12466 research outputs found
Sort by
Trends in Extreme Precipitation: Identifying Historical and Projected Patterns Using Multi-Source Precipitation Datasets
No full textThis work is embargoed by the author and will not be publicly available until May 2028.Understanding past precipitation trends and the likelihood of such trends persisting or changing in the future is crucial to optimize water demands, designing infrastructure, and preparing a climate-resilient society. This dissertation investigates historical and projected changes (magnitude, frequency, and intensity) in annual, seasonal, and extreme precipitation across different regions in the world by leveraging datasets from a range of sources: Earth observations (i.e., satellite remote sensing), models of earth system science (e.g., atmospheric reanalysis and community earth system models, CESM), as well as ground-based measurements. Precipitation patterns are evaluated in terms of temporal trends (and their statistical significance) and spatial patterns from the regional (Southern Mid-Atlantic and High Mountain Asia) to the continental scale (i.e., Contiguous United States). This work adopts hydroclimatic extreme indices from ETCCDI (Expert Team on Climate Change Detection and Indices) to characterize precipitation trends and the relative contribution of extreme events to the total precipitation. At a regional scale, using 40 years (1980-2018) of high-resolution reanalysis data from the North American Land Data Assimilation V2 (NLDAS-2, 12 km/ hourly), a significant (0.1 significance level) increase in annual precipitation (+3~+5 mm/year) is identified over Northern Virginia accompanied with an increase in summer precipitation. An annual increase (at 0.05 significance level) in extreme events (95th and 99th percentiles) is also identified in the southern Mid-Atlantic region. An investigation into the proportion of annual precipitation occurring on wet and extremely wet days (95th and 99th) also indicates a significant increase. Next, this dissertation evaluated the projected changes until the end of the 21st century in the probability distribution of hydroclimate extreme indices across the National Climate Assessment (NCA) regions in the contiguous US. This was performed using a large ensemble simulation (70 members) in a new scenario-matrix-architecture (Shared Socioeconomic Pathway 3, SSP3-7.0) of CESM v2 (100km/daily, 2015-2100) from the Coupled Model Intercomparison Project Phase 6 (CMIP6). A projected increase in the northeastern regions in spring and winter and consistent drying in the Midwest summer precipitation were identified. Next, spatiotemporal patterns of both precipitation averages and extremes were identified over a region characterized by complex terrain, i.e., HMA, during 1990-2018. However, finding a reliable dataset in this region can be challenged by several factors, including the lack of ground observations, the diverse climate zones, and the sharp orographic gradients. Therefore, evaluating the quality and reliability of different precipitation products before analyzing their trends and patterns is fundamental. A comprehensive assessment of high-resolution satellite-based, model reanalysis precipitation estimates and their blended product (ensemble) was conducted using ground observations from a suite of rain gauge networks at different elevation ranges. The last chapter of this dissertation transitions from pattern-based analyses to an event-based analysis that focuses on the heavy downpours that triggered devastating floods in Pakistan in the summer of 2022. Results confirm the singularity of this event with abnormal daily rain rates compared to climatology in western Balochistan and large anomalies across Pakistan’s southern provinces after August 16th. Outputs from this dissertation provide insights into the changing distribution of precipitation extremes, which are likely to trigger hydroclimatic hazards. In conclusion, this dissertation highlights the importance of assessing changes in extreme precipitation patterns across scales and from a variety of data sources to improve our understanding of the changing water cycle and enhance regional resilience to extreme climatic events.2028-05-1
Special Education Teachers' Perceptions of Students with Intellectual Disabilities' Participation in the Individualized Education Program Process
Get PDFThe purpose of this qualitative basic interpretive study was to explore the perspective of high school special education teachers regarding the factors that support or hinder student participation in the IEP process. Data sources included semi-structured interviews, memos, documents, and electronic correspondence and were coded using Braun and Clarke’s (2006) six phase thematic analysis. Five themes emerged from the data: (a) teacher beliefs and student characteristics, (b) teacher experiences with families, (c) school culture, (d) methods and resources, and (e) training. Implications for practice and policy are presented and recommendations for future research are discussed
It's About Time! A Time-Domain Multi-Wavelength Study of Nearby Active Galactic Nuclei
