8413 research outputs found
Sort by
Exploring regionalization techniques: Simulation and application in ecology
No full textIn this thesis, I examine regionalization algorithms in the context of spatial discretization for ecological research. Regionalization refers to the process in which areal units are created based on spatial relationships between smaller polygons and the similarity or dissimilarity of characteristics between those polygons. Chapter 2 consists of a non-exhaustive literature review detailing multiple regionalization algorithms and cluster optimization or scoring indices, briefly discussing their history, applications within the literature, and the statistical and computational methods underlying their operation. Cluster optimization indices are used to determine how many clusters are best for a certain set of data and a given regionalization algorithm. Chapter 3 applies these algorithms and indices to three simulated datasets, or landscapes, generated with low, medium, and high amounts of variation. There, I compare the performance if different combinations of regionalization algorithms and cluster optimization indices across this range of landscape variation. I demonstrate the tremendous variety of outcomes that different combinations of algorithms and cluster optimization indices can produce. I discuss the differences in clustering between each algorithm/index combination both generally and according to level of variation in the simulated data and compare how similar or dissimilar of results each algorithm produces. Results indicated that the AZP and REDCAP algorithms tended to produce regionalization schemas with the best fit, regardless of the level of variation in the simulated landscape. Among clustering algorithms, the SD index tended to optimize regionalization algorithms to have a number of clusters with the highest goodness of fit. However, this high goodness of fit may result from a large amount of clusters being recommended by the SD index. Chapter 4 uses one of the regionalization algorithms, REDCAP, in with real data to discretize the Prairie Pothole Region in North America. I data relevant to the management of waterfowl species in the area and then discuss the importance of non-statistical science in the evaluation of algorithm results. While often treated as a unified area to waterfowl ecologists, this region shows a tremendous amount of landscape diversity. This diversity may generate differences in wildlife population dynamics across the region. As such, creating discrete spatial units in this area, based on relevant ecological data, may aid ecologists studying waterfowl population trends. The final results indicate that 25 regions in the PPR may exist in which it is reasonable to separately study waterfowl populations
Distributed Machine Learning in Heterogeneous Edge Networks
No full textThe development of edge devices (e.g., smartphones, IoT sensors, and wearables) has led to a surge in data generated at the network's edge. Meanwhile, the complexity of machine learning models has increased significantly, with state-of-the-art models for tasks like natural language processing and computer vision now containing billions of parameters.Distributed machine learning addresses the challenges posed by massive datasets and high computational demands by distributing the training process across multiple devices. This approach enables parallel computation and reduces the overall computational burden. However, distributed machine learning poses unique challenges, including the straggler problem (i.e., where slow devices hinder training progress), communication overhead (i.e., the high cost of transferring data between devices), and device heterogeneity (i.e., variability in device capabilities).
This dissertation explores the challenges of distributed machine learning in heterogeneous edge networks, focusing on the design and optimization of algorithms to accelerate training while enhancing resource efficiency.
