Digital Commons Network^™

From Neural Networks To Large Language Models: Innovations In Financial Ai, Mathematical Reasoning, And Structured Data Representation, Junyi Ye

Dissertations

This dissertation explores the evolution and application of artificial intelligence techniques across three critical domains: financial modeling, mathematical reasoning, and structured data analysis. The dissertation presents seven research projects that chart a progression from specialized neural architectures to sophisticated large language models (LLMs), contributing novel methodologies and frameworks at each stage.

In the financial domain, the research first introduces TS-Mixer, a MLP-based architecture for time-series forecasting that captures both feature relationships and temporal dependencies through a simple yet effective design, outperforming more complex models in S&P500 index prediction. The dissertation then presents DySTAGE, a dynamic graph representation learning framework that …

Gamified Gait Rehabilitation Via Real-Time Biofeedback And Adaptive Hip-Exoskeleton Control, Mariya Huzaifa Tohfafarosh

Theses

Gait impairments arise from systemic diseases, age-related degeneration, musculoskeletal dysfunctions, or neurological conditions. While traditional rehabilitation can be effective, they often face challenges such as high costs, inaccessibility, and low patient engagement. To address these challenges, my work introduces a virtual reality-based rehabilitation (VRBR) system, integrating real-time motion and electromyographic (EMG) muscle activation feedback with a gamified virtual environment for enhanced adaptability and engagement. The system includes a custom-designed hip-exoskeleton that provides adaptive spring-like assistance or resistance, supporting both mobility-impaired users and strength training. Assistance levels can be tuned to match the user's progress. Additionally, a custom pressure insole was …

Quantum-Enhanced Training Of Large Language Models: A Hybrid Approach, Nan Wu, Fangmin Song, Xiangdong Li

Publications and Research

The training of large language models (LLMs) presents significant computational challenges, particularly regarding efficient convergence. This paper presents a hybrid quantum-classical framework designed to address the significant computational challenges associated with training large language models (LLMs). By integrating quantum computing principles superposition, entanglement, and tunneling with classical deep learning methods, we propose an approach to accelerate convergence, enhance optimization efficiency, and improve model generalization. Specifically, quantum feature mapping is employed to project classical data into high-dimensional Hilbert spaces, facilitating more expressive data representations. Quantum-assisted optimization algorithms, such as Quantum Approximate Optimization Algorithm (QAOA) and Variational Quantum Eigensolver (VQE), efficiently navigate …

Plm-Dbps: Enhancing Plant Dna-Binding Protein Prediction By Integrating Sequence-Based And Structure-Aware Protein Language Models, Suresh Pokharel, Kepha Barasa, Pawel Pratyush, Dukka B. Kc

Michigan Tech Publications

DNA-binding proteins (DBPs) play a crucial role in gene regulation, development, and environmental responses across plants, animals, and microorganisms. Existing DBP prediction methods are largely limited to sequence information, whether through handcrafted features or sequence-based protein language models (PLMs), overlooking structural cues critical to protein function. In addition, most existing tools are trained for general DBP predictions, which are often not accurate for plant-specific DBPs due to the unique structural and functional properties of plant proteins. Our work introduces PLM-DBPs, a deep learning framework that integrates both sequence-based and structure-aware representations to enhance DBP prediction in plants. We evaluated several …

Photojournalism In The Age Of Deepfakes: The Role Of Media Literacy And Ethical Standards In Restoring Trust In Visual Reporting, Ionnnis Kontos, Katerina Chryssanthopoulou, Ioannis Galanopoulos-Papavasileiou

All Works

This article explores the impact of deepfake technology on photojournalism, highlighting its role in undermining trust in visual media. As deepfakes allow for the creation of highly realistic manipulated content, they pose significant challenges regarding the authenticity of journalistic imagery and erode the authority of visual truthfulness. The widespread use of deepfakes has led to a decline in public confidence in the credibility of news, raising concerns about the future of photojournalism in an era of digital deception. As a solution to regaining viewers’ trust, this article suggests a twofold approach: First, it emphasizes the importance of media literacy in …

Artificial Intelligence Use In Medical Education: Best Practices And Future Directions, Rasheed A. M. Thompson, Yash B. Shah, Francisco Aguirre, Courtney Stewart, Costas D. Lallas, Mihir S. Shah

Department of Urology Faculty Papers

PURPOSEOF REVIEW: This review examines the various ways artificial intelligence (AI) has been utilized in medical education (MedEd)and presents ideas that will ethically and effectively leverage AI in enhancing the learning experience of medical trainees.

