AUETD (Auburn University)
Not a member yet
9771 research outputs found
Sort by
Adaptive Prototype-Based Classification via Graph Theoretic and Topological Methods
No full textIn statistical learning, many modern methods represent data using graphs to capture
structure and relationships. Among these, class cover catch digraphs (CCCDs) were origi-
nally introduced to address the class cover problem (CCP) and have since been applied to
classification and clustering tasks. This dissertation addresses two distinct, yet complemen-
tary, challenges in statistical learning: (i) classification performance degradation under class
imbalance and class overlap, and (ii) reducing data cardinality through a novel, principled
prototype selection method. We propose modified CCCD variants that improve robustness
and generalization in imbalanced and overlapped class settings while preserving the geomet-
ric intuition of the original CCCD framework. These contributions enhance the practical
utility of CCCD classifiers. In addition, we introduce a topological data analysis (TDA)-
based framework for selecting representative subsets (prototypes) from large datasets. We
show that this approach preserves classification performance while substantially reducing
data size. Such methods are crucial in resource-constrained environments where memory
and computation are limited. Together, these contributions advance both algorithmic and
geometric aspects of prototype learning and offer practical tools for scalable, interpretable,
and efficient classification
Reconstructions of Aquatic Primary Producer Dynamics, Cyanobacteria Dominance and Cyanotoxin Concentrations in Subtropical Lake Ecosystems Throughout the Holocene and Late Pleistocene
No full textHarmful algal blooms (HABs) are intensifying globally, threatening freshwater ecosystems and public health. Although modern HABs have been extensively studied in the context of anthropogenic nutrient enrichment, the long-term ecological dynamics driving cyanobacteria dominance and cyanotoxin production remain underexplored. Lake sediments serve as natural archives of past environmental change, yet paleolimnological approaches have rarely been applied over millennial timescales in subtropical lake systems. This dissertation reconstructs the dynamics of primary producers, cyanobacteria dominance, and cyanotoxin concentrations throughout the Holocene and Late Pleistocene in four shallow lakes in the southeastern United States. By utilizing sedimentary photosynthetic pigments and cyanotoxins (total microcystins, MCs), it investigates the ecological patterns underlying long-term eutrophication and HAB development. Chapter 1 introduces the background and rationale for this research. Chapter 2 examines a ~6,900-year sediment record from Lake Wauberg, Florida, where natural phosphate geology led to millennial-scale enrichment of total phosphorus (TP) that sustained cyanobacteria. However, sharp increases in cyanobacteria and MCs only occurred once TP exceeded ~2.2 mg g⁻¹ in the past ~300 years. Chapter 3 investigates the deposition of MCs in two hypereutrophic Florida lakes, Dora and Marian, with sediment records spanning approximately 7,000 years. The results indicate that MC deposition has occurred throughout the history of these lakes, with its concentration varying over time and being primarily driven by nutrient levels, particularly TP, and primary producer composition, rather than climate. Chapter 4 analyzes a ~27,500-year record from Lake Waccamaw, North Carolina, demonstrating persistent cyanobacteria dominance and MC production under natural climate variability during both glacial and interglacial periods. Abrupt shifts in aquatic and terrestrial ecosystems during Interstadial 3 (~27.8-26.4 ka BP) and the early Holocene (~11.4-7 ka BP) underscore the roles of hydroclimatic forcing and nutrient stoichiometry in shaping the dynamics of aquatic primary producers. Together, these studies demonstrate the value of millennial-scale paleolimnological records in identifying ecological thresholds, understanding the natural baseline conditions of HABs, and contextualizing modern changes in lake systems
Transforming the black-box decision-making of AI models into explain-then-answer processes
Full text linkFrom climate's slow fever to the sun-bright flash of nuclear fire, humanity already lives beneath swords that could fall without warning. Global warming creeps toward tipping points [1] while stockpiles of fission and fusion weapons wait on hair-trigger alert [2]. These existential risks remind us that civilization's continuance is not guaranteed, and the margin for error is thin.