This work's key contributions include the development of Dynamic Tiering-based Federated Learning (DTFL), Communication-Efficient Decentralized Multi-Agent Learning (ComDML). DTFL is a federated learning algorithm designed to address the inherent heterogeneity of edge networks, where devices vary widely in computational power, communication bandwidth, and task sizes. By dynamically assigning clients to tiers based on their capabilities, DTFL mitigates the straggler problem and accelerates training. Clients in each tier offload portions of the global model to a central server, enabling parallel updates through split learning and local-loss-based training. A dynamic tier scheduler continuously profiles clients, estimating training times based on observed resource metrics such as network speed and dataset size, and adjusts tier assignments accordingly. This low-overhead approach prevents straggler problems and ensures efficient resource utilization in dynamic environments.Extensive experiments with DTFL on large models such as ResNet-56 and ResNet-110, using datasets like CIFAR-10, CIFAR-100, CINIC-10, and HAM10000, validate its efficacy. Results demonstrate up to an 80\% reduction in training time compared to advanced federated learning methods while maintaining model accuracy. DTFL reduces training time in both IID and non-IID data settings and maintains high performance even under privacy-preserving measures. Theoretical analysis further demonstrates its convergence properties across convex and non-convex loss functions. ComDML extends DML to decentralized scenarios, eliminating the need for a central aggregator. By doing so, ComDML operates in peer-to-peer configurations, enhancing resilience and security by removing single points of failure. In heterogeneous networks with varying computational and communication resources, ComDML optimizes training efficiency by enabling slower agents to offload tasks to faster ones, ensuring a more balanced distribution of workloads. A decentralized pairing scheduler dynamically matches agents based on their real-time capabilities using lightweight profiling to assess communication overhead, computation capacity, and task size for each pairing, thereby reducing idle times for faster agents. Through local-loss-based split training, where agents train different portions of a model concurrently, ComDML mitigates the communication synchronization bottlenecks typical of traditional split learning. To further enhance communication efficiency during workload offloading, ComDML incorporates SplitPair, a technique that compresses the intermediate feature maps exchanged between paired agents using Singular Value Decomposition (SVD). SplitPair applies a dynamic rank adjustment mechanism that starts with a low SVD rank for initial communication savings and increases it only when model accuracy plateaus, optimizing the trade-off between communication cost and model fidelity throughout training. Experimental results show significant improvements, reducing training time by up to 71% while maintaining model accuracy. ComDML can seamlessly integrate privacy-preserving techniques, such as differential privacy, without substantial performance loss, demonstrating its robustness in dynamic, edge network environments
Social-Psychological Sources of Resistance to Eating Plant-Based Protein: The Effects of Framing and Masculinity
No full textOver recent years, there has been growing awareness of the effects of the widespread, rapid production and consumption of meat, including risks to personal and public health and environmental pollution. Reducing the public's meat consumption has not kept pace with concern over these issues, a pattern this proposal sought to understand better. Extensive research has examined structural and personal barriers, such as policy, cost, taste preferences, perceived nutritional inadequacies, cultural upbringing, and the social-psychological barriers addressed in this thesis proposal. While literature is abundant on such barriers, there is far less research on how different message framings influence acceptance or resistance to eating more plant-based. Although research specifically linking framing to plant-based diets is limited, prospect theory, well established in behavioral economics and social psychology, provides a strong foundation for designing persuasive dietary messages. This study examined how gain and loss message framing influenced willingness to eat more plant-based and reduce meat consumption and tested whether masculinity moderates these effects. Cisgender men were recruited through Prolific to take a brief survey, and were randomly assigned to a gain-frame, loss-framed, or control message condition, followed by questions about their willingness to change their diet, their masculinity beliefs, and behavioral and demographic covariates. Regression analysis revealed that the loss-frame message increased willingness to change diet, but neither the gain-frame message nor masculinity had any effect. This research contributes to the literature on the role of message framing in behavior change with implications for diet, public health, and behavior change messaging
Toward Interpretable, Data-Driven Frameworks for 3D Point-Cloud Analysis in Forest Ecology
No full textThree-dimensional point clouds have transformed our ability to capture natural environments at scale, offering detailed representations of forest structure, canopy architecture, and understory composition. Yet despite this promise, the use of deep learning on ecological point clouds remains limited by three challenges: the scarcity of annotated data, reliance on models developed for non-ecological domains, and feature representations that lack biological interpretability. This dissertation addresses these challenges through three contributions. First, it introduces a modular pipeline for generating synthetic vegetative point clouds at scale, combining procedural tree models, randomized environmental factors, and simulated LiDAR acquisition. This pipeline alleviates data scarcity by enabling the creation of diverse, labeled datasets spanning species, growth stages, and canopy conditions. Second, it advances point-cloud architectures with two innovations: semantics-aware diffusion conditioning, which generates high-fidelity point clouds with per-point labels, and PointRTD, a noise-robust transformer whose regularizer improves stability under occlusion while enhancing class separation. This design accelerates convergence during training and improves discriminative accuracy on standard benchmarks. Third, it proposes an interpretable embedding framework inspired by L-systems, aligning latent spaces with parametric growth descriptors such as branching angles and crown spread. This framework regularizes reconstructions while producing embeddings that map directly to ecologically meaningful traits. Together, these contributions provide a unified framework for ecological machine learning that mitigates data scarcity, reduces model-domain mismatch, and grounds latent representations in biological structure. The result is a set of methods that both advance the state of 3D machine learning and open new possibilities for ecological analysis
Evaluation of the Cortical Subplate in Autism Using Postmortem Tissue and Machine Learning as an Indicator of Long-Range Connectional Alterations in Development.