Get PDFSupermassive black holes (SMBHs) are known to reside at the center of nearly everymajor galaxy, and there is a well-established correlation between the black hole mass and stellar velocity dispersion, indicating a co-evolution between the SMBH and the host galaxy. Gravity alone cannot explain this correlation since the gravitational sphere of influence of the SMBH extends out to only a few parsec, meaning there must be some other mechanism at play. One such mechanism is when gas and dust is accreted onto the SMBH, generating enormous amounts of energy in the form of outflowing winds or collimated jets called active galactic nuclei (AGN). AGNs can produce ’feedback’ on the host galaxy by heating the gas of the surrounding interstellar medium, via radiative and mechanical energy, which in turn can cause quenching of star formation. In the case of mechanical feedback, collimated jets can propagate out to kilo-parsec and even mega-parsec scales. There is still ongoing debate as to the exact driving mechanisms in AGNs that contribute to the launching of jets, as jets are generally inferred to originate very close to the SMBH, in a possibly synchrotron self-absorbed corona. This is a cloud of hot free electrons confined to an area that resides somewhere above the accretion disk where UV photons emitted by the accretion disk are inverse Compton scattered, bumping their energies to the X-ray regime. There have been many relationships, or correlations, between radio and X-ray emission found in the litera- ture, applying to both stellar-sized black holes and SMBHs. One relationship in particular is the so-called ’fundamental plane’, which purports to unify X-ray and radio emission from accreting black holes in general by introducing black hole mass as a third parameter; how- ever, this relationship has recently been shown to break down on the smallest physical scales where it was expected to be strongest. Correlations between the X-ray and radio emission in accreting black holes suggest that the X-ray corona is also the origin of the radio emis- sion, and may also be associated with the launching of jets, but the exact mechanism is still a matter of ongoing research. The X-ray emission can also be used to estimate the bolometric luminosity of the AGN, which is indicative of the accretion rate of the SMBH. With a high enough accretion rate, the AGN becomes radiatively efficient to drive winds and referred to as radiative feedback. Several outstanding questions exist, such as whether or not both the primary radio and X-ray emission of AGNs originate in the corona. If they do have the same origin, are they coupled? Which AGN feedback mechanism is dom- inant in the local universe? A powerful method to address these questions is simultaneous multi-wavelength observations, which include X-ray observations and very long baseline in- terferometery (VLBI) that can resolve radio emission in local AGNs on parsec to sub-parsec scales, providing exquisite views into the inner accretion region. In this PhD dissertation, I make several contributions to our understanding of the different physical mechanisms of ra- dio and X-ray emission in AGNs. This includes finding evidence of anticorrelation between the core radio and X-ray emission and finding the mechanical feedback produced in nearby AGNs may not be as significant as other modes of feedback such as that from radiative feedback. I also include future research in this field that aims to address some remaining questions
Barely Visible Inner Constructions
No full textThis thesis is a culmination of the work I have done in my three years of this program. The first section is a complete chapbook, titled Barely Visible Inner Constructions. This book is an exploration of desire and the body, specifically through the lens of luxury industries and my relationship with my mother, who introduced me to the world of fashion. Many of the poems utilize the sapphic stanza in order to frame the desire present as something more closely tied to queerness and femininity. Others use a dropped line form that I began as a take off on the sapphic stanza, using what would have been the shorter, 5 syllable line in the original as a moment for a clearer interiority and image. Section two is a collection of poems more pointedly about desire. The final section collects many of my poems that are in conversation with other art, especially film.This thesis has been embargoed for 10 years. It will not be available until May 2032 at the earliest
Jointly Improving Performance of Human Annotators and Models in Human-in-the-loop Machine Learning Systems
Get PDFThere is a rise in leveraging streaming data in data science applications across a variety of domains, such as disaster management and humanitarian work. Some of these applications focus on mining and analyzing human behavior from public-generated data streams online. For example, filtering relevant help-request messages for enhanced disaster management using online social media platforms, extracting malicious messages from online social media platforms on gender-based violence for informed policy-making, and so on. In recent years, due to time-critical decision-making requirements, the concerned organizations have started using Artificial Intelligence (AI) based systems to automatically analyze and mine human behaviors from online social media platforms in real-time at scale. However, given the lexical ambiguity, sparsity of behaviors, and concept drift of data streams, the current AI systems fail in the tasks requiring a higher level of cognition and reasoning necessary for analyzing such behaviors from natural language text of online social data that humans can efficiently perform. Thus, human knowledge to assist purely data-driven methods of AI can help design effective AI systems for the complex tasks of mining human behavior from online social data. In this dissertation, I focus on optimizing