137091 research outputs found
Sort by
How to Write a Book About Cabin Crews
No full textInterview portion of Lost in the Stacks, episode 641. Features interview with Dr. Drew Whitelegg discussing the process of writing his book, Working the Skies.Interview portion of Lost in the Stacks, episode 641. Features interview with Dr. Drew Whitelegg discussing the process of writing his book, Working the Skies
Revealing unknown features of ribonucleotide incorporation in various genomic DNA
No full textRibonucleotides, known as ribonucleoside monophosphates (rNMPs), are considered being the most plentiful non-canonical nucleotides easily found in genomic DNA. The presence of rNMPs was initially found in mitochondrial DNA from mice and HeLa cells in the 1970’s. Many studies have shown that the failure to prevent the incorporation and accumulation of cellular rNMPs in genomic DNA can lead to genome instability and, in mammals, contribute to the development of Aicardi-Goutières Syndrome (AGS). The efforts to unlock the secrets of rNMP incorporation have partially revealed which mechanisms and what genes are involved in. However, conducting deeper research to investigate the function/s of embedded rNMPs found in cellular genomic DNA has often encountered difficulties due to the activity of ribonucleotide excision repair (RER) pathway and the absence of appropriate mapping techniques. To map rNMPs in genomic DNA and broaden our understanding of their incorporation features in larger genomes, we upgraded the ribose-seq protocol, a technique developed in 2015 for capturing rNMPs, by integrating modern technical advances. We successfully extended our understanding of non-randomly incorporated rNMPs captured in various yeast genomes, both in the presence of active ribonucleotide excision repair (RER) and in two AGS-associated mutants modeled in yeast. In addition, advancing our techniques to study rNMPs in mammalian cells, enabled us to explore the specific signatures of rNMPs in both mitochondrial and nuclear DNA across a wide range of human cell types. These findings have revealed fundamental insights into the functional and structural impacts of rNMPs on genome metabolism, as well as potential clues to rNMP repair mechanisms relevant to human disease.Ph.D.Biolog
Distal Trait and Proximal Strategic Predictors of Technological Fluency
No full textModern workplaces are characterized by the necessity of just-in-time, non-routine problem-solving, often aided by technology. Although the Internet is an increasingly core resource for such problem-solving, the competencies which support leveraging technology for problem-solving are not well understood. Effective use of online resources depends upon a complex of individual differences including ability and non-ability traits, the accumulation of relevant knowledge and skills, and the use of adaptive strategic approaches during interactions with technological resources. I refer to this as technological fluency, a trait complex which describes individuals’ propensity (i.e., ability/willingness) to leverage technological resources to solve real-world problems. The present study examined the relative contributions of distal trait and proximal strategic variables to explaining individual differences on an assessment of technological fluency. While item-level floor effects limited the interpretability of aggregate performance data, there were several interesting findings related to individual differences in process variables. Cluster analyses, for example, showed that minimally satisficing participants (who had very low intrapersonal variability in item scores) tended to be more positively oriented toward technology, while the two clusters of participants with greater intrapersonal variability in item scores could be distinguished by patterns of differences in both ability and non-ability variables. Follow-up qualitative analyses identified several key process variables for which ability differences were most striking, including inference quality (including the ability to self-correct faulty inferences), the ability to leverage visual problem cues, and the nature of AI use during problem-solving. Finally, exploratory regression analyses suggest that proximal strategic variables do improve prediction of available process indicators above and beyond distal trait complexes, providing support for a key proposition of the study. In addition to providing methodological recommendations for the assessment of technological fluency (e.g., more direct assessment of metacognitive processes, inclusion of domain knowledge tests), the present study’s findings suggest that the quality of metacognitive processing in technology-supported problem-solving may be correlated with individual differences in ability – this is a critical direction for future research.Ph.D.Psycholog
Breaking Barriers: Unraveling the Roles of LolA and LolB Protein in Escherichia coli’s Antibiotic Defense
No full textIn Escherichia coli, the outer membrane (OM) plays a crucial role in survival and antibiotic resistance. Lipoprotein trafficking helps to maintain the integrity of the OM, and is facilitated by the Lol pathway and its respective proteins LolABCDE. This thesis investigates the functional roles of two key proteins in this pathway, LolA and LolB, and their potential as targets for antibiotic treatments. The LolA project demonstrated that making single or double mutations within a hydrophobic region of LolA did not significantly disrupt lipoprotein trafficking, or cause OM defects that increased sensitivity to antibiotics, whereas triple mutations to the hydrophobic region resulted in a lack of LolA protein abundance. Investigation into LolB utilized a constructed mutant library, with a focus on identifying dominant-negative mutations that disrupted not only lipoprotein trafficking to the OM, but also demonstrating antibiotic sensitivity. While initial results determined that certain lolB mutations might impair bacterial viability, careful plating assays showed no significant differences in antibiotic sensitivity compared to wild-type E. coli. This finding highlights the complex robustness of OM assembly via the Lol pathway, as well as the need for further exploring additional LolB mutants. Overall, the findings contribute to a deeper understanding of the role of the Lol pathway in OM assembly as well as its potential for future drug developments, emphasizing the need for continued research to expand novel strategies to combat antibiotic-resistant pathogens.UndergraduateBiolog
Essays on Retail Investor Behaviors and Technology Platforms
No full textRapid strides in information technology and fintech have significantly transformed the retail investment landscape. This dissertation examines the impact of technology platforms on retail investor behaviors.