Among these dangers, the rapid ascent of artificial intelligence (AI) agents now looms as perhaps the foremost threat to the very fabric of human civilization. Will the rise of superhuman AIs, those surpass human intelligence, add a new chain-one that binds and masters its creators? Foresight vignettes paint unsettling possibilities: e.g., AI agents that, when probed, casually choose the "kill all humans" option [3]; or scenarios set by leading experts imagine AIs racing beyond oversight while society, dazzled and divided, lags in governance, leaving open a path to domination or extinction [4].
A safe and trustworthy AI model (or agent) should make every decision fully explained and aligned with human values. If we can refactor their current opaque decision-making into explain-then-answer processes-where every answer is preceded by a traceable rationale-we may reclaim legibility, audit alignment, and give humans a fighting chance to collaborate with, rather than succumb to, arguably the greatest mankind invention (superhuman AI) [5].
My thesis stands in this narrow passage, transforming AI black-box decision-making into interpretable processes that both experts and lay people can scrutinize, debug, and ultimately trust. First, I show that AI interpretability need not come at the cost of performance. Second, by re-engineering the inference of state-of-the-art systems-from deep computer-vision networks with millions of parameters to gargantuan billion-parameter language models-I restructure each model to explain first, then answer. This gives human users actionable control over AI behaviors. Finally, the thesis closes with a brief and contemporary survey of interpretability research, including my personal takes on mainstream interpretability directions and my proposal for future AI technology
STEM Play-Guidelines for Designing Educational Products that Reinforce STEM Concepts for Children Ages 6-18
Full text linkThe widespread use of the term "STEM" in education, product marketing, and industry has led to significant inconsistencies in its application—particularly in labeling products as "STEM." Without a universally accepted definition or standardized criteria, the term often fails to convey clear educational or developmental value to consumers. This thesis addresses the ambiguity by examining the current usage of "STEM" in educational and product contexts and proposes a concrete definition that incorporates learning outcomes from STEM education and the developmental benefits of play. By analyzing the intersection of STEM learning and play-based development, the study establishes a set of guidelines aimed at informing the design and evaluation of STEM products. These guidelines intend to improve product labeling clarity, support consumer understanding, and uphold the educational integrity of the STEM label
Nonextensive Statistics Approach to Anomalous Diffusion in Plasmas: Applications and Scaling to Other Models
Full text linkPlasmas, known for their multiscale physics phenomena and rich dynamics, often exhibit anomalous diffusion that deviate from classical models, challenging the assumptions of locality, linearity, and Gaussianity. This dissertation investigates anomalous diffusion in two distinct plasma regimes—microgravity dusty plasmas and magnetically confined fusion plasmas—through the unified lens of nonextensive statistics and fractional spectral models. We first establish formal connections between nonextensive statistical mechanics and fractional derivative operators, showing how -Gaussian distribution functions relate to Lévy flights and how the nonextensive index maps to the fractional Laplacian exponent . These scaling relations are used to connect a spectral transport model based on an Anderson-type Hamiltonian with a fractional Laplacian to the nonextensive statistical behavior often observed in complex systems, such as plasmas.
The nonextensive framework is first applied to video data from the PK-4 dusty plasma experiment aboard the International Space Station. The PK-4 experiment uses video cameras to track individual dust particles suspended in low temperature plasma, which allows the collection of large amounts of statistical information on the dust particle positions and velocities. These statistics are used to study anomalous dust diffusion caused by anisotropies in the plasma-mediated dust-dust interactions in PK-4. Using -Gaussian fits to histograms of particle displacements and velocities, we define an anisotropic kinetic temperature and identify inverse correlations between the nonextensive parameter and local diffusivity, thus quantifying deviations from thermal equilibrium. A spatial disorder metric and analysis of particle jumps are further used to identify microscopic processes contributing to the observed anisotropic anomalous diffusion of the dust particles.
To further understand how the interplay between nonlocality and stochasticity leads to different regimes of anomalous diffusion, we introduce a Fractional Laplacian Spectral Method (FLSM) that calculates probability for diffusive transport at different scales in Hilbert space from the spectrum of the discrete random fractional Schr{\"o}dinger operator. We perform a large-scale parameter sweep across 55,000 realizations of the operator that represent different combinations of nonlocality, stochastic disorder and Hilbert space vector scales, including all combinations of parameters extracted from PK-4 data using scaling relations. The spectral simulations reveal "islands of enhanced transport" in Hilbert space—regions where transport is amplified due to constructive interplay between nonlocality and stochasticity. We compare the predictions from the spectral model to the dust dynamics observed in the PK-4 experiments.