RECENT FINDINGS: AI has improved accessibility to learning material in a manner that engages the wider population. It has utility as a reference tool and can assist academic writing by generating outlines, summaries and identifying relevant reference articles. As AI is increasingly integrated into MedEd and practice, its regulation should become a priority to prevent drawbacks to the education of trainees. By involving physicians in …

Are Cycles Of Neural Activity The Algorithm Of The Brain?, Edwin Omondi Onyango

Computer Science Senior Theses

We propose that precisely timed neural activity cycles can serve as structural primitives for memory and computation in a system that exhibits associative learning like the brain. Inspired by biologically grounded mechanisms such as calcium-dependent plasticity, spike-timing-dependent learning, and phase-sensitive excitability, we construct a spiking neural network model in which repeated temporal coincidences drive the formation of self-sustaining activity loops. These cycles, once formed, persist as dynamic memory traces: not stored as static weights, but as reverberating patterns that replay in time when these loops are restarted. We show that noise alone fails to induce stable structure, but even sparse, …

Some Studies On Information Set Decoding Algorithms And Universal Hash Functions, Sreyosi Bhattacharyya

Doctoral Theses

This thesis presents some studies on Information Set Decoding algorithms and Universal Hash Functions. In the context of Information Set Decoding (ISD) the thesis studies time/memory trade-off of ISD algorithms and in the context of universal hash functions, the thesis studies design and efficient implementations of polynomial hash functions defined over prime order fields. A cornerstone of ISD algorithms is the algorithm proposed by Stern and it introduced the meet-in-the-middle collision search approach to ISD algorithms. Though this algorithm is more efficient in terms of asymptotic time complex- ity than the preceding algorithms proposed by Prange, Lee and Brickell and …

Detection And Mitigation Of Out-Of-Band Channel Wormhole Attack In Wireless Network Using Propagation Delay, Harry May

Doctoral Dissertations

Wireless networks, susceptible to a range of attacks due to their simplicity and ease of evasion, face a significant threat from control data attacks, notably the elusive wormhole attack. Detecting and mitigating such attacks poses challenges, particularly in the absence of a digital signature. This dissertation introduces an innovative approach that utilizes the propagation delay associated with malicious nodes’ timing characteristics for detection, employing the Ad-hoc On-Demand Distance Vector (AODV) algorithm as its foundation. The inherent propagation delay in the AODV protocol is calculated for each node link along the entire communication path, offering a distinctive timing method that provides …

Who Should Take Responsibility For Artificial Intelligence Actions And Outcomes? Perception Of Auditors As Users Of Ai Systems, Hanh Hoang Le

Doctoral Dissertations

As artificial intelligence (AI) systems become increasingly embedded in auditing processes, questions arise regarding how professional auditors perceive and allocate responsibility for AI-assisted decisions. This study investigates the effects of AI explainability and auditors’ perceived autonomy on perceived responsibility in the context of audit decision-making. Drawing on theories of moral responsibility and professional judgment, the study employs a 2x2 experimental design using hypothetical audit scenarios to manipulate levels of AI explainability and auditors’ autonomy. Hierarchical regression analysis reveals that perceived autonomy statistically significantly increases auditors’ perception of responsibility for AI-assisted decisionmaking, whereas AI explainability is not a significant predictor. Additionally, …

Design And Optimization Of Low-Loss, High-Gain Metamaterial-Based Log-Periodic Dipole Array Antenna For Full Ka-Band Coverage, Mohamed El Moniar, Ahmed Abd El Hady, Yasser Ismail, Nihal Areed

Turkish Journal of Electrical Engineering and Computer Sciences

This work describes a microstrip log-periodic dipole array (MLPDA) antenna that uses metamaterials and operates across the whole Ka-band. The suggested MLPDA antenna layout provides a wide bandwidth with fewer dipole elements than traditional MLPDA antennas while maintaining the same resonance frequencies. To reduce size while covering a wide operational spectrum, the antenna design includes bending dipoles as radiating elements, as well as an incomplete ground plane. Furthermore, the proposed MLPDA antenna’s energy loss has been reduced while boosting its signal strength (gain) by inserting a metamaterial-based structure in front of it at a certain distance and on the same …

Helmholtz Cage: Software Development And Implementation For Cubesat Testing, Gustavo A. Cotom Lopez