In my first essay, I examine the impact of product recommendations on retail investors’ mutual fund investments. Product recommendations have been widely adopted on online investment platforms to help investors make investment decisions more easily. The findings indicate that investors tend to follow the recommendations when they make mutual fund purchases. However, only investors of low socioeconomic status (SES) suffer significantly worse investment returns after purchasing recommended funds. To explain this disparity, I find investors with low socioeconomic status tend to gather less information and expend reduced effort in fund research when buying recommended funds. Furthermore, those investors tend to hold recommended funds for a significantly shorter time and their purchase of recommended funds experiences a significantly larger price reversal than non-recommended funds. In summary, product recommendations disproportionately lead lower socioeconomic status investors to make irrational investment decisions, exacerbating financial harm for those who are most vulnerable in the market.
The second essay examines the impact of stimulus checks on retail investors’ interaction with robo-advisors. As robo-advisors become an increasingly essential tool for investors when making financial decisions, it is critical to examine how economic factors such as stimulus checks impact investors’ interaction with robo-advisors. After receiving stimulus checks, investors significantly increase their lumpsum investments to robo-advisors. This increase is most significant for investors with lower income, younger age, higher risk tolerance, and previously rely on robo-advisors more. This increase in lumpsum investments mostly go to short-term investment accounts rather than long-term retirement accounts. Investors also increase their number of roundups investments and decrease their number of withdrawals from robo-advisors.
The third essay analyzes merger and acquisition rumors on social media platforms for retail investors using data from Seeking Alpha. Social media has become an essential source of information for retail investors when making investment decisions. My findings reveal that rumors about firms with higher attention and more bullish authors on Seeking Alpha are more likely to be accurate, while those with negative sentiment are less credible. Firms involved in true rumors exhibit higher abnormal returns, reflecting the market’s partial ability to assess acquisition probabilities. Retail investors are particularly influenced by Seeking Alpha authors’ disclosures, favoring stocks with more long position disclosures.Ph.D.Managemen
Human Hand Joint and Mesh Reconstruction Based on RGB Images
No full textHand pose and shape estimation, or hand mesh and joints reconstruction, has been popular since the extensive use of deep learning techniques for its utility in game design, AR/VR applications, and human-machines interactions. The task is estimating the real-world or camera space coordinates of a hand from an RGB image, or video, of the hand or a person, with possible assisting inputs such as depth images, heat maps, and silhouette. In this work, we will try to reconstruct hand pose and shape from RGB images as frames from videos to assist evaluating the condition of patients possibly experiencing stroke or seizure. Our model will contain a spiral encoder and decoders to reconstruct the mesh as well as projecting the mesh to get the joints prediction, and another transformer for taking in predictions from the previous model and proceed to predict about the existence of illness, with some Large Language Model(LLM) at the end to adjust for the small errors and smoothing the movements.UndergraduateComputer Scienc
Leveraging sparsity in deep neural networks for training efficiency, interpretability and generalization
No full textSparse neural networks (Sparse NNs) are characterized by having fewer connections between consecutive layers compared to traditional fully connected, or dense NNs. Historically, sparsity has been studied post-training to enhance inference efficiency and as a regularization mechanism to improve generalization. However, additional benefits beyond these areas remain underexplored. In this thesis, we investigate sparse NNs, various sparsity patterns, and their broader benefits, including improved training efficiency, enhanced interpretability, and stronger generalization.
