Computer Sciences Commons^™

Trustworthy Reinforcement Learning And Communication-Efficient Multi-Agent Systems, Rui Zuo

Dissertations - ALL

The rapid proliferation of autonomous systems, such as Unmanned Aircraft Systems (UAS) transitioning to large-scale Beyond Visual Line of Sight (BVLOS) operations, demands a paradigm shift toward reliable, transparent, and resilient autonomy. While Deep Reinforcement Learning (DRL) and Multi-Agent Reinforcement Learning (MARL) have demonstrated exceptional capabilities in complex decision-making and coordination, their real-world deployment in safety-critical domains is hindered by two fundamental challenges. First, the opaque, "black-box" nature of DRL models prevents human operators from understanding and trusting the agents' underlying logic. Second, as multi-agent operations scale, they become severely constrained by the stochastic link qualities and strict bandwidth limitations …

Scaling Logical Reasoning On Modern Hpc Hardware, Yihao Sun

Dissertations - ALL

Deductive logic reasoning has evolved from a theoretical symbolic artificial intelligence tool into a massive computational workload. Modern domains, ranging from static program analysis and binary reverse engineering to knowledge-graph reasoning, rely on Datalog, a logical query language, to express deeply recursive, declarative specifications. However, scaling logic reasoning to these industrial workload exposes two interlocking ceilings in traditional engines. Architecturally, the memory bandwidth and parallel throughput of single-node CPUs fall drastically short of the read- and write-heavy demands of semi-naive evaluation. Asymptotically, traditional query processing algorithms, such as binary join algorithms generate massive intermediate relations that exhaust device memory on …

From Detection To Segmentation: A Foundation Model Approach To Organoid Brightfield Image Analysis Using Sam 3, Hiren Manani

Theses - ALL

This thesis presents the first systematic application of SAM 3, a unified foundation model for promptable segmentation, to mouse small intestinal organoid brightfield microscopy image analysis. The work spans the complete pipeline from zero-shot baseline evaluation through domain-specific fine-tuning on a GPU cluster, and documents the full engineering process required to adapt a state-of-the-art foundation model to a novel biomedical imaging domain. A comprehensive literature review of over 20 papers spanning detection-based, classical segmentation, foundation model, and morphological analysis approaches identified a clear research gap that this thesis addresses. Five critical compatibility patches were developed to deploy SAM 3 on …

Constrained Neural Intelligence: Investigations Into Deep Multi-Objective Optimization, Naveed Tahir

Dissertations - ALL

Neural intelligence is identified with complex learning problems optimized over the high-dimensional, non-convex parameter spaces of deep neural networks. Solving such problems generally requires handling competing objectives and conflicting constraints. This is traditionally dealt with using heuristic methods that collapse such complexity into unconstrained, singular objectives. While computationally convenient, this invites a tradeoff against the precision and stability afforded by non-heuristic, geometry-aware approaches. This dissertation explores constrained multi-objective learning in various applied and theoretical contexts and highlights the feasibility of approximate methods as well as the necessity of exact methods. We propose a framework for equality-constrained deep learning via approximate …

Sage: Contention-Aware Sensitivity Adjustment For Spin Lock Exiting In Vms, Sam Thomas Stowers

Theses - ALL

To address the severe multi-threaded performance degradation caused by lock contention in oversubscribed cloud environments, this thesis examines and improves on the Pause Loop Exit (PLE) mechanism used by hypervisors to deschedule unproductive virtual CPUs (vCPUs). These oversubscribed environments introduce contention which frequently leads to Lock Holder Preemption (LHP), where a hypervisor preempts a vCPU holding a critical spinlock, causing waiting vCPUs to waste cycles spinning. To mitigate LHP, hardware mechanisms trigger PLE events to inform hypervisors of spinning vCPUs, relying on a PLE_gap parameter to differentiate spinlock instances. The PLE_gap dictates spinlock detection sensitivity—a component of the virtualization stack …