No full textAutism spectrum disorder (ASD) is a neurodevelopmental condition characterized by challenges in social communication and repetitive behaviors. These behaviors have been suggested to arise from an imbalance in brain connectivity, specifically between long- and short-range connections in ASD. The development of these crucial connections begins with subplate neurons (SPns), which are vital for forming and maturing brain circuits from early development into adulthood. Located beneath the cortical plate, SPns are essential for organizing the cerebral cortex. Similarly, glial cells—including astrocytes, microglia, and oligodendrocytes—are critical for establishing healthy neural circuits and connections between different cortical regions. Studies on other neurodevelopmental and neurodegenerative disorders, such as Alzheimer's disease, have found abnormalities in both the number and shape of glial cells. Given the strong evidence for altered connectivity in autism, it is plausible that dysfunctions in both SPns and glial cells contribute to the abnormal neural networks seen in ASD. However, our understanding of these cells' specific roles is limited due to a lack of research on this particular brain region. This project contains three aims: Aim 1 assessed SPns changes in ASD and neurotypical (NT) control subjects in the subplate of the parietal cortex. Aim 2 evaluate glial subtypes and their differences between ASD subjects and NT subjects within the subplate of the parietal cortex. Aim 3 evaluate the efficacy of recently developed, and commercially available, machine learning techniques for quantitative classification of both neurons (Aim 1) and glial cells (Aim 2) relative to more traditional statistical analysis. Analysis for Aim 1 and 2 suggest cellular morphology and texture revealed preliminary differences in the subplate and layer VI between ASD and NT with a main effect in diagnosis for intensity and texture and in cell bodies across layers, but no significant layer-by-diagnosis interactions were found. Results for Aim 3 quantification and classification using machine learning software revealed a greater number of neurons in the NT subplate compared to the ASD group, and a significantly greater amount of glia in ASD for layer VI compared to the NT group. Cluster analysis attempts to distinguish cell types resulted in poorly rated and imbalanced clusters that did not align with expected glial/neuronal ratios, thus limiting comparison with machine learning classification results. This research, validated by cutting-edge machine learning analysis, delivers robust, initial quantitative evidence of a specific cellular pathology underlying the ASD parietal cortex. The discovery of diagnostic-specific alterations in neuron and glia counts within the subplate and layer VI establishes a critical, anatomical basis for the widely accepted etiology of cortical unbalance in ASD. These pivotal findings empower researchers with new, quantifiable data points, significantly improving our cellular understanding of this key etiology and accelerating future mechanistic investigations
Enhancing Processing Strategies of Claystone Ores: Li-Bearing Smectites from Oregon, USA
No full textThe demand for lithium has dramatically increased in the last few decades, as the unique metal finds itself at the forefront of efforts to produce cleaner energy across the globe. Meeting the demand requires new sources of lithium, spurring an interest among the mineral processing community in Li-bearing sedimentary claystones, an ore type that has never been processed on a commercial scale, largely due to the notoriously difficult nature of processing clay minerals. However, both academic and industrial leaders alike have recently begun tackling this challenge, a collaborative effort that continues here. This study highlights lithium-bearing smectitic ores obtained from the McDermitt Caldera in southeastern Oregon, USA, that have never before been analyzed in a metallurgical setting. Eight total samples were studied, two of each from four ore categories: 1) high grade, 2) rich clay, 3) clay-carbonate, and 4) miscellaneous. A “scoping study” approach was adopted for the claystones, through which they were subjected to experiments across multiple stages of a conventional mineral processing flowsheet to inform future research on related materials. The Li-claystones were tested in four major stages: 1) characterization, 2) comminution, 3) beneficiation (via enhanced gravity concentration), and 4) extraction (via direct acidification). Characterization using x-ray diffraction (XRD) and inductively-coupled plasma, mass spectrometry (ICP-MS), revealed that the clay mineral phase in all samples was smectite with no observable illitic