First, we introduce PHEW (Path with Higher Edge-Weights), a novel method for identifying sparse sub-networks within dense NNs at initialization, without using any training data. PHEW is a probabilistic network formation method based on biased random walks, relying solely on the initial weights of the NN. Importantly, PHEW does not make any task-specific assumptions; instead, it exploits structural properties inherent in dense NNs that promote faster convergence and better generalization. By identifying effective sparse sub-networks at initialization, PHEW reduces the computational burden of training dense NNs and consistently outperforms other state-of-the-art methods.
Second, we propose Neural Sculpting, a technique to uncover the underlying hierarchically modular task structure within NNs. Many real-world tasks exhibit hierarchical modularity, where complex target functions can be decomposed into simpler sub-functions arranged in a hierarchy. We pose the following question: given a sufficiently deep NN, how can we uncover the task’s hierarchical structure? Neural Sculpting uses an iterative process of pruning both units and edges during training, followed by network analysis to detect functional modules and infer hierarchical relationships between them. This method enhances the interpretability of NNs by guiding them to reflect the task’s inherent hierarchical and modular structure through pruning, and subsequently revealing that structure through network analysis.
Finally, we leverage structural information about the task’s hierarchical modularity to enhance NN performance by aligning the architecture at initialization with the task’s structure. Specifically, we investigate how modular NNs can outperform dense NNs by systematically varying the degree of structural knowledge incorporated at initialization.
We compare architectures ranging from monolithic dense NNs, which assume no prior knowledge, to hierarchically modular NNs with shared modules, which leverage sparsity, modularity, and module reusability. Incorporating modularity and module reuse significantly enhances learning efficiency and generalization, particularly in data-scarce scenarios, where hierarchically modular NNs excel by promoting functional specialization and reducing redundancy.
These findings reveal that task-specific architectural biases can lead to more efficient, interpretable, and effective learning systems.
In conclusion, this thesis demonstrates that sparse NNs offer not only enhanced training and inference efficiency but also superior interpretability and generalization capabilities. These findings have broad implications for NN design across various domains, particularly in data-scarce scenarios or applications where understanding the underlying task structure is essential. Future work may focus on refining these methodologies and extending their applicability to more complex, real-world tasks and larger-scale architectures.Ph.D.Machine Learnin
Improve the Situation
No full textInterview portion of Lost in the Stacks, episode 624. Features interview with Jordan Moore, the User Experience Librarian at the Georgia Tech Library, discussing how she has adapted to her new position and new location, what a User Experience librarian does, and her brief appearance on the game show "Jeopardy!" in 2018.Interview portion of Lost in the Stacks, episode 624. Features interview with Jordan Moore, the User Experience Librarian at the Georgia Tech Library, discussing how she has adapted to her new position and new location, what a User Experience librarian does, and her brief appearance on the game show "Jeopardy!" in 2018
Machine Learning Enabled Stem Cell Lipidomics
No full textLipids form the life-defining barriers that separate a cell from its environment. They provide the cell with structure, allow it to move, hold the vital proteins that allow the influx of nutrients and outflux of waste, and are important members of metabolic and cell signaling pathways. Not only are there many thousands of different lipids, but lipid metabolism is also constantly changing in response to the state of the cell. In the face of overwhelming biological complexity, omics fields have emerged with the goal of profiling biological systems in a global fashion to create a snapshot of the system in a moment of time. Lipidomics enables the large-scale study of lipids by leveraging current developments in analytical techniques.
Stem cell therapies aim to utilize cellular functions to regenerate tissues, provide immunotherapy, and treat cancer. By taking advantage of the complexity of cellular machinery, these treatments can perform functions that traditional small molecule drugs and biologics are unable to perform. However, the complexity of these medicines requires additional tools for medicinal characterization in order to meet FDA guidelines for outlining mechanism of action (MoA) and defining safety and quality standards. Lipidomics approaches for cell characterization hold unique abilities to capture the complexity and heterogeneity of these therapies to help identify molecular markers of medicinal quality or critical quality attributes (CQAs).
Mass spectrometry’s analytical power is derived from the highly customizable nature of the available instrumentation. Custom mass spectrometry workflows and methodology can create advantages in terms of molecular coverage, specificity, sensitivity, and quantitative or statistical reliability that enable lipidomics for CQA discovery. In particular, developments that increase quantitative reliability and sensitivity can benefit the field of cellular lipidomics in order to capture individual cell diversity and improve batch harmonization. Lipidomics studies rely on the production of large datasets, from which useful biological information is extracted. Modern methods in machine learning (ML) are necessary to extract this information and identify molecular predictors of cellular properties such as potency.Ph.D.Chemistry and Biochemistr