18525 research outputs found
Sort by
Manipulating the Brain During Problem-Solving: Direct-Current Stimulation of the Prefrontal Cortex Prior to a Classic Problem-Solving Task
No full textFinding ways to enhance cognitive function in learning and problem-solving has been a focal point of psychological research due to its significance in human life and advancement. I utilized targeted Transcranial Direct Current Stimulation (tDCS) to explore its effects on complex problem-solving abilities. Twenty volunteer students from Vanderbilt University completed the Tower of London (ToL) task under both anodal and sham conditions. Task performance was assessed for time to task completion and move count, comparing outcomes between conditions. Based on the critical role of the prefrontal cortex in working memory manipulations, I hypothesize that anodal tDCS would enhance accuracy and reduce completion time relative to the sham condition. However, my results did not support this hypothesis, with no significant differences observed between conditions. These findings contribute to the ongoing debate on tDCS efficacy and highlight that cognitive enhancement through prefrontal stimulation may depend on factors such as individual neurophysiology, task complexity, and stimulation parameters. Future research should explore alternative brain targets and diverse problem-solving tasks to better understand the potential of tDCS as a cognitive enhancement tool.
Keywords: transcranial direct current stimulation, problem solvingThesis completed in partial fulfillment of the requirements of the Honors Program in Psychological Science
Scale-up Unlearnable Examples Learning with High-Performance Computing
No full textThe rapid evolution of artificial intelligence (AI) systems, exemplified by architectures like ChatGPT, has raised significant concerns about data privacy in healthcare applications. Modern AI models frequently retain user interactions, creating potential vulnerabilities where sensitive medical imaging data - such as those processed by AI-driven diagnostic tools in radiology - could be inadvertently stored and repurposed for model training without explicit consent. This paradigm poses critical challenges to patient confidentiality, institutional intellectual property rights, and regulatory compliance in clinical environments.
To combat unauthorized data exploitation in machine learning, Unlearnable Examples (UEs) have emerged as a promising defense mechanism by systematically degrading model training through optimized perturbations. Among these approaches, Unlearnable Clustering (UC) has demonstrated particular promise through cluster-wise perturbation strategies that enhance protection efficacy with larger batch sizes. However, previous implementations have been constrained by computational limitations, typically operating on single workstations that restrict batch size scalability and dataset diversity. In this study, we present a groundbreaking implementation of UC leveraging High-Performance Computing (HPC) resources through Distributed Data Parallel (DDP) training on the Oak Ridge National Laboratory's Summit supercomputer. Our scaled implementation enables unprecedented batch sizes and comprehensive experiments across multiple medical and natural image datasets, including Oxford-IIIT Pets, MedMNIST, Flowers102, etc.
Through systematic evaluation, we demonstrate that UE effectiveness exhibits complex relationships with batch size configuration. Experiments conducted on datasets such as Pets, MedMNist, Flowers, and Flowers102 reveal that both overly large and overly small batch sizes can lead to performance instability. Furthermore, the optimal batch size is dataset-specific, highlighting the need for tailored strategies to maximize data protection. These findings establish practical guidelines for deploying UEs at scale and provide empirical evidence that HPC-enabled perturbation strategies can significantly enhance healthcare data security in AI applications. The complete implementation is openly available at https://github.com/hrlblab/UE_HPC to support reproducibility and clinical adoption
Exploring the Effects of Editing Techniques via Event-related Potential
No full textIn this study, we explored the neural correlates of the movie-viewing experience. Previous studies have shown that shot changes in a video consistently elicited event-related potential (ERP) components associated with semantic disruptions such as the N400 and P600 component, and we attempted to replicate the findings under a naturalistic viewing experience by showing the participants a feature film. Additionally, we also explored how the ERP components interacted with the various continuous editing techniques in the movie. Consistent with prior research, we found that shot changes in a movie reliably elicited distinct ERPs, characterized by a negative frontal deflection (N400) and a positive posterior deflection (P600). The amplitude of these components, however, did not differ significantly across different editing techniques, or during the start and the end of the movie. Notably, we found a significant right-lateralization of the posterior component, and this effect was weaker in ERPs elicited by match on action cuts, and right-to-left shot/reverse shots. We tested if the posterior lateralization was elicited by detection in bilateral spatial changes and received nonsignificant effects. Finally, we discussed the implications of the results, and potential connection between the posterior lateralization and spatial perception during shot changes
