22689 research outputs found
Sort by
OUT-OF-DISTRIBUTION LEARNING
Get PDFTraditional machine learning approaches assume data points are independent and identically distributed (iid). However, in practice, a learning agent may face data points which drift significantly from the training distribution. In this thesis, we address the problem of out-of-distribution (OOD) learning from two fronts. First, we explore lifelong learning where in-distribution (ID) as well as out-of-distribution data are used to improve performance not only on the present task, but also on past and future tasks. While typical transfer learning algorithms can improve performance on future tasks, their performance on prior tasks degrades upon learning new tasks (called forgetting). Many recent approaches for lifelong learning have attempted to \textit{maintain} performance on old tasks given new tasks. But striving to avoid forgetting sets the goal unnecessarily low. The goal of lifelong learning should be to use data to improve performance on both future tasks (forward transfer) and past tasks (backward transfer). we show that a simple approach---representation ensembling---demonstrates both forward and backward transfer in a variety of simulated and benchmark data scenarios, including tabular, vision (CIFAR-100, 5-dataset, Split Mini-Imagenet, and Food1k), and speech (spoken digit), in contrast to various reference algorithms, which typically failed to transfer either forward or backward, or both. Moreover, the proposed approach can flexibly operate with or without a computational budget. In the second part, confidence calibration for deep discriminative models throughout the feature space is addressed. Deep models, including both random forests and deep-nets, learn internal representations which are unions of polytopes with affine activation functions and they both can be conceptualized as partitioning rules of the feature space. By replacing the affine function in each polytope populated by the training data with a Gaussian kernel, we can enhance the predictive uncertainty estimation of the model. Experiments on both tabular and vision benchmarks show that the proposed approaches obtain well-calibrated posteriors while mostly preserving or improving the classification accuracy of the original algorithm for ID region, and extrapolate beyond the training data to handle OOD inputs appropriately
ALS/FTLD-Linked Mutant FUS Oligomerization in Cells
Get PDFFused in sarcoma (FUS) is a nuclear RNA-binding protein (RBP) implicated in amyotrophic lateral sclerosis (ALS) and frontotemporal lobar dementia (FTLD). FUS plays essential roles in maintaining genome stability, transcription, splicing, RNA processing and localization, and regulating cellular stress responses. FUS can undergo liquid-liquid phase separation (LLPS), establishing biomolecule concentration to perform proper cellular functions. ALS/FTLD-linked FUS mutants can be detrimental to the cell by causing mislocalization and improper LLPS of FUS. FUS cytosolic mislocalization results in a loss of or defective nuclear FUS function and the formation of pathological aggregates in cytoplasm. Although previous studies have reported on subcellular mislocalization, defective nuclear function, and improper LLPS activity of FUS causing toxic aggregates, we lack the quantitative analysis of the oligomeric status of cellular FUS and if it could be linked to pathogenic features of FUS aggregation. Therefore, we sought to investigate an array of ALS/FTLD-linked FUS mutants in their subcellular localization, granule pattern, oligomerization, and potential for aggregate dissolution under normal and varying cell stress conditions. This work also implements a newly established protocol for the single molecule pulldown of fluorescently tagged oligomers from cell lysates, which can be adapted for related RBPs
RETHINKING SOFTWARE SYSTEM RELIABILITY WITH PERSISTENT MEMORY
Get PDFAs modern systems grow larger and more complex, they face exponentially increasing volumes of state and data, while requiring high performance. Persistent Memory (PM) emerges as an appealing new storage option in this landscape. PM is a newly emerging storage technology that has fast reads and writes comparable to DRAM, while being able to persistently store data. As a result, many developers have opted to adapt PM
in their applications. However, in addition to software development, PM yields new implications for software reliability in several aspects.
Many existing works in the PM reliability landscape have extensively researched the crash consistency problem, a PM unique issue where partial/reordered PM writes lead to inconsistent state after crashes. However, there are other aspects of PM software reliability that remain underexplored despite their significance and avenues for interesting research directions.