University Honors Theses

This paper details the successful development and deployment of a Helmholtz cage system, designed to produce precisely controlled magnetic fields for testing and calibration purposes. The core focus was on creating a robust, modular software architecture enabling independent current modulation on each axis, comprehensive serial communication between multiple microcontrollers, and real-time data acquisition from the MR3 magnetometer. All software components, including the serial communication drivers, control algorithms, command line interface, and calibration routines, were developed from the ground up. The system was fully operational upon completion: all hardware components functioned as intended, serial communication with each subsystem was reliable, and …

Blockchain-Enabled Master Data Management, Shakhawat Hossain

Theses and Dissertations

Master Data Management (MDM) is essential for maintaining data quality, accuracy, consistency, and governance within organizations. However, traditional centralized MDM systems continue to face challenges related to data integrity, security, and scalability. This research presents a blockchain-enabled MDM framework designed to overcome these limitations by leveraging blockchain’s decentralized, immutable, and secure architecture. The study aims to identify and address the shortcomings of conventional MDM practices, examine the applicability of blockchain technology in enhancing these systems, and develop a functional prototype to validate the proposed model. The framework incorporates decentralized review mechanisms that improve auditability and ensure trusted data verification by …

Hotlangbench, A Tiny Benchmark Suite For Higher-Order Statically Typed Languages, Konstantin Laufer

Computer Science: Faculty Publications and Other Works

This work in progress aims to compare various HOT (higher-order and statically typed, a term coined by Phil Wadler) through reproducible course-grained, wall-time benchmarks. Our overall goals include simplicity, agility, and reproducibility.

There is currently only one benchmark, but it brings out substantial performance differences among the various languages and platforms. It uses function composition and other higher-order constructs to build a pipeline of transformations, along with a brute-force iteration that is computationally expensive for input files specifying large ranges as function domains. We currently include versions in Modern C++, C#, Go, Haskell, Kotlin, Modern (stream-based) Java (24), OCaml, Scala …

Transformer Decoder-Enhanced Swin Unetr For Multi-Organ Semantic Segmentation On Openkbp: Improving Radiotherapy Planning Accuracy, Zainab Adnan Jwad, Israa Hadi Ali

Karbala International Journal of Modern Science

Accurate segmentation of organs-at-risk (OARs) in head and neck CT scans is crucial for radiotherapy planning. The CNN-based decoder limitation of Swin UNETR hinders its capacity to process meaningful information from multiple organ positions essential for accurate medical segmentation. The proposed Transformer Decoder-enhanced Swin UNETR model targets the OpenKBP dataset multi-organ segmentation through its dedicated design for this purpose. The model utilizes transformers along with cross-attention approaches in its decoder to improve segmentation mask outputs through analysis of extensive global information. The model gets additional feature representation power through the addition of squeeze-and-excitation (SE) blocks linked with spatial attention mechanisms …

Computational Approach: 3d-Qsar, Molecular Docking, Molecular Dynamics Simulation Investigations, Drug- Like-Ness, And Dft Score Evaluation Of A Potential Novel And Retrosynthesis Of Some Tr-H Derivatives As Streptococcus Pneumoniae Drug., Belaidi Mustapha, Tchouar Noureddine, Djelilate Mohammed, Saleh Bufarwa, Dalal K. Thbayh

Karbala International Journal of Modern Science

Streptococcus pneumoniae is the main source of hospital-acquired pneumonia and meningococcal pneumonia in children, adults, and the elderly, especially the immunocompromised. Recently, it has attracted the attention of many research studies around the world as a potential target for remediation. TR-H are considered SP antibacterial agents, as they can inhibit. To describe their mode of action and to identify new antibacterial drugs against Streptococcus pneumoniae exhibiting the TR-H scaffold, the present work included 3D-QSAR, in- silico pharmacokinetic evaluation and molecular simulation modelling of TR-H. Using an atom-based method for quantitative structure-activity relationship (QSAR) study, a comprehensive 3D-QSAR model …

Dynamate: Leveraging Ai-Agents For Customized Research Workflows, Orlando A. Mendible-Barreto, Misael Díaz-Maldonado, Fernando J. Carmona Esteva, J. Emmanuel Torres, Ubaldo M. Córdova-Figueroa, Yamil J. Colón

Computer and Data Science Faculty Publications

No abstract provided.