Lingua Franca No More: The Political Erosion of French In Louisiana
No full textHistory Department Honors ThesisCollege of Arts and ScienceDepartment of Histor
Network Analysis and Visualization of Disease Multimorbidity Using Electronic Health Records and Genetic Biobank Data
No full textDisease multimorbidity, the co-occurrence of multiple diseases within an individual, presents complex challenges for both public health and precision medicine. Advancing our understanding of multimorbidity can illuminate disease mechanisms, reveal patient heterogeneity, and enable biomarker discovery and treatment repurposing. Large-scale Electronic Health Records (EHR) and EHR-linked genetic biobanks offer unique opportunities to quantify phenome-wide multimorbidity, uncover shared genetic mechanisms among co-occurring conditions, and define multimorbidity-based disease clusters. However, major analytical and methodological challenges remain. To address these, we present three key contributions. First, we introduce a phenome-wide multimorbidity network that quantifies nonrandom disease-disease co-occurrences while accounting for potential confounding factors. Second, we develop a genetic discovery platform that integrates polygenic scores for predicted transcriptomic, proteomic, and metabolomic traits with phenome-wide association studies (PheWAS) to uncover shared biological mechanisms among multimorbid conditions. To support exploration, we also develop an interactive network visualization tool featuring dynamic cluster analysis of biological pathways linked to diseases with similar multimorbidity patterns, enabling intuitive exploration of complex disease relationships and their shared biological mechanism. Third, we propose a model-based clustering framework using a bipartite stochastic block model (biSBM) with a stability-driven post-processing step to identify robust disease clusters and patient subgroups from individual-level EHR data. This framework demonstrates superior performance in simulations and replicates coherent, interpretable multimorbidity structures across independent datasets, including UK Biobank and Vanderbilt BioVU. A case study of JAK2V617F somatic mutation carriers reveals genetic heterogeneity across patient subgroups with distinct multimorbidity patterns, illustrating the potential of our data-driven approach to uncover mechanistic insights into patient heterogeneity through EHR-derived multimorbidity networks
Detection of Single Nucleotide Polymorphisms in Resource-Constrained Settings
No full textNearly half of known genetically linked human disorders are caused by single base variations, often referred to as single nucleotide polymorphisms (SNPs). Therefore, SNPs serve as a molecular biomarker which, when detected, can provide invaluable insights into disease susceptibility, progression, and treatment responses. Beyond the human genome, SNPs in pathogens, such as HIV-1, SARS-CoV-2, and Mycobacterium Tuberculosis, indicate important traits like drug resistance, transmissibility, and vaccine efficacy. Advancements in biotechnology and science have created powerful genetic tools for the identification of SNPs and other genetic variations. Yet, gold-standard technologies remain expensive and complex, making them unfit for application in resource-constrained settings.
This dissertation focuses on innovative approaches to SNP detection that prioritize accuracy, affordability, accessibility, and adaptability. The oligonucleotide ligation assay (OLA), a high-fidelity, ligase-facilitated reaction, was employed for highly specific detection of known SNPs. The OLA was coupled with polymerase chain reaction (PCR) to attain robust and sensitive detection of important SNPs for pathogenic traits and human diseases. During the COVID-19 pandemic, my work focused on SARS-CoV-2 variant-typing to achieve sequencer-free identification of locally circulating variants of concern in clinical human-derived samples. While the variant-typing efforts were performed in a quick, high-throughput, and adaptable manner, some technical burden of assay implementation remained. Magnetic bead processing was integrated into the OLA-PCR coupled workflow to address this burden within the context of HIV-1 drug resistance detection, significantly improving assay sensitivity and provided the foundation for future automation. To further enhance accessibility, OLA and PCR assay design challenges were overcome through development of a software tool for automated nucleic acid reagent design. The tool streamlined assay customization for a variety of genetic targets, including M. Tuberculosis drug resistance mutations and SNPs indicative of disease risk.