This dissertation aims to analyze the new implications PM yields for software reliability research outside of crash consistency. It explores novel failure classes brought by PM
integration and the tradeoff between performance and recovery with respect to porting design decisions. However, PM not only presents new issues, but also opportunities,
such as leveraging PM for debugging purposes. This dissertation offers three major contributions to the PM reliability space. 1.) First, it delves into complex design tradeoffs involved in porting legacy KV stores to incorporate PM. 2.) The second
part addresses a new class of PM faults we define as “soft-to-hard faults”. It then introduces Arthas, a tool designed to effectively recover PM systems from soft-to-hard faults using fine-grained checkpointing and rollback. 3.) The final part explores
leveraging PM to assist with post-crash debugging via Steno, a new tool for debugging production failures that uses PM to log intermediate program state. In conclusion, this dissertation seeks to provide extensive studies, advanced techniques, and intricate
systems to reinforce and enhance software reliability with PM
Perceptions Versus Realities of Parental Engagement
Get PDFResearch is saturated with studies on parental engagement, its definition, effectiveness, and methods to approach parental engagement in schools. This dossier explores parental engagement, specifically in a secondary school in a socioeconomically vulnerable community. It explores parental engagement frameworks, the barriers to parental engagement, specifically in socioeconomically vulnerable communities, and stakeholders’ perceptions of it. Using one mixed method and one qualitative study, this dossier found that teachers in a socioeconomically vulnerable school in the southeastern region of the United States believe in the necessity and importance of parental engagement; however, they acknowledge that the current levels are insufficient. This dossier also explored whether misalignments of stakeholders’ perceptions of parental engagement contribute to insufficient levels of parental engagement at a school, an area not widely explored. This research provides implications for staff training and reexamining a school’s parental engagement plan as outlined by The National Network of Partnership Schools
AN EXAMINATION OF TEACHER AND STUDENT TEST-TAKING STRATEGIES: A MIXED METHODS STUDY
Get PDFAbstract
Many students in middle school English language arts courses exhibit disparity in academic achievement as measured by standardized tests. This study aimed to explore teachers’ perceptions of students’ self-efficacy as test-takers, teachers' self-efficacy beliefs about teaching middle school, and teachers' familiarity with and inclination to teach test-taking strategies. Five middle school English language arts teachers, a middle school/high school literacy coach, and a special education teacher participated in a survey and follow-up semi-structured interviews in this concurrent mixed-methods design study. The study context was a combined elementary/middle school in a rural area in the mid-Atlantic region. Findings suggest that teachers deeply care about their students’ academic sense of self-efficacy, engage in various types of cultural capital transmission in test-taking strategies instruction, and that teachers need support in helping develop instruction on test-taking strategies. The study also focused on understanding middle school students’ perceptions of test-taking strategy instruction, knowledge of and inclination to use test-taking strategies, and students’ self-efficacy beliefs related to taking tests. Fifty-four seventh grade students participated in a student survey. The findings portray a complex interplay of how aspects of students’ knowledge and inclination to use test-testing strategies, metacognition, and perceptions of test-taking strategies instruction influence students’ test-taking strategy use, self-efficacy, and reading achievement
DEVELOPMENT OF AN NFAT-MEDIATED INDUCIBLE RETROVIRAL VECTOR SYSTEM IN CD19-TARGETED Vδ1 γδ T CELLS FOR ENHANCED CANCER IMMUNOTHERAPY
