71446 research outputs found
Sort by
Whispers in the Data: Exploring Advice-Driven Sketching Algorithms
No full textIn modern systems, sketching algorithms are a common tool to analyze vast amounts of data. Recent work has explored ways to apply an advice model on top of these algorithms, where machine learning models provide hints when processing elements. This enables these sketching algorithms to provide even better performance as they are guided by the models.
This thesis focuses on these sketching algorithms that use advice. I examine and consider the limitations of existing algorithms, then design a new sketching framework to work around these limitations, which I dub the Bucketing sketch. Using this framework, I focus on three key computational challenges: estimating high-frequency moments for increment-only data streams, handling turnstile streams, and quantile estimation for aggregate values. For high-frequency moments (defined as f(\boldx) = \sum x^d for ), I prove that my approach can achieve precise estimates using logarithmic space, requiring only minimal assumptions about the underlying data-generating process. The remaining two problems are challenging for existing sketching algorithms; nevertheless, I show strong experimental results for all three problems, highlighting the flexibility and power of the Bucketing sketch.Computer Scienc
LESIONS ASSOCIATED WITH HYPERSOMNIA MAP TO A COMMON BRAIN NETWORK
No full textAbstract.
Background: Hypersomnia is characterized by excessive daytime sleepiness, severely impairing daily functioning and often a significant source of distress for affected individuals. Brain lesions associated with new-onset hypersomnia may provide causal insights about neuroanatomy critical for sleep regulation. Moreover, mapping the whole-brain functional connectivity of these lesions may identify the brain networks that regulate sleep and wakefulness. The hypothesis that lesions causing hypersomnia map to a common brain network has not been formally tested.
Methods:
We conducted a systematic review of the literature and identified 63 hypersomnia-related lesions. Each lesion was mapped onto a common brain template, and functional networks were computed using resting-state functional connectivity data from 1,000 healthy individuals. Sensitivity analyses were performed at two thresholds (T > 4.7 and T > 7). A one-sample Welch’s t-test was performed to account for multiple comparisons. Two independent specificity analyses were conducted by comparing the hypersomnia network with lesion connectivity associated with other neuropsychiatric conditions (n = 997) and a cohort of stroke lesions (n = 135). We created a conjunction map and used it as a region of interest (ROI) to derive a hypersomnia brain network. We formally tested the hypothesis that the hypersomnia network exhibits preferential functional connectivity with the posterior hypothalamus, rather than the anterior hypothalamus, using two different ROIs from two separate atlases. The functional connectivity of the entire hypersomnia cohort to these ROIs was averaged, and a paired t-test was performed. Given the phenotypic similarities of hypersomnolence and coma, a one-sample t-test (FWE-corrected) was conducted to test the similarity of hypersomnia and coma-associated brain networks. An exploratory voxel-wise-two-sample t-test (FWE-corrected) was conducted to identify brain network patterns differentially associated with hypersomnia and coma, respectively. Lastly, we conducted an exploratory assay to compare the spatial topography of our hypersomnia network to brain networks associated with other neurological and psychiatric conditions.
Results: Hypersomnia-causing lesions are part of a common brain network (in >95% of cases) connected to relevant arousal centers, which is both specific and consistent. The posterior hypothalamus shows preferential functional connectivity to hypersomnia-associated lesions, in contrast to the anterior hypothalamus. There is a significant similarity in functional connectivity networks between hypersomnia and coma. However, exploratory analysis reveals distinguishing properties between them. Our hypersomnia network demonstrates strong, significant spatial similarity with lesion networks associated with coma.