By innovating upon and around existing molecular technologies to overcome technical, logistical, and financial barriers, the work expands the reach of the OLA for highly specific SNP testing – improving the practicality of the assay for resource-constrained environments and situations. With this dissertation, efforts were focused toward not only advancing the field of genetic diagnostics but also toward providing a foundation for more equitable healthcare solutions worldwide
H.R. Knickerbocker and the Structures of Interwar Journalism
No full textHistory Department Honors ThesisCollege of Arts and ScienceDepartment of Histor
Expanding EPSO Access: A Co-Design Approach to Addressing Disparities in Early Postsecondary Opportunities
No full textLeadership and Learning in Organizations capstone projectThis capstone, developed in partnership with Nashville PEER, addresses persistent inequities in Early Postsecondary Opportunity (EPSO) participation for Black, Hispanic, low-income, English learner, and special education students. Using a convergent mixed methods design, the study combined enrollment trends, survey data, interviews, and artifact review to examine how school teams used co-design to expand access. Findings show that co-design fostered stronger advising systems, more intentional communication, and equity-centered scheduling practices, particularly when supported by real-time data and shared leadership. While schools made meaningful progress, capacity constraints and systemic barriers highlight the need for continued investment to achieve lasting, equitable change
"It's Much:" Teacher Collective Sensemaking and Affective Experience with Mandated Literacy Curriculum
No full textThis dissertation is an ethnographic study of teachers’ sensemaking and experience with
literacy curriculum and reform policy while planning. It was implemented with a rhizomatic
approach which frames planning meetings as potential events and considers the plans,
understandings, feelings and possibilities that emerge and endure through them. Analysis reflects the literacy planning of a second-grade team over the course of a school year. While a variety of data was generated and collected , the analyses reported in this dissertation all connect to formal and informal planning meetings, which were the only video-recorded interactions of the study. Analysis was conducted with a variety of discourse-analytic, post-structrual, and participatory approaches around the teaching team's sensemaking and experience with planning and curricula, and the identities, concepts, and capacities that endured beyond them. Findings traced discursive
strategies employed during planning, how curricula and present and past policies constrained sensemaking, and narrowed but contested identities and capacities that emerged
Intact Audiovisual Spatial Integration in Autism: Lessons from Brain and Behavior
No full textIn addition to the core features of social communication impairments (SCI) and patterns of restricted and repetitive behaviors and interests (RRBIs), autism is characterized by changes in sensory processing, including in the integration of information across the different sensory modalities. While there is significant evidence that audiovisual (AV) integration is altered in the temporal domain in autism—and that these differences are associated with presentation of SCI and RRBIs—very little is known about how autistic individuals integrate AV information in the domain of space. We conducted a free-response AV spatial localization task in autistic and non-autistic children aged 7-17 years with EEG collection to understand AV localization abilities, as well as how these abilities relate to SCI, RRBIs, and other clinical features of autism. While we found no group differences in AV spatial performance, an exploratory analysis revealed three clusters of autism features that were significantly predictive of performance across groups: a motor/somatosensory cluster, a sensory responsivity cluster, and a social communication cluster. Similarly, a model comparing participant performance to optimal integration as measured by maximum likelihood estimation (MLE) revealed no group differences, but rather that children more broadly may be suboptimal in their integration of spatial stimuli. Interestingly, EEG power analyses did reveal differences in how autistic and non-autistic children may be processing AV spatial information, including different patterns of posterior and anterior activity in AV conditions. From our primary studies, we suggest that while autistic children with low support needs may have similar ability to integrate AV signals in space, they may be using alternative neural mechanisms to achieve this. Other chapters discuss autism in the context of predictive coding, as well as the lack of sampling diversity in multisensory research to date. We provide suggestions for more inclusive sampling and experimental approaches and possible directions for future research