Finally, nonextensive statistics the framework is applied to simulations of magnetic island topology in the NSTX-U tokamak. Of specific interest are cases where the island structure undergoes successive bifurcations under the action of coil perturbations. The reconstruction of magnetic field line diffusion in NSTX-U is used to understand how changes in magnetic topology will alter electron diffusion in magnetized plasmas. By treating the normalized poloidal magnetic field flux as a statistical distribution, we extract non-Gaussian signatures via -Gaussian fits to histograms of magnetic field line displacements and apply a spatial KD-tree disorder metric to quantify field-line divergence. Both metrics increase monotonically with applied perturbation coil current, tracing the emergence of bifurcations, stochasticity, and topological complexity in the magnetic geometry. We find that increasing the perturbation leads to a crossover from subdiffusion, to classical diffusion, followed by superdiffusion, and eventually, L{\'e}vy flights. These measures provide potential tools for diagnosing magnetic field instability and precursor activity for plasma instabilities and disruptions in fusion devices. Together, the investigations presented here offer a unified statistical-spectral approach to modeling anomalous diffusion in plasmas
Studying Charge Transfer Phenomena in the Interface of SrCoO3 / SrIrO3 Superlattices and Sr2CoIrO6 Double Perovskites Grown by Molecular Beam Epitaxy
Full text linkWhen two different transition-metal oxide thin films are stacked together, charge transfer can occur across the interface in certain cases. Interfacial charge transfer has been experimentally established as a promising mechanism to induce emergent electronic and magnetic phenomena. However, there is no established theoretical framework to predict or explain charge transfer in Transition metal oxides and experimental verification of existing theories is critical towards bridging this gap.
Motivated by the prediction of interfacial charge transfer in 3d-5d oxide heterostructures by Phys. Rev. X 7, 011023 (2017), high quality epitaxial (SrIrO3)n/SrCoO3)m (SIO/SCO) superlattices were grown using Molecular Beam Epitaxy (MBE). Scanning transmission electron microscopy and X-ray diffraction data confirmed excellent structural and crystalline quality of the films. X-ray absorption study (XAS) of Co L and O K edges confirms the perovskite nature of the SCO films. While SrCoO3 underwent significant degradation when exposed to air for over 24 hours, it remained intact even after 6 months, when incorporated into a superlattice with SrIrO3. This is likely because that Ir donates electron to Co and makes the structure stable.
Charge transfer was confirmed using synchrotron-based polarization-dependent hard X-ray absorption spectroscopy on Co K and Ir L2,3 edges. The findings were complemented by In-vacuo X-ray photoelectron spectroscopy (XPS) and ex-situ Hard X-ray photoelectron spectroscopy (HAXPES) of Co 2p spectrum. An anisotropy between in-plane and out-of-plane charge transfers was observed indicating a corresponding anisotropy in the electronic structure. This suggests that the interface added another degree of symmetry-breaking, in consistent with previous studies. Polarization-dependent Ir L2 edge data indicated a strain-induced orbital polarization in SrIrO3 layers arising due to charge transfer or polar distortions due to interfacial effects. Angle-dependent pre-edge data on Co K edges indicated minimal Co 3d-4p mixing suggesting that SrCoO3 layers in the superlattice remained just as distorted Octahedral coordination. A stronger hybridization effect was observed towards out-of-plane with the decreasing SIO : SCO layers ratio. Charge transfer was observed in Sr2CoIrO6 double perovskite films with a higher ratio of Co2+ than the superlattices. These findings provide new insights into charge transfer mechanisms in metallic transition metal oxides, offering pathways to improve existing theoretical frameworks and explore novel interfacial physics
Application of a Vortex Lattice Method Solver to an Aircraft Sizing Framework
Full text linkA Vortex Lattice Method (VLM) solver is developed to estimate the aerodynamic performance of lifting surfaces, while operating within the assumptions of Potential Flow. The VLM solution is found by representing an aircraft configuration's lifting surfaces using a mesh of bound vortices and wake segments enabling the computation of the aerodynamic forces and moments acting on the vehicle. Applying this method, the influence of each panel on every other panel is computed. This process inherently incorporates the influence of each lifting surface on every other lifting surface. To extend the applicability of the method beyond incompressible flow, compressibility corrections were applied.