No full textChimeric antigen receptor (CAR) T cell therapy has brought about a revolutionary transformation in the treatment of B cell malignancies. Nonetheless, there is an ever-pressing demand for the development of CAR T cell therapies that are not only more potent but also flexible in their applications. An innovative strategy lies in the combination of CAR signaling with the inducible expression of cytokines, a concept known as "T cells redirected for universal cytokine-mediated killing" (TRUCK). Previous studies have demonstrated the potential of TRUCK through using an NFAT-mediated inducible vector system in both T cells and NK cells to enhance the effectiveness of both CAR T cells and CAR NK cells therapies. Vδ1 gamma delta (γδ) T cells represent a unique subset with remarkable antitumor properties, rendering them attractive candidates for CAR T cell therapy. However, not much is known regarding the application of TRUCK principles to Vδ1 γδ T cells. Here, we tested the hypothesis that TRUCK approach can be applied to Vδ1 γδ T cells by integrating the modular retroviral all-in-one inducible vector system delivering a CD19-specific CAR and an NFAT-
inducible transgene. Upon activation, CD19-CAR transduced Vδ1 γδ T cells will exhibit regulated increase in NFAT-mediated transgene expression. By elucidating the mechanisms of gene expression control using this all-in-one retroviral inducible vector system, our research aims to contribute to the advancement of CAR T cell therapies that are not only more efficacious but also versatile in their applications. This study serves as a crucial step in harnessing the potential of Vδ1 γδ T cells and expanding the repertoire of CAR T cell therapies
Search For Electrophysiological Biomarkers Of Arousal In Anesthesia And After Cardiac Arrest
Get PDFElectrophysiological recordings, notably Electroencephalography (EEG) and Somatosensory Evoked Potentials (SSEP) play pivotal roles in clinical settings, offering insights into the functioning of the nervous system and aiding in diagnosing and monitoring various neurological conditions. EEG captures the brain's electrical activity, enabling clinicians to assess brain function, diagnose epilepsy, monitor anesthesia, and evaluate sleep disorders. SSEP measures the electrical responses of the nervous system to sensory stimuli, providing crucial information about the integrity of sensory pathways, aiding in the diagnosis of spinal cord injuries and peripheral nerve disorders, and assessing intraoperative neurophysiological integrity during surgeries.
Nevertheless, there is much space for improvement regarding the applications of these techniques to clinical practices, for example, to more precisely monitor anesthetic depth and to guide the use of hypothermia therapy and the prognosis of neurological recovery for cardiac arrest patients. Cardiac arrest often leads to significant neurological damage, and predicting the neurological outcome accurately is challenging yet essential for guiding clinical management and decision-making. By identifying specific electrophysiological signatures associated with favorable or unfavorable outcomes, clinicians can make timely and informed decisions regarding therapeutic interventions, prognostication, and potentially tailoring individualized treatment strategies.
This thesis develops a novel recording modality named the Extended Evoked Response Potential (EERP). Based on the well-validated and widely used EEG and SSEP techniques, EERP attempts to combine both and offer a more comprehensive perspective of the status and dynamics of neural activities. Chapter 1 provides a comprehensive literature review of the neurobiological foundations of arousal and the applications of electrophysiological tools for clinical purposes. Chapter 2 describes this thesis study's theoretical principles and experimental settings. Chapter 3 demonstrates the use of EERP in both anesthesia and cardiac arrest experiments, and the results are shown. Chapter 4 evaluates the ear as a potential location for the non-invasive recording of SSEP. Chapter 5 concludes the findings of the thesis work and discusses future investigations
Molecular Uptake and Transport in iPSC-derived Brain Endothelium