components, and major gangue minerals were calcite, alkali feldspars, quartz, and zeolites. Comminution through attrition scrubbing and rod milling were both used to grind the ores for separate tests in Falcon enhanced gravity concentration and direct acidification. Samples varied in recovery after beneficiation with the Falcon, with a maximum Li recovery achieving over 90%, while most samples fell in the range of 50 – 60%. Recoveries for attrition-milled samples were slightly higher than that of the rod-milled samples at averages of 62.6% and 55.2%, respectively. In direct acidification experiments, 2 M sulfuric acid was used to leach the claystones. Most samples saw > 99% Li extraction after just 30 minutes of leaching at 80 °C, and rod-milled samples consistently recovered lithium slightly faster than attrition samples, although they achieved similar recoveries. Results show that both attrition milling and rod milling are effective strategies for grinding the ore and that it is amenable to sulfuric acid leaching. Future studies are recommended to focus on optimization of enhanced gravity separation parameters
Revisiting the Measurement and Analysis of the Implicit Relational Assessment Procedure at the Individual Organism Level
No full textImplicit Relational Assessment Procedure (IRAP) research has historically had a predominant focus on group designs to evaluate data. This dissertation started with Study One, a group design experiment involving training verbal relations among experimental symbols. Predictions were generally met when analyzing the group data, but when individual data was considered, the proportion of data meeting predictions was lower than expected. As such, Study Two, the focus of the dissertation, analyzed data at the individual organism level using two different extended IRAPs intended to be more suitable for individual organism data. One version had substantially more blocks of trials than typically used in IRAP research. Study Two was primarily concerned with comparing different scoring calculations with respect to stability and predicted data patterns based on the Differential Arbitrarily Applicable Relational Responding Effects (DAARRE) Model. The different calculations included latency, rate, accuracy, slope and probability analyses. In addition to calculating standardized difference scores, raw difference scores, block data, and individual trial data were analyzed. These data were subsequently evaluated in terms of quantifiable variability formulas, including the percentages of deviation, and coefficient of variation. Finally, different aggregation methods were evaluated in terms of DAARRE model predictions. The different IRAP versions, aggregation methods and formulas for analyses resulted in varying degrees of stability and in differences in the extent to which DAARRE-based predictions were met. Several methods resulted in adequate stability levels and yielded a higher proportion of DAARRE-based predictions being met for individual participant data. Implications for the expansion of single subject approaches in IRAP research and proposed strategies are discussed
Investigating the cellular and molecular mechanisms by which YPEL regulates synaptic function in Drosophila
No full textThe Yippee-like (YPEL) gene family is evolutionarily conserved across species, from yeast to humans. It consists of five members: YPEL1 through YPEL5. The high degree of conservation among YPEL genes across diverse species suggests that they serve important biological functions. Various studies have shown that YPEL proteins are involved in processes ranging from cellular senescence to tissue development. Previous studies suggest that YPEL plays a role in nervous system function; however, the underlying cellular and molecular mechanisms remain poorly understood. In Chapter 2, we demonstrate that mutations in YPEL result in defects in neuromuscular junction development. We further demonstrate that YPEL negatively regulates p62 puncta formation in both motor and sensory neurons without altering total p62 protein levels. Notably, reduced p62 gene dosage significantly alleviates NMJ defects. Moreover, we discovered that reducing CncC activity in Drosophila fully rescues the neuromuscular junction defects caused by the YPEL mutation. CncC is the Drosophila ortholog of mammalian Nrf2, which functions as a transcription factor that regulates the stress responses, particularly oxidative stress. In addition, we observed that loss of YPEL function reduces reactive oxygen species levels in motor neurons. Collectively, these findings uncover a novel cellular and molecular mechanism underlying synaptic development, in which YPEL regulates NMJ formation through the p62-Nrf2 antioxidant pathway in Drosophila.