Semantic Entanglement In Vector-Based Retrieval: A Formal Framework And Context-Conditioned Disentanglement Pipeline For Agentic Rag Systems, Nick Loghmani

iSchool - All Scholarship

Retrieval-Augmented Generation (RAG) systems deployed in agentic environments depend on the geometric properties of vector representations to retrieve contextually appropriate evidence for autonomous reasoning. When source documents conflate multiple topics within contiguous text regions, standard vectorization pipelines produce embedding spaces in which semantically distinct content occupies overlapping geometric neighborhoods — a condition we term semantic entanglement. This paper formalizes semantic entanglement as a model-relative measure of cross-topic overlap, defines an Entanglement Index (EI) as a quantitative proxy, and argues that higher EI is associated with reduced attainable Top-K retrieval precision under cosine similarity retrieval. We introduce the Semantic Disentanglement …

A New Tool For Handling Multiracial And Multi-Identity Data In Health Research, Gabriel J. Merrin

Population Health Research Brief Series

When surveys ask about race or ethnicity, a growing number of Americans select more than one category. The multiracial population now represents over 10% of the U.S. population and is the fastest growing racial group in the country. Yet researchers routinely collapse these individuals into an “other race” category for statistical analysis, rendering specific subgroups invisible. This brief introduces CATAcode, a free software tool that helps researchers systematically explore, document, and prepare check-all-that-apply demographic data for statistical modeling. In a demonstration with over 8,000 high school students, CATAcode revealed 85 distinct racial identity combinations from just eight response options. The …

Agentic Intelligence Under Constraint: Energy, Context, And The Expansion Of Exchange, Nick Loghmani

iSchool - All Scholarship

Recent advances in agentic artificial intelligence have been driven primarily by scale: larger models, increased data, and expanding computational resources. However, rising energy costs, inference latency, and hardware constraints increasingly challenge this trajectory. This paper argues that intelligence—biological or artificial—does not primarily scale through raw computational expansion, but through the management of exchange under constraint. Drawing on cognitive science, systems theory, and prior work on exchange-based models of intelligence, the paper proposes a theoretical framework in which agentic intelligence scales through context management, proceduralization, and the assembly of reusable units of exchange. Unlike approaches that focus solely on model compression …

Provoking Generative Ai Futures: Merging Theory And Praxis, Regina M. Luttrell Ph.D., Nick Bowman

Media Studies - All Scholarship

An accessible exploration of the myriad applications and challenges of generative AI for media and communication students, scholars, and practitioners alike.

The latest emergence of increasingly low-cost and scalable AI technologies presents a point of both celebration and concern for the contemporary media and information ecosystem. To this end, this edited volume gathers media and communications scholars and practitioners to engage in discussions and exchange ideas about current trends and developments in the field. Questions this volume asks include: What are the essentials of generative AI from a media and communication perspective? How has generative AI influenced research and scholarship? …

Robust Security Assurance In Cyber-Physical Systems: From Attack Diagnosis To Attack Resilience, Zifan Wang

Dissertations - ALL

Cyber-Physical Systems (CPS) have become integral to critical infrastructure, from autonomous vehicles to industrial control systems. However, their increased connectivity and sophistication introduce new vulnerabilities, making security a paramount concern. This dissertation presents a comprehensive pipeline for enhancing CPS security, focusing on accurate diagnosis of trustworthy time frames and subsystems, followed by system resilience measures to restore safe states. The foundation of system resilience lies in an advanced checkpointing protocol for real-time multi-process systems. A novel three-step approach uniquely addresses both logical and timing correctness, which are crucial for modern CPS applications. By partitioning processes into directed acyclic graphs, implementing …

Model Inference And Sparse Network Analysis: Machine Learning In The Gene Regulatory Network Framework, Youchuan Wang

Dissertations - ALL

This thesis explores sparse network inference from high-dimensional, noisy, and underdetermined data—a fundamental challenge in many scientific domains. We focus on the development of evolutionary computation methods for discovering underlying network structures that are both interpretable and biologically plausible. Our methods are applied to the domain of Gene Regulatory Network (GRN) inference, where sparsity, indirect interactions, and limited observations pose significant hurdles. The framework models both steady-state and time-series gene expression data, with particular emphasis on biological sparsity and regulatory dynamics. We approach the problem from three perspectives: (1) edge-level analysis using transitive reduction to distinguish direct from indirect regulation; …