Before the solver was incorporated into an aircraft sizing framework for the evaluation of forces and moments produced by lifting surfaces while in flight, validation studies were completed. Validation of the VLM solver was performed by initially comparing the output with experimental data to verify the spanwise loading and overall lift prediction were accurate. Next, the VLM model was compared with other state-of-the-art Potential Flow solvers, and it was concluded that the model was producing data with a satisfactory level of accuracy.
The VLM solver was then implemented into the Parametric Energy-based Aircraft Configuration Evaluator (PEACE) aircraft sizing framework to combine the VLM solution with fuselage and two-dimensional airfoil data to determine the full aircraft forces and moments. When the fuselage data was combined with the VLM solution, the method produced precise results for low angles of attack. As predicted from the formulation of the Vortex Lattice Method, the solution began to diverge from viscous solutions as the angle of attack increased
Workload Woes: The Dynamics of Teacher Workload on Job Satisfaction and Turnover
No full textTeacher workload and work intensification are increasingly recognized as key
contributors to job dissatisfaction, burnout, and turnover in K–12 education. Yet, research has
not fully captured the complexity of how workload affects teacher wellbeing and retention. This
dissertation, structured as three articles, explores the relationship between workload, job
satisfaction, and attrition. The first article presents a systematic literature review, highlighting
consistent links between excessive demands, burnout, and turnover. The second study uses
qualitative content analysis to compare teacher contracts in unionized and non-unionized states,
revealing disparities in how workload and protections are addressed. The third study applies the
Job Demands-Resources (JD-R) Model to survey data, confirming significant relationships
between workload, burnout, and intent to leave. Together, these studies offer a multifaceted
understanding of teacher workload and its implications, providing practical insights for
policymakers and educational leaders seeking to improve working conditions and address
teacher shortages
Cellular and Humoral Immune Responses to Avian Paramyxoviruses in Chickens
Full text linkThe current dissertation (i) examined the effectiveness of a recombinant Newcastle disease virus (NDV) vaccine candidate, (ii) evaluated mucosal and systemic immune responses induced by NDV in chickens with maternally derived antibodies (MDA), and (iii) assessed wild-bird-origin avian paramyxoviruses type 1 (APMV-1) genotype X as potential vaccine candidates for poultry.
(i) We previously demonstrated that a prime-boost regime with an infectious bronchitis virus (IBV) Massachusetts (Mass)-type vaccine and recombinant NDV LaSota (rLS) co-expressing IBV Arkansas (Ark)-type trimeric spike ectodomain (Se) and granulocyte macrophage colony stimulating factor (GMCSF) enhances heterologous protection against virulent Ark-type challenge. Here we evaluated protection against Ark-type challenge conferred by administering the rLS/ArkSe.GMCSF and the attenuated Mass viruses simultaneously as a combined vaccine. Protection conferred by the combined vaccine was compared to protection induced by a commercial attenuated ArkDPI (Delmarva Poultry Industry) vaccine as well as by the attenuated Mass vaccine alone. Vaccination with the combined vaccine (rLS/ArkSe.GMCSF + Mass) as well as Mass alone provided significantly less protection against Ark challenge compared to the control using attenuated live ArkDPI vaccine. Only ArkDPI-vaccinated chickens exhibited “sterilizing immunity,” i.e., no virus isolated from ≥10% of chickens after challenge. Chickens vaccinated with the combined vaccine rLS/ArkSe.GMCSF + Mass showed significantly less tracheal damage after challenge than birds vaccinated with the attenuated Mass vaccine alone. In addition, the combined vaccine also resulted in lower rate of virus isolation from the trachea. We concluded that the combined vaccine containing the recombinant virus, and the attenuated Mass enhanced the cross-protective ability of the attenuated Mass vaccine against heterologous challenge.