In Chapter 3, we demonstrate that YPEL physically interacts with F-box protein FBXL2, a component of the Skp1-Cullin-F-box ubiquitin ligase complex, responsible for substrate recognition and plays a crucial role in the ubiquitin-proteasome pathway. We demonstrate that mutations in FBXL2 lead to neuromuscular developmental defects. Notably, double mutations of YPEL and FBXL2 do not produce additive NMJ defects, suggesting that they function in a common pathway. Furthermore, similar to YPEL, FBXL2 negatively regulates p62 puncta formation in neurons, and reducing p62 gene dosage in the FBXL2 mutant background ameliorates the NMJ defects. Collectively, these findings suggest that YPEL regulates neuromuscular junction development by interacting with FBXL2 to modulate p62 dynamics
Determinants of Willingness to Seek Mental Health Support: A Comparison of Mexican/Mexican-American and White College Students
No full textMexican/Mexican-American (M/MA) college students consistently underutilize mental health services compared to their White peers, despite experiencing comparable or higher levels of stigma, perceived need, and depression symptoms. To understand the cultural and structural factors contributing to this disparity, this study analyzed data from the 2023 – 2024 Healthy Minds Study to examine how M/MA undergraduate college students seek mental health support and how these predictors differ from White undergraduate college students. They study examined whether personal stigma, public stigma, perceived need, knowledge of formal services on campus, mental illness diagnosis, and current mental health symptoms predicted informal and formal help-seeking as well as how they differ between each group. Independent sample t-tests determined that M/MA students utilize both informal and formal support less than their White peers. Logistic regression models indicated that M/MA students reported higher stigma, had lower perceived need of help, and had less knowledge of mental health services available on campus. Interaction effects further showed that personal stigma had a stronger negative effect for M/MA students while perceived need had a positive effect for both kinds of students, but greater for White students. These findings highlight the influence of cultural and structural barriers on help-seeking intentions while promoting the need for culturally responsive mental health outreach and support systems within higher education
Tectonic and Environmental Influences on Crustal Deformation in the Western US
No full textThis dissertation studies crustal deformation in the western United States using secular and seasonal components of high-precision GPS geodesy. The western U.S. is a zone of active deformation with diverse geological and climatic characteristics, covered by a dense network of GPS stations. It serves as an ideal natural laboratory to address a variety of fundamental questions on fault mechanics, deep magmatic intrusions, as well as the interaction of the environment with the solid earth. The first chapter examines the secular part of the signal in the Walker Lane, a zone of transtensional shear along the west margin of the Basin and Range. A comparison of two competing fault models reveals a profound lack of evidence that locked faults accumulate strain across their surface expressions. Instead, horizontal velocities show uniformly linear deformation, demonstrating that the geodetically observed deformation reflects distributed shear below the brittle crust rather than discrete fault dislocations. This has major implications for seismic hazard assessment, which relies on geodetic strain accumulation rates to estimate earthquake potential on individual faults.The second chapter examines the seasonal part of the GPS signal to demonstrate that hydrologically driven seasonal crustal strain modulates the timing of deep dike openings. The study analyzes all three known deep magmatic intrusions in the western U.S. and finds that each is optimally oriented to receive seasonal reduction in normal stress from hydrological loading, helping the magma to overcome the host rock's resistance and move through the lower crust. Stress reductions of less than 1 kPa are sufficient to facilitate dike opening, demonstrating the sensitivity of deep crustal processes to small environmental stress changes.
The third chapter dissects the seasonal GPS signal in the western U.S. with special attention to the arid Basin and Range. The study describes spatial characteristics of various sources of seasonal elastic deformation and quantifies the relative contributions of the two largest environmental loading sources. Results show that after correcting for global and regional elastic loading signals, the residual deformation adequately captures local hydrological variations. The last two chapters demonstrate coupling between the solid Earth and the hydrosphere and atmosphere, showing that hydrological forcing is significant in the western U.S., even in arid regions, and that weather and climate can affect processes 30 km underground