Balancing Robustness And Practicality In Model Compression, Personalization, And Healthcare Interventions, Sawinder Kaur

Dissertations - ALL

Advances in machine learning have enabled significant progress in generating robust solutions. However, designing solutions that balance robustness while meeting the practical constraints of diverse applications, such as limited resources and data, remains a challenge. We explored three key aspects of this interplay: verified robust compressed neural networks, context-wise robust personalization, and reliable counterfactual interventions for healthcare. First, we introduce VeriCompress, a novel framework to streamline the synthesis of compressed neural networks with formal guarantees of adversarial robustness. This enables the deployment of reliable and efficient models in resource-constrained environments, such as smartphones. Second, we developed CRoP (Context-wise Robust Static …

Adaptive Generation In Evolutionary Robotics: From Adversarial Objects To Guided Optimization, Unknown Akshay

Dissertations - ALL

Evolution-inspired algorithms have proven effective for complex optimization problems butsuffer from computational inefficiency due to their reliance on random variation operators. This is problematic in domains where fitness evaluation depends on expensive procedures such as training a neural network or running a robot, either in simulation or on hardware. This dissertation presents novel approaches for evolutionary robotics that replace stochastic evolutionary operations with learned, adaptive strategies using reinforcement learning (RL), significantly improving search efficiency while maintaining population diversity.The first contribution is a voxel-based evolutionary framework for generating adversarial objects that challenge robotic grasping systems. By evolving objects with controlled similarity …

Is Ai Replacing Human Mental Health Professionals?, Michiko Ueda

Population Health Research Brief Series

An increasing number of people are turning to generative artificial intelligence (AI) tools and AI-assisted chatbots to manage mental health concerns. This data slice presents findings from a national survey of U.S. adults aged 18-49 (N = 1,805) conducted in October 2025. Among respondents, 35.2% reported using AI tools more than once a week for mental health support. Among those who had ever seen a human mental health professional, 28.4% reported visiting human providers less often since beginning to use AI for the same purpose. The findings suggest that a subset of users may be using AI to replace, rather …

Protecting Networks Against Entry-Point Attacks, Shivani Mishra

Theses - ALL

In many network applications, it is critical to protect sensitive nodes from discovery by malicious crawlers. This thesis addresses the network protection problem from the data protector’s perspective, focusing on strategically deleting edges to hide target nodes from entry-point attacks. Earlier work on this problem proposed node-level scores to identify key edges for deletion. We propose two novel edge-level scoring functions to identify critical edges for removal: the Shortest Path Change Score (SPCS), which quantifies the damage an edge’s removal causes to shortest paths, and the PageRank Edge Flow Score (PEFS), which estimates an edge’s usage in random walks from …

A Model Of Blind Quantum Diffusion, John Carver Vining

Dissertations - ALL

We introduce the Blind Quantum Diffusion Network (BQDN), a quantum computational framework designed for distributed systems operating under conditions of partial observability and asynchronous communication. BQDN generalizes the structure of Generalized Boolean Networks (GBNs) by replacing classical multivalued logic nodes with quantum registers, enabling the use of superposition, entanglement, and unitary evolution to model distributed computation and state propagation. Building on the properties of W-states and quantum teleportation protocols, the BQDN leverages entangled coin registers and a token-based control mechanism to implement blind diffusion across a graph. The system achieves synchronization-free evolution without classical coordination by embedding decision-making within quantum …

Broadband Resilience By Zero Trust Community Network Policy Design, Lee W. Mcknight, Danielle Smith

The Lender Center for Social Justice

This paper proposes a Zero Trust framework for broadband policy design to enhance community network resilience. It provides a governance and policy perspective for ensuring secure, equitable broadband access.