Integrative Computational Modeling Of Distinct Binding Mechanisms For Broadly Neutralizing Antibodies Targeting Sars-Cov-2 Spike Omicron Variants: Balance Of Evolutionary And Dynamic Adaptability In Shaping Molecular Determinants Of Immune Escape, Mohammed Alshahrani, Vedant Parikh, Brandon Foley, Gennady M. Verkhivker

Mathematics, Physics, and Computer Science Faculty Articles and Research

In this study, we conducted a comprehensive analysis of the interactions between the receptor-binding domain (RBD) of the SARS-CoV-2 spike protein and four neutralizing antibodies—S309, S304, CYFN1006, and VIR-7229. Using integrative computational modeling that combined all-atom molecular dynamics (MD) simulations, mutational scanning, and MM-GBSA binding free energy calculations, we elucidated the structural, energetic, and dynamic determinants of antibody binding. Our findings reveal distinct dynamic binding mechanisms and evolutionary adaptation driving the broad neutralization effect of these antibodies. We show that S309 targets conserved residues near the ACE2 interface, leveraging synergistic van der Waals and electrostatic interactions, while S304 focuses on …

Historical Perspectives In Volatility Forecasting Methods With Machine Learning, Zhiang Qiu, Clemens Kownatzki, Fabien Scalzo, Eun Sang Cha

All Faculty Open Access Publications

Volatility forecasting for financial institutions plays a pivotal role across a wide range of domains, such as risk management, option pricing, and market making. For instance, banks can incorporate volatility forecasts into stress testing frameworks to ensure they are holding sufficient capital during extreme market conditions. However, volatility forecasting is challenging because volatility can only be estimated, and different factors influence volatility, ranging from macroeconomic indicators to investor sentiments. While recent works show promising advances in machine learning and artificial intelligence for volatility forecasting, a comprehensive assessment of current statistical and learning-based methods is lacking. Thus, this paper aims to …

Towards Multi-Modal Multi-Document Understanding Capabilities In Foundation Models, Chuhan Li

Computer Science Theses

Contemporary foundation models are predominantly evaluated on isolated documentor image-understanding tasks, thereby overlooking the inherent multimodal multi-document reasoning that characterizes scientific inquiry. To bridge this gap, M3SCIQA is introduced, aMulti-Modal,Multi-document Scientific Question Answering benchmark crafted to test foundation models in practical scientific research settings. A comprehensive evaluation of 18 leading foundation models shows a substantial performance gap between models and human experts. Detailed error analysis reveals persistent deficiencies in both scientific visual reasoning tasks and long-range retrieval. Addressing the former, SPACECUE offers a concise yet effective visual prompting that overlays grid coordinates and Semantic-SAM masks …

Towards Practical And Real-Time Decoding Of Quantum Hypergraph Codes, Binghong (Leo) Li

Computer Science Theses

Quantum error correction (QEC) enables scalable quantum computation by detecting and correcting physical errors. However, decoding remains a key bottleneck—particularly for quantum low-density parity-check (qLDPC) codes, whose hypergraph structures demand complex reasoning. Most existing decoders are either too slow for real-time use or lack formal guarantees, and often struggle to generalize across diverse quantum hardware.

This thesis introduces the Minimum-Weight Parity Factor (MWPF) algorithm, a unified and certifiable decoding formulation that extends minimum-weight perfect matching to general hypergraph-based stabilizer codes. We focus on making MWPF practical and performant through a two-phase decoding architecture, consisting of a fast search phase and …

Programming A More Efficient Onboarding Process For New Employees, Long H. Pham

Undergraduate Honors Theses

The current onboarding process for new hires in the University of San Diego’s Shiley-Marcos School of Engineering is inefficient. There is no central location where new hires and administrators can track onboarding progress. Both parties have to manage multiple email chains and write their own reminders to keep track of everything. This leads to delays, missing deadlines, and confusion for both parties. A web-based onboarding application has been developed recently to address these issues and streamline the onboarding process for new hires. However, this application contains several accessibility issues and does not follow all of the standards for effective employee …

Simulating 3d Humanoid Ragdoll Physics Using Velocity Verlet Integration, Pin Constraints, And Rigid Body Collision Systems, Son D. Nguyen

Programming Theses and Dissertations

Ragdoll physics simulates realistic character collapse with physical realism by responding to environmental forces rather than using predefined animations.

Data Encoding, Compilation, And Algorithms For Quantum Machine Learning, Aviraj Sinha

Computer Science and Engineering Theses and Dissertations

Quantum computing enables new approaches to data processing, especially in quantum machine learning. Unlike classical systems, quantum data must be synthesized through operations and can exist in superposition. Encoding choices affect efficiency, noise resilience, and trainability—key factors in quantum machine learning models. This dissertation enhances quantum data encodings by extending quantum read-only memory (QROM) beyond binary representations, improving efficiency and parallelism. It introduces new compilation methods for quantum random number generators (QRNGs), supporting non-parametric distributions for post-quantum cryptography. Additionally, it explores Cayley graph-based encodings to extract spectral features for quantum machine learning.