knowledge acquisition in Human-In-The-Loop Machine Learning (HITL-ML) systems for processing online social data streams. My goal is to enhance the machine learning (ML) processes by emphasizing the importance of high-quality human annotations and effective labeled data collection. By optimizing knowledge acquisition, minimizing human annotation errors, and improving the quality of labeled data, my research contributes to the development of HITL-ML-based streaming analytics systems. Additionally, I highlight the significance of considering the properties of human memory and cognitive processes when designing hybrid HITL-ML systems. This understanding provides valuable insights into creating accurate classifiers and reducing errors in human annotation, thereby enhancing the overall performance of ML-driven streaming analytics systems. Specifically, my research proposes an efficient system design for HITL-ML system for mining relevant documents from streaming data, which relies upon active learning and reinforcement learning techniques with inspiration from psychological theories. This research hypothesizes that continuous labeling feedback from humans helps the active learning-based AI model gain better performance when provided with carefully curated labeling task sequences. Additionally, taking inspiration from psychological theories of human errors, my research investigates the modeling of human error in combination with machine/model errors, thus, minimizing the chances of non-experts making errors when giving labeling feedback to a model. I validate the proposed system design on classification tasks of filtering online social data streams for relevant messages posted during natural disasters. The findings from this research help design more effective AI systems for mining human behavior from online social data streams, benefiting various applications, including disaster management, social media analytics, and informed data-driven policy-making. It fills the gap in the existing works on AI system design that focus on better optimizations for the ML model and system performance alone
ROLE OF POLY‐GAMMA-GLUTAMIC ACID SYNTHESIS LOCUS IN FRANCISELLA MICROBIAL PHYSIOLOGY
No full textThis work is embargoed by the author and will not be publicly available until August 2028.Francisella tularensis is the causative agent of tularemia. It is a facultative intracellular bacterium and part of its life cycle outside of a eukaryotic host. Poly-gamma glutamic acid (γ- PGA) is produced via the genes in the pgs locus. γ-PGA is one of the major virulence and survival factors of B. anthracis and enables S. epidermidis to escape phagocytosis. It is common for gram-positive bacteria to produce an attached γ-PGA capsule, but it is unknown for gram-negative bacteria. Due to the presence of the outer membrane, the mechanism for attaching this type of capsule to the bacteria in gram-negative bacteria is not possible. However, all strains of Francisella require this locus for virulence. The physiological role of pgs (cap) locus in Francisella is unclear but the deletion of these genes had a significant effect on intracellular growth and resulted in an attenuated mutant strain. Characterizing the potential formation of γ-PGA when Francisella is under stress is important in understanding the resistance to stress and the ability to grow in harsh environments. The goals of this study are to characterize this pgs locus and identify the ability of this operon to produce γ-PGA.2028-08-1
Automated Search, Classification, and Metadata Annotation of Peer-Reviewed Publications for NeuroMorpho.Org
Get PDFMotivation: The biomedical literature is expanding at ever-increasing rates, and it has become extremely challenging for researchers to keep abreast of new data and discoveries even in their own domains of expertise. For example, NeuroMorpho.Org, a sharing platform for digital reconstructions of neural morphology, must evaluate more than 6000 potentially relevant articles per year to identify data of interest. We introduce a crawler based on periodic full-text searches across publisher web portals that automatically finds and assess the likelihood of a publication to be relevant for the project. Furthermore, it extracts key elements of the metadata such as tracing system, species, cell type, and brain region.Results: Without user interactions, the tool retrieves and stores the bibliographic in- formation (full reference, corresponding email contact, and full-text keyword hits) based on pre-set search logic from a wide range of sources including Elsevier, Springer/Nature, PubMed/PubMedCentral, and Google Scholar. Although different publishing sites require different search configurations, the common interface unifies the process from the user per- spective. Once saved, the tool automatically identifies articles describing digitally recon- structed neural morphologies with high accuracy. In addition, the ability to automatically extract key metadata from neural tracing reduces the risk of errors or inconsistencies in the analysis and interpretation of the data. Conclusions: Since deployment, the tool helped NeuroMorpho.Org more than quintu- ple the yearly volume of processed information. Its processing rate of 900 publications per hour is not only amply sufficient to autonomously track new research, but also allowed the successful evaluation of older publications backlogged due to limited human resources. The number of bio-entities found since launching the tool almost doubled while greatly reducing manual labor. The tool is open source, configurable, and simple to use, making it extensible to other biocuration projects