Get PDFBrain diseases represent a significant and widespread challenge, affecting millions of people around the world. Understanding the mechanisms of molecular uptake in the blood-brain barrier (BBB) and developing effective therapeutic strategies to cross BBB are critical to solve this universal healthcare burden. However, the uptake of potential drug carrier proteins in human brain endothelium remains poorly understood due to the lack of experimental models to recapitulate human BBB characteristics. Using human induced pluripotent stem cell derived brain microvascular endothelial cells model, the internalization kinetics of the molecular uptake of serum albumin, low-density lipoprotein (LDL), and transferrin was investigated. The mechanisms of endocytosis were studied by the inhibition of caveolae-mediated and clathrin-mediated endocytosis pathways. The time- and concentration-dependent molecular uptake reveals the kinetics of serum albumin, LDL, and transferrin uptake. The chemical inhibition of endocytic pathway indicates that serum albumin could be taken up via caveolae-mediated endocytosis pathway, and transferrin could possibly be taken up via both caveolae-mediated and clathrin-mediated endocytosis pathway. Taken together, this research on molecular internalization in human brain endothelium provides an effective model and method for investigating endocytosis across the human blood-brain barrier, paving the way for revealing the structure of the blood-brain barrier and enhancing therapeutic drug delivery for brain disorders
Garden, Gourd-Vine, and Graft: Conjured Chronologies in the Root-Works of Martin Delany and Zora Neale Hurston
Get PDFThis dissertation analyzes uses of conjure across the careers and imaginative repertoires of Martin Delany and Zora Neale Hurston. In the process, I identify their shared interests in using African and Euro-Christian religious traditions to reconfigure historical narratives and author redemptive visions of historical salvation that depend on a confounding rhetoric of hiddenness and revelation. Each of the three chapters identifies one or more crucial botanical metaphors that configure each author’s attempted refiguration of providential history. Applying phenomenology-of-reading approaches including Paul Ricoeur’s interlinked theories of metaphor and imagination, alongside Ricoeur and Theophus Smith’s post-secular methodologies for examining Christian myth and African American conjure traditions respectively, this dissertation brings these analytical models to bear on African American texts charged with activist aspirations and implications. These texts in turn become legible as open invitations to consider conjure traditions, both as a historical phenomenon and as a potential model for literary practice. The readings made possible through this conjunction of methods and texts ultimately oppose positivist or reductionist readings of conjure as a form of compensatory—rather than effective and enactive—imagination
DEVELOPING A NOVEL SARS-COV-2 DNA VACCINE TO ELICIT AN ADAPTIVE IMMUNE RESPONSE USING A PADRE MOTIF
Get PDFThe height of the COVID-19 pandemic saw the rapid evolution and advancement of a revolutionary vaccine platform that was instrumental in decreasing the morbidity and mortality of the SARS-CoV-2 virus. The two primary vaccines that utilized this platform, BNT162b2 (Pfizer-BioNTech) and mRNA-1273 (Moderna), both saw high levels of efficacy and were responsible for the substantial decrease in severe SARS-CoV-2 infection.
However, over time due to ongoing mutations to the spike (S) glycoprotein that they each targeted, neutralizing antibody levels have begun to wane as more mutations have begun to surface and increase their viral load. Logistically, the mRNA platform has also proven to be a hurdle for underprivileged areas where access to therapeutics may be an issue due to their cost of development, production, shipping, and storage.
To combat these issues, our lab developed a SARS-CoV-2 DNA vaccine that targeted a highly conserved region of the virus known as open reading-frame 3a or ORF3a (“Cov”). Although still difficult to stimulate a strong systemic response, the preliminary results of this study served as a proof of concept to illustrate the viability of the ORF3a protein as a target for vaccination. Here, we present the development of a novel SARS-CoV-2 DNA vaccine aimed at enhancing both the overall humoral and systemic immune response to the ORF3a antigen by conjugating a CD4+ T helper cell epitope (“PADRE”) to our established MIP- 3αORF3a DNA vaccine.
Although the humoral response was observed to be about the same as our ii
previous vaccine formulation, the overall CD4+ T cell response was about 5% greater. In addition, the presence of cells that were triple positive for CD4, IFN-γ, and TNF-α were much greater than in our previous vaccine. This study highlights the possibilities that PADRE can offer in future studies when combined with the chemokine MIP-3α and the ORF3a peptide, and can hopefully provide a lower barrier to access for communities that are still suffering from severe SARS-CoV-2 infection