Noise-Embedded Image Processing Based On Quantum Data Encodings, Yayu Mo

Multidisciplinary Studies Theses and Dissertations

Advancements in quantum information have significantly impacted the field of image processing, although challenges remain. Especially in the edge detection and image encoding area, distorted feature and noises would affect the further classification or super resolution tasks. In our work, we conduct researches on two stages to both evaluate the potential of Quantum-based Convolutional Structure in extracting distorted feature and further explore the effects of quantum noise channels on quantum image encodings.

In the first stage, we propose a method to extract distorted edge features by applying shallow layers in quantum convolutional neural networks (QCNN). By combining the advantages of …

Bucket-Based Priority Queues For A* And Related Bounded-Suboptimal And Anytime Search Algorithms: Theoretical And Practical Advancements, Garrett Michael Fereday

Theses and Dissertations

For shortest-path problems with a small number of integer transition costs, it is well-known that the performance of the classic A* algorithm can be improved by using bucketing to reduce priority queue overhead—in particular, by using a bucket queue data structure for the priority queue, instead of a binary heap. This dissertation describes several theoretical and practical extensions of this approach. First, the traditional two-level bucket queue data structure is modified in simple ways to improve the worst-case complexity of its operations, which leads to the first demonstration that the priority queue operations of a two-level bucket queue for A* …

Promoting Collaboration And Multi-Directional Reliance By Sharing Mental Model Information For Effective Multi-Agent Teaming, Audrey L. Aldridge

Theses and Dissertations

Successful human-agent teaming requires teammates to form and maintain a shared or common understanding of several attributes regarding taskwork and teamwork. With enhanced information sharing, mental model development, and team functionality, teammates (human, autonomous) can learn to anticipate each others' behaviors, preferences, and needs as well as understand their capabilities and limitations. In designing a framework to support this type of cooperative teaming, there is a need to determine how sharing knowledge, mental models, and common understandings impacts teaming dynamics and performance. By incorporating each individual's understanding into a human-agent interface, this research enables better team coordination and performance through …

Optimizing Mars Terrain Segmentation With Weakly Supervised Learning: A Focus On Weighted Loss From Annotation Metadata, Malika Dutta

Theses and Dissertations

The study of planetary surfaces heavily depends upon space rovers that gather detailed images of terrain needed for analysis and navigation. Deep neural networks and other sophisticated machine learning techniques are necessary for autonomous navigation in challenging terrain. However, the inconsistent annotations by citizen scientists frequently hinder the performance of these models. This study seeks to optimize terrain segmentation to improve the autonomous capabilities of future Mars rovers by presenting a novel weakly supervised learning framework to handle noise and unreliability in datasets. Using factors like number of clicks, pixel accuracy, and annotator dependability, the method utilizes annotation metadata in …

Cyber Security Threat Recognition And Preparedness Of Undergraduate Students, Litany Hope Lineberry

Theses and Dissertations

Cybersecurity awareness and preparedness are critical competencies for individuals across academic disciplines and professional sectors. However, undergraduate students often lack sufficient knowledge and skills to recognize and mitigate cybersecurity threats. This dissertation examines cybersecurity threat recognition and preparedness among undergraduate students through a three-phase research approach. Study 1 explores faculty perspectives on students' cybersecurity awareness, identifying gaps in knowledge and preparedness across various fields of study. Study 2 investigates industry professionals' perceptions of new hires’ cybersecurity readiness, assessing the alignment between academic training and industry expectations. Study 3 evaluates the effectiveness of an online intervention designed to enhance students' cybersecurity …

Ai Enabled Autonomic, Safe, And Interactive Intrusion Response System, Damodar Panigrahi

Theses and Dissertations

The exponential rise in internet usage has precipitated a corresponding surge in cyber threats, underscoring the urgent need for advanced cybersecurity solutions. While traditional intrusion detection systems (IDS) can identify these threats, their inability to self-recover leaves systems vulnerable. Intrusion response systems (IRS) have been developed to address this, aiming to auto- matically restore systems to their desired state post-security breach. However, current IRSs often necessitate manual intervention and may not be su!ciently robust against sophisticated threats. To overcome these limitations, we propose an AI-powered Autonomic, Safe, and Interactive Intrusion Response System called ‘Intrusion Response System Digital Assistant (IRSDA)’. IRSDA …