Conclusion: Hypersomnia-associated lesions map to a common brain network, encompassing critical arousal centers and regions outside the classical ascending reticular activating system (ARAS). The functional connectivity networks are significantly similar between hypersomnia and coma. However, our exploratory analysis reveals that the functional connectivity networks of coma originate from the lower brainstem, whereas the functional connectivity networks of hypersomnia originate in the upper brainstem. These findings offer new insights into the functional networks of hypersomnia specifically and conscious arousal generally, and lays the groundwork for proposing new stimulation targets that could alleviate hypersomnia symptoms.Medical Scienc
Entrepreneurial Strategy and Learning
No full textLearning is a cornerstone of competitive advantage, especially in entrepreneurial contexts. Despite extensive research on entrepreneurial learning methodologies, we still lack understanding of fundamental constraints that limit learning effectiveness. This dissertation investigates key structural obstacles that hinder effective entrepreneurial learning: resource constraints, uncertainty, interdependencies, and bounded rationality. It provides novel insights into how these barriers manifest and interact in entrepreneurial contexts. The first study reveals a counterintuitive relationship between experience and strategic foresight: as entrepreneurs add features to successive products, the total interdependencies grow faster than their ability to anticipate them, leading to increasingly inaccurate forecasts. The second study reveals how novel recombination, while often associated with opportunities for outsized returns, also incurs the cost of managing novel complexity—obscuring the time and resources required for execution and increasing the risk of bridge financing and shutdown. The third study identifies a “mediocrity trap," in which firms launching lower-quality products receive ambiguous feedback and, as a result, persist longer despite negative market signals. This study challenges assumptions about learning with minimum viable products and rapid pivoting. Together, these studies highlight critical tensions in entrepreneurial learning and demonstrate that many learning challenges stem from structural barriers rather than purely cognitive limitations. This integrated perspective contributes to scholarly understanding of entrepreneurial strategy while offering practical approaches for entrepreneurs seeking to navigate these barriers and learn under constraint.Business Administratio
Path to Protein Dynamics: Advancing Crystallographic Data Analysis via Deep Generative Models
No full textArtificial neural networks and machine learning methods have transformed numerous scientific research paradigms. A milestone in this transformation was the release of AlphaFold2 in 2021, DeepMind's machine learning model for protein structure prediction. By delivering an accurate and accessible computational solution for predicting static protein structures from sequence and coevolutionary data, AlphaFold2 marked a new era for structural biology. This breakthrough, beyond its technical wizardry, owes much of its success to the availability of extensive protein structure databases built over decades of community effort, particularly the Protein Data Bank. This underscores a boarder lesson: foundation models thrive on large, well-curated data. In structural biology, the next frontier is to understand protein dynamics, their responses to perturbations such as binding and mutations, and their interactions with biomolecules like nucleic acids. While deep learning approaches hold promise, they are limited by the scarcity of large, high-quality datasets that capture this complexity. At the same time, advancements in hardware and techniques have greatly increased the throughput of raw data reporting on protein dynamics and conformational heterogeneity across modalities such as X-ray crystallography and cryo-EM. However, inherent pathologies and noise in biophysical data, as well as the inherently rugged nature of protein (free) energy landscapes, continue to make model building a major bottleneck, limiting our ability to interpret these datasets with confidence and efficiency. Significant human effort and intuition are still required. This dissertation presents new methods to expedite automated model constructions with experimental data under Bayes' principle. First, I introduce SFCalculator, a GPU-accelerated, differentiable function for calculating atomic model agreement with crystallographic data, bridging crystallographic data analysis and machine learning techniques. Second, I describe how this interface can aid in interpreting crystallographic data by incorporating priors from molecular mechanics or pretrained predictive models, while using experimental data to guide generative models for improved accuracy. I then introduce VAE-Assisted Ligand Discovery (VALDO), a method aimed at boosting the signal-to-noise ratio in crystallographic fragment screening for drug discovery. Collectively, these developments demonstrate how generative models can be applied to advance crystallography data analysis in a principled manner. Finally, I offer insights into the importance and potential pathways for inferring protein dynamics by integrating experimental data from diverse modalities.Engineering and Applied Sciences - Applied Physic
