Many detection systems have been developed with great success under spatial contexts; however, there is still significant room for improvement when it comes to temporal context. While there is substantial work regarding this task, there is minimal …

Using Facial Recognition For Selective Pose Detection, William J. Parker

Master's Theses

Pose detection involves locating and identifying key body points for all individuals within a frame. This enables the ability to convert the pose into a digital format, which can then be recorded and analyzed for a variety of purposes. Advancements in the field have already opened applications in areas such as digital fitness coaches, fall detection, and virtual reality. Existing approaches primarily focus on tracking all detected individuals, which limits the practical applications when attempting to analyze a single or specific subject when there are other people in frame. Previous work has discussed integrating identification, but these approaches use identification …

Quantum Algorithm Emulation Using Fpgas, Samuel Petruescu

Master's Theses

Field Programmable Gate Arrays (FPGAs) have been used in most of the physics sub-fields for various unique purposes. This includes particle physics, quantum optics, and, more recently, quantum computing. FPGAs boast many benefits over previous experimental and computational setups. They are versatile, easy to program, and cost-effective, leading to an understandable desire to incorporate them into the new field of quantum computing. While FPGAs have been used to help control the readout and control of physical qubits, they can also be a good tool for improving algorithm simulations, which is the focus of this paper. Different algorithms have different computational …

Sirilla: Predicting Traffic Flow Via Stacked Decentralized Federated Learning, Andrew Selvia

Master's Theses

Billions of people today rely on traffic predictions to optimize their travels. Digital mapping services deliver accurate predictions by learning from vast troves of historical data. Impressive as these systems are, their assumptions do not always apply. They depend on an endless flow of sensitive user data to a central authority, a stable Internet connection, and trustworthiness on both sides of the traditional client-server model. This thesis explores a novel architecture which bucks those assumptions. In the proposed model, traffic data remains on edge devices which individually train models via federated learning. Beyond the obvious privacy benefits, this architecture enables …

Reverse Engineering Bring Up And Profiling Of Multi-Fpga Systems, Henry A. Evans

Master's Theses

FPGAs have long been used for prototyping and verifying high-speed digital designs in industry and in academic research. As ASIC designs have grown in complexity and size, prototyping those designs on FPGAs has required multiple FPGAs that sometimes span multiple servers. Western Digital donated multiple FPGA-based systems to Cal Poly in 2023. These servers contain multiple high-end AMD FPGAs that are ideal for prototyping large high-speed digital designs, however the full documentation on how to use the servers and how the servers work was not provided. The servers did not come with any information on how to program the FPGAs, …

Assessment Of The Imx7ulp Heterogeneous Soc For Use In A Next Generation Student Cubesat Obc, Lorenzo A. Pedroza

Master's Theses

CubeSats represent a rapidly evolving platform for space research, industry, and education, demanding increasingly sophisticated onboard computing solutions that balance performance, fault tolerance, size, weight, and power constraints. Although the design of the Cal Poly's CubeSat Lab's (PolySat) current On-Board Computer (OBC) keeps up with many of these factors, the demands of modern workloads like fine attitude determination will start to outpace available compute. While the selection of a more capable processor to address this would have traditionally resulted in increased energy usage, modern system-on-chip (SoC) architectures offer novel ways to trade available compute for power savings on-the-fly. With this …

Griddle: A Novel Hardware Based Matrix Multiplier Architecture, Seth Kiefer

Master's Theses

Matrix multiplication is a computational cornerstone in modern artificial intelligence and scientific computing, yet general-purpose processors struggle to perform these operations efficiently at scale. This thesis presents Griddle, a novel hardware architecture for matrix multiplication implemented on a Xilinx Artix-7 FPGA. Griddle focuses on flexibility and scalability by adopting a purely iterative approach that supports arbitrarily shaped input matrices without requiring padding or strict dimensional constraints. The architecture uses computational pipelines to execute a multiplication operation. Each pipe consists of a multiplication core and accumulation buffer that compute matrix products in parallel. The multiplication core contains a set of multiplier …

Deeppanorf: Deep Prior For Neural 3d Reconstruction From Sparse Panoramas, Edward Du

Master's Theses

Advances in neural field representations have led to a significant improvement in view synthesis quality. However, many current novel view synthesis methods rely on a dense set of input views, which can be impractical and inefficient in real-world applications. We propose DeepPanoRF, a novel method for 360◦ scene reconstruction from a sparse set of input equirectangular panoramas. Built upon K-Planes, a radiance field representation that encodes explicit features on orthogonal feature planes, our method does not directly learn feature grids. Instead, we parameterize the feature grids to enable sparse view reconstruction without pretraining or additional regularization. We implement a custom …

Adversarial Deep Reinforcement Learning For Tank Duel Simulation Using Lidar-Based Observations, Braedan Kennedy

Master's Theses

Previous research has demonstrated that reinforcement learning agents can learn to steer differential-drive robots around obstacles using 2D lidar scans as observations. However, these studies typically treat all range returns as undifferentiated obstacles—objects to avoid—without distinguishing between different object types. This thesis builds upon previous research by introducing an adversarial task in which an agent must interpret raw range readings to both avoid static obstacles and identify, pursue, and engage a hostile target.

To investigate this problem, this thesis introduces TankGame, a novel, lightweight 2D tank duel simulator. Each agent receives a 360° lidar scan, controls its motion via tread …

Sentry V3: Extending Context Switches On A Trusted Secure Coprocessor, Mark Kong

Master's Theses

Software correctness and integrity is only ensured through trust in the un- derlying hardware. However, modern computer systems are complex to design and secure. Thus, given the choice between performance and security, companies will often prioritize performance, resulting in vulnerable systems. This creates exploitable systems that must be patched retroactively because business value performance over security. One approach to this issue is to separate the root of security from the rest of the system to create a minimal trusted computing base. Trustguard is one instance of this. Trustguard implements a Containment Architecture with Ver- ified Output (CAVO) model which shows …