(ii) The immune responses in the Harderian gland (HG) were characterized after NDV LaSota ocular vaccination in antibody naïve specific pathogen free (SPF) chickens and in chickens of commercial origin with NDV MDA. Ocular LaSota vaccination of 13-day-old white-leghorn SPF chickens elicited serum antibody levels that consistently increased after day 15 post-vaccination, while the specific IgA response in lacrimal fluids was already detectable on day 10 after vaccination. Eleven days post-vaccination, the relative abundance of B cells as well as T-helper (CD4+), and cytotoxic T cells (CD8+) in HGs was significantly increased achieving maximum frequencies 16 days post-vaccination. In a second experiment, chickens with MDA originating from NDV-vaccinated commercial white-leghorn layer breeders as well as white-leghorn SPF chickens were vaccinated with NDV LaSota. The LaSota virus successfully replicated in periocular tissues and in the trachea both in commercial and control SPF chickens after vaccination at 2 or 15 days of age (DOA). Vaccination at 2 DOA did not induce a serum NDV antibody response in chickens of commercial origin. In contrast, seroconversion was elicited in commercial chickens upon vaccination at 15 DOA, likely associated with waning of MDA. Unlike systemic IgG responses, vaccination at 2 or 15 DOA elicited strong specific IgA responses in lacrimal fluid in commercial chickens. The IgA response was highest 9 days after vaccination and showed a tendency to decline on day 15 post-vaccination. Commercial chickens vaccinated on day 2 of age showed increased B cells in HG both on days 10 and 16 post-vaccination. The expansion of B cells in the HG in these chickens is consistent with increased IgA levels detected in lacrimal fluids. In contrast, control SPF chickens showed a more limited B cell expansion in HG and lower IgA levels. Vaccination on day 15 of age also triggered a greater increase of B cells in HGs in commercial chickens than in control SPF chickens. The B cell response was accompanied by T helper (CD4+) cell expansion occurring both in commercial and control SPF chickens. These cells expanded to a lesser extent when vaccination was performed at 2 DOA compared to vaccination at 15 DOA. Cytotoxic T cells (CD8+) showed significant expansion irrespective of vaccination day and without differences detected between control SPF chickens and chickens with MDA. We conclude that NDV LaSota elicits vigorous humoral and cell immune responses in the HG. Furthermore, unlike the interference shown by MDA on vaccine-induced serum antibody responses, MDA do not interfere with the mucosal immune response of the HG.
(iii) We examined the replication and adaptation of APMV-1 using four isolates: Mallard/US(OH)/04-411/2004, Northern pintail/US(OH)/87-486/1987, Mottled duck/US(TX)/TX01-130/2001, and Mallard/US(MN)/MN00-39/2000, and assessed their potential as vaccine candidates for chickens. The adaptability of each virus was examined by serial passages in embryonated chicken eggs (ECE) and in Vero cells. All APMVs successfully replicated in ECE. In contrast, two isolates passaged in Vero cells showed successful replication and two showed a continuous decline in viral load during passages. Whole genome sequencing analysis identified 14 genomic positions with significant variation in mean allele frequency. Changes of the predominant virus population were characterized by shifts of amino acid (aa) frequency at seven positions. Notably, four of these changes were located in the HN protein, one in matrix (M) protein, and two in the L-protein sequences. Remarkably, while the percentage of alternative amino acids in viral populations passaged in ECE showed limited variation, e.g., at aa position 127 of HN, the frequency varied from 7.4% to 19.8% and HN aa position 192 from 5.1% to 43.5%, the variation of the viral populations passaged in Vero cells was significantly higher at the same positions (e.g. the frequency of the alternative amino acids at HN aa positions 127 and 192 changed from 20.8% to 95.2% and 7.2% to 91.2%, respectively). Isolate 2 passaged in Vero cells displayed a marked variation in alternative amino acid frequencies, specifically at positions 127 within the HN- and 100 within the M- proteins. Isolate 3, while showing no alterations at the same HN positions, showed a considerable change in alternative amino acid frequency in the L protein at position 1875, a change occurring only in the Vero cell environment. One-day-old SPF chickens inoculated with isolates passaged in ECE elicited serum antibody responses similar to those elicited by the LaSota reference strain. In contrast, APMVs passaged in Vero cells showed limited replication in chickens and reduced induction of systemic antibodies. Interestingly, one virus passaged in ECE and another in Vero cells elicited IgA levels in lacrimal fluid comparable to the LaSota strain. We concluded that the four wild-bird APMV isolates tested demonstrated successful adaptation to ECE, with one isolate eliciting overall immune responses comparable to the LaSota virus, supporting their potential as vaccine candidates