Machine Learning and Reinforcement Learning in Computer Aided Design
Get PDFIn the Clock Tree Synthesis (CTS) phase of physical design, the clock network is generated.Traditionally, design tools prioritize balanced clock trees and minimal clock skew, which can increase power demand. To tackle this issue, our study proposes two solutions: Representation-learning Architecture for Path-based Timing-Analysis (RAPTA) and a Reinforcement Learning approach for reducing peak current through clock skew engineering. Our learning solution, RAPTA, utilizes path-based static-timing-analysis to predict timing-slacks. It is employed during the place-and-route design stage to identify and annotate discrepancies between delays reported by path-based and graph-based timing analysis for desired Process, Voltage, and Temperature (PVT) corners. This enables comprehensive optimization of Power, Performance, and Area (PPA). Notably, RAPTA offers significant advantages: superior accuracy (errors ranging from 3.9ps to 16.05ps in 32nm technology), architecture unaffected by feature-set size changes, and elimination of manual feature engineering by directly utilizing gate and net properties from Electronic Design Automation (EDA) tools. The reinforcement learning agent employed in our study learns to adjust the clock arrival time of each register to maximize the distribution of clock arrival. This approach enables us to explore optimization opportunities in clock tree synthesis beyond the limitations of heuristic algorithms used in modern EDA tools. Our experimental results strongly support this claim, demonstrating a significant 35% reduction in peak current and a substantial decrease in IR drop (from package to transistor) across the selected benchmarks. The second experiment modified the Q-table updating technique, which resulted in another additional 10% improvement compared to the first experiment. In both experiments, the agent traverses the environment and explores different options despite creating timing violations and obtaining a huge negative feedback reward for the actions taken. However, the timing violation fixed later results in the agent obtaining a future reward for modifying the clock arrival time of other registers. The overall process resulted in the wider spread of clock arrival distribution. While the reinforcement learning agent explores potential optimizations, it may create timing violations, resulting in a large negative reward. However, if the timing violation is subsequently resolved by adjusting the clock arrival time of other registers, the agent can receive a bonus reward, leading to a wider spread in clock arrival distribution
Frozen Justice: Detecting and Recovering Firearms Evidence at Snow Scenes
Get PDFThis research project identifies and explores the complexities associated with the detection and recovery of firearms evidence, namely casings, from areas in which snow is prevalent. By reviewing established crime scene processing methodologies, this research sought to determine what, if any, effect snow has on an investigator’s ability to find and collect casings when they are concealed underneath a layer of solid precipitation. Casings of eight different sizes were used to establish a mock scene at Fort Drum, New York, where the average annual snowfall regularly exceeds 100 inches. Two different brands of metal detectors were used to recover the casings, a Garrett CSI Pro and a Bounty Hunter Tracker IV. Bench tests performed with both detectors in controlled indoor conditions showed consistent responses to a distance of approximately eight inches across the eight calibers and gauges examined. During the recovery experiment approximately one month after initial deposition, the less expensive detector performed poorly when pitted against the more expensive model. The experiment resulted in the recovery of 48 of 80 casings, a 60% recovery rate. Neither detector was able to locate .22 Long Rifle casings at a depth of approximately 7”, indicating a relationship between size of the casing, depth of snow, and strength of the detectors
Rapid and Precise Tumor Classification: A Neural Network Approach for Malignant and Benign Tumor Distinction
Get PDFThe early and accurate diagnosis of cancer significantly impacts the effectiveness of treatment strategies. Traditional diagnostic methods, while effective, are often time-consuming and may not always reach optimal accuracy. We aim to demonstrate the application of a neural network model, developed using TensorFlow that is able to distinguish between malignant (cancerous) and benign (non-cancerous) tumors, addressing the critical need for rapid and precise diagnosis in oncological care. Utilizing a large dataset containing categories consisting of the mean, standard error (SE), and worst (or largest) measurements of each major tumor characteristic, we preprocessed the data to separate features from labels and split it into training and testing sets. We developed the model on Google Colab, then tested and trained through thousands of epochs before finalizing on a model that is monitored through metrics including loss (binary cross entropy) and accuracy. The final neural network accurately predicted tumor diagnosis (whether it is malignant or benign) with an accuracy rate of 99.12%, and only having a loss (the measure of error in prediction) of 0.0468, showcasing both high reliability and precision in its predictions