Enhancing Security And Healthcare Through Continuous Monitoring On Wearable Devices, Sicong Chen

Dissertations - ALL

Wearable and mobile devices are becoming deeply embedded in daily life, supporting a growing range of tasks - from communication and entertainment to health monitoring and productivity. As people increasingly rely on these devices, it is essential to ensure both the protection of the sensitive data they store and the ability to derive meaningful insights that enhance users’ well-being. This dissertation investigates how wearable and mobile devices can be leveraged to provide robust and user-friendly solutions through continuous monitoring across two critical domains: security and mental health. The foundation of this dissertation is the development of a security infrastructure, as …

Online And Offline Learning For Embodied Ai In Autonomous Systems, Kun Wu

Dissertations - ALL

Embodied Artificial Intelligence (AI), which integrates physical embodiment with intelligent decision-making, is increasingly critical in advancing autonomous systems across diverse domains such as autonomous driving and robotic manipulation. This dissertation presents a comprehensive exploration of online and offline learning approaches for Embodied AI in autonomous systems, addressing both algorithmic innovations and dataset construction to overcome fundamental challenges in perception, decision-making, and control. Through five interconnected studies, we systematically advance the state of the art in deep reinforcement learning (DRL) and imitation learning for embodied control. First, we introduce CADRE, a cascade online DRL framework for vision-based autonomous urban driving that …

Data-Efficient 3d Deep Learning, Minmin Yang

Dissertations - ALL

3D data, whether represented as point clouds, volumetric data or meshes, plays a critical role in domains such as autonomous driving, robotics, and medical imaging. However, the complexity of 3D data acquisition and the high cost of annotation often make it impractical to curate large, fully labeled 3D datasets. Furthermore, the unstructured nature of point clouds and the high dimensionality of volumetric data pose additional challenges for designing effective deep learning models. While existing architectures, like PointNet and DGCNN, have made significant progress in learning directly from raw 3D inputs, they typically rely on data-rich scenarios. This dissertation focuses on …

Towards The Next Generation Of Storage Stack For Nand Flash Memory-Based Systems, Ziyang Jiao

Dissertations - ALL

The explosive growth of data has led to increased attention on NAND flash-based solid-state drives (SSDs), which offer high performance, low power consumption, and significant capacity per unit volume when compared to traditional hard-disk drives (HDDs). However, as NAND flash memory density continues to scale, modern SSDs suffer from what is known as fail-slow symptoms, and their performance degrades over time as they wear out. In this dissertation, we focus on understanding and addressing the performance, reliability, and sustainability challenges of modern flash-based storage systems by modeling key metrics, analyzing the design tradeoffs between these metrics, and optimizing existing storage …

Automated Flaw Discovery In Decentralized Systems Via Semantic Fuzzing Tools, Yibo Wang

Dissertations - ALL

This dissertation studies the security challenges of blockchain transaction processing before consensus. While prior work has focused on consensus protocols and smart contract bugs, the pre-consensus infrastructure, such as the mempool and off-chain batching, remain underexplored. This dissertation aims to bridge that gap by systematically analyzing both transaction processing in the mempool and off-chain transaction batching for cost-optimization. This dissertation focuses on two core contributions. First, it presents MPFUZZ, an automated software fuzzing tool designed to uncover denial-of-service vulnerabilities in Ethereum mempools. This work is the first to formally define the mempool fuzzing problem and introduce bug oracles that detect …

Exploring Higher-Order Networks, Hao Tian

Dissertations - ALL

Networks are natural representations of interactions in the real world (social networks, bio-networks, road networks, and the like) and are utilized across various disciplines. By default, network interactions are pairwise; in recent years, the demand for model ing higher-order interactions has kept increasing. For example, in collaboration networks, we aim to distinguish between one publication coauthored by three or three publications coauthored by two in a triangle. In this work, we perform higher-order network analysis in the following two directions. First, we explore the influence of higher-order structures on dyadic (pairwise) graphs; second, we model higher-order interactions by ordered hy …

Harnessing Llms To Detect Hate Speech, Weibin Cai

Theses - ALL

Hate is a sentiment, while hate speech refers to the expression of hate in a form that targets and attacks specific groups, such as race, religion, or gender. With the rise of the internet and social media, hate speech has spread rapidly, gaining wide exposure and posing threats to individual well-being, the profits of major tech companies, and social stability. As a result, both industry and academia have turned their attention to the study of hate speech. One of the most active areas is hate speech detection, which involves training models to predict whether a given piece of content is …