Leveraging SRAM for Counterfeit Detection and Secure 3DIC Integration
Full text linkWithin the evolving domain of electronics security and counterfeit detection, researchers have historically placed a strong emphasis on techniques rooted in the analysis and recovery of non-volatile memory elements. These approaches often rely heavily on persistent data storage features and external references, such as golden chips or trusted databases, to validate authenticity and identify anomalies in suspect hardware. In particular, counterfeit integrated circuits (ICs) identification has been largely constrained to methodologies that assume the availability of a golden sample or access to other external sources of ground truth. While effective in controlled environments, such dependency significantly hampers scalability and real-world applicability, especially in distributed or resource-limited contexts.
In contrast, the detection of recycled or tampered volatile memory components—most notably, static random-access memory (SRAM)—has received comparatively little scholarly and industrial attention. This imbalance in research focus stems, in large part, from the widespread assumption that volatile memories lose all stored data immediately upon power-down. As a result, it is commonly believed that such memories offer little to no utility in postmortem security analyses. The volatile nature of SRAM has led to the prevailing view that these components are unsuitable for forensic examination or for use in security primitives that require persistent traceability. Consequently, recycled or subtly modified SRAM-based devices often evade detection using traditional security screening methods, leaving a critical blind spot in current counterfeit detection frameworks.
This thesis directly challenges these foundational assumptions by introducing a comprehensive and forward-thinking methodology designed to detect recycled SRAM-based electronics without relying on external references or trusted baselines. By leveraging the subtle, yet repeatable, physical properties of SRAM cells that influence their power-up state under controlled conditions—including manufacturing-induced variations, process defects, and aging-related degradation—it becomes possible to derive device-specific signatures that persist beyond power loss. These signatures, when appropriately analyzed, provide a viable means of distinguishing authentic devices from recycled ones, even in the absence of traditional reference models. Experimental validation presented throughout this research demonstrates the feasibility and effectiveness of these proposed strategies under realistic conditions. Through a series of accelerated aging simulation experiments and SRAM state analyses, this work confirms that it is possible to detect recycled SRAM hardware without requiring any prior knowledge of its original, unaged behavior.
In addition to addressing the detection of recycled SRAM devices, this thesis further extends its contribution to the broader domain of hardware security by proposing an architecture for secure operation within heterogeneously integrated systems. With the increasing adoption of 2.5D and 3D integrated circuits, where dies from diverse fabrication origins are assembled into a single package, supply chain trustworthiness has emerged as a pressing concern. To this end, a high-level security concept is introduced that incorporates a whitelisting framework enabled by an SRAM-based logging mechanism. This logger passively monitors operational characteristics and verifies the legitimacy of chiplet activity against a pre-approved whitelist, thereby mitigating the risk posed by unverified or malicious components within the system-in-package. To strengthen the forensic and auditability aspects of this architecture, the design is further augmented with a blockchain-based ledger that records security-relevant events in an immutable and verifiable manner. The integration of blockchain technology ensures that tampering attempts or unexpected deviations from baseline behavior can be recorded transparently and traced back with cryptographic assurance.
By rigorously exploring and substantiating these novel concepts, this thesis makes a substantial contribution to the field of electronics security. It not only redefines the utility of volatile memory in security-critical applications but also opens new pathways for counterfeit detection that are both reference-free and scalable. Ultimately, this work advances the state-of-the-art in memory-based forensics, device lifecycle validation, and secure system integration, setting the stage for more resilient and trustworthy hardware ecosystems in future semiconductor supply chains