Noise-Enhanced Network Science, Reyhaneh Abdolazimi

Dissertations - ALL

Graphs are a versatile and powerful data structure used to model complex relationships in diverse domains such as social networks, biological systems, and transportation networks. In graphs, entities are represented by nodes, and interactions or relationships between them are represented by edges. For example, in the World Wide Web, web pages are considered vertices, and if there is a link from one page to another page on the web, there will be a directed edge between those pages in its graph data structure. These structures allow us to analyze and solve important problems like community detection (identifying closely related clusters …

Enhancing Graph Neural Networks By Editing Graphs, Jiayu Li

Dissertations - ALL

Graphs are pervasive in both the natural world and various domains of science and engineering. Numerous advanced classifiers, such as Graph Neural Networks (GNNs), have been developed to perform node classification on these graphs. However, as graphs become denser with an increasing number of edges, GNNs often suffer from suboptimal generalization performance due to the presence of task-irrelevant connections. These redundant connections can introduce noise, consume excessive computational resources, and degrade performance. Identifying and preserving critical connections in large-scale graphs, while pruning unnecessary ones, is crucial for enhancing the efficiency and accuracy of GNNs in node classification, particularly for GCNs. …

Privacy Protection In Machine Learning Via Exploitation Of Constrained Adversarial Evasion, Brian Testa

Dissertations - ALL

Deep neural networks are extensively applied to real-world tasks. In many cases, the data feeding these tasks are human generated content, which emphasizes the criticality of privacy and data protection. The diverse modalities of this user data provide fertile ground for 3rd parties to monetize a user’s data. This work considers two such modalities: user speech when interacting with smart speaker voice assistants (VAs) and images shared with online service providers. In these cases, the user would like to support some form of machine learning (ML) inference without allowing others. A user interacting with a smart speaker would like the …

Memetic Memory As Vital Conduits Of Troublemakers In Digital Culture, Alexander O. Smith, Jordan Loewen-Colón

School of Information Studies - Post-doc and Student Scholarship

Recent fears of data capitalism and colonialism often argue using implicit assumptions about cybernetic technology’s ability to automate data about culture. As such, the level of data granularity made possible by cybernetic engineering can be used to dominate society and culture. Here we unpack these implicit assumptions about the datafication of culture through memes, which both act as cultural data and cultural memory. Using Alexander Galloway’s critical method of protocological analysis and descriptions of media tactics, we respond to fears of cybernetic domination. Protocols – the source by which cybernetic technologies enable automated datafication – enables us to respond to …

Stochastic Algorithms For Simulating Protein Assembly In Biological Cells, Advait Shukla

Theses - ALL

Understanding membrane protein assembly using computational methods is critical in developing drug delivery strategies for tight junction diseases such as Alzheimer’s and Crohn’s diseases. While experimental methods, such as freeze-fracture micrographs, provide microscopic details of the protein strand networks, they cannot offer molecular-level information, which is necessary for drug design. This work attempts to close the gap by computationally predicting strand networks of the claudin family of proteins that form tight junctions across neighboring cells in epithelial and endothelial tissues. Using claudin dimer data collected by the Protein AssociatioN Energy Landscape (PANEL) method coupled with some Markovian approximations, we built …

On Multi-Sensor Adaptive Birth Theory For Labeled Random Finite Sets Tracking, Anthony Trezza

Dissertations - ALL

This dissertation provides a scalable, multi-sensor measurement adaptive track initiation technique for labeled random finite set filters. The lack of a well-defined, systematic approach is problematic for many applications, especially when fusing ambiguous sensor measurements. We begin by showing that a naive solution leads to an exponential number of newborn components in the number of sensors. An efficient solution is derived by formulating a ranked assignment truncation problem. A truncation criterion is established for a labeled multi-Bernoulli random finite set birth density that has a bounded L1 error in the generalized labeled multi-Bernoulli posterior density. This criterion is used to …