Distributed encoding of natural and drug-induced physiological states in the insular cortex
No full textInteroception—the sensing of internal bodily signals—is crucial for maintaining homeostasis and plays a significant role in pathological states including drug addiction. Both the rewarding aspects of drug consumption and the aversive effects of withdrawal are experienced as salient body states. The insular cortex (InsCtx) is a key interoceptive region that integrates external sensory, visceral, and limbic information, and has been implicated in the maintenance of nicotine dependence. Here, we combine chronic two-photon imaging of hundreds of InsCtx neurons with physiological recordings of heart rate, pupil area, and body temperature during repeated nicotine exposure. We find that InsCtx neurons exhibit stable, distributed encoding of natural physiological states, enabling accurate predictions of arousal and cardiovascular variables across days. Nicotine administration triggers unique, centrally mediated physiological changes, which are reflected in distinct patterns of InsCtx activity. Longitudinal nicotine administration resulted in physiological tolerance, which contrasts with nicotine-evoked InsCtx neural responses that did not adapt across days. This study highlights the InsCtx’s role in tracking and distinguishing between natural and drug-induced states, and offers insight into the neural basis of tolerance -- a core feature of addiction.Neuroscienc
At the Center of It All: Applying Network Analysis to Authoritarian Elite Shareholders in Singapore
No full textAuthoritarian regimes concentrate power in an exclusive elite. When the elite depends on state access for power, the regime is able to discipline and reward these elites through interventions on their supply and demand. When the regime faces a contestation to power, authoritarian theory reveals two possibilities. One, the regime may expand distribution of benefits to co-opt outsiders. Two, the regime may further concentrate benefits to trusted elites. I test these hypotheses in Singapore, a state with clear authoritarian origins, an elite of government-linked companies (GLCs), and recent existential political threats to power. Utilizing data from nearly 30,000 government contracts across six years, I find evidence of concentration of benefits over the period as changes in total spend outpace changes in supplier count. I then bring novel applications of network analysis to the shareholders who own the suppliers. Through nodal representations and robust centrality indices, I find that shareholder centrality grows in importance, indicating that contracts go to more connected suppliers. The top shareholders at all times are GLCs, affirming that the developmental elite still dominates government contracts today. These findings support the latter hypothesis of concentration of benefits, though future work is needed to understand causal relationships and fringe elites.Computer Scienc
A Biography of Ancrene Wisse
No full textIn the wider field of narrative nonfiction and the narrower genre of biography, there is a trend in recent years to write of the “life” and cultural import of inanimate objects, from spoons to sugar. This has extended to works of literature, and it is in this tradition that I am writing a biography of Ancrene Wisse (A Guide for Anchoresses), a book with 800 years of influence over a variety of audiences. A precise date for the creation of this text is difficult to pin down, but the scholarly consensus has landed on 1220–1230. It was written mainly for an audience of anchoresses, women who lived a reclusive religious life inside a cell attached to a church. Its influence in England quickly grew, and as it did, the practice of anchoritism grew as well. I argue, in this biography of the text, that the popularity of the Ancrene Wisse and the anchoresses who were its initial (but not ever exclusive) audience grew in tandem until the Protestant Reformation in the sixteenth century. While the practice of anchoritism all but died out, Ancrene Wisse has continued to have an active afterlife among scholars in several disciplines.Extension Studie
Human-Centric Mobility: Exploring Non-Work Travel Patterns in D.C.
No full textPublic transportation in the U.S. has been designed to serve worker travel patterns. However, only 18% of urban mobility trips are for the purpose of work. Further, post-pandemic, the proportion of trips taken for the purpose of work is decreasing. In order to boost ridership and more equitably serve their communities, public transportation agencies need to consider non-work travel patterns. Using mobile phone data and tap-in/tap-out public transportation data from D.C., this thesis explores the convenience disparities and changes in demand for non-work travel patterns on public transportation. We found that approximately 10% more trips are for the purpose of travelling from the home to a non-work destination post-pandemic compared to pre-pandemic. We found there is a significant convenience burden for non-work trips on public transportation. Non-work trips take 85% longer in travel time and are 55% slower in travel speed than work trips.
This project is in collaboration with the Washington Metropolitan Area Transit Authority
(WMATA).Computer Scienc
Portrait of the arzet: The Doctor in German Literature of the High and Late Middle Ages
No full textThe healing arts in Western Europe of the traditionally termed High and Late Middle Ages, the period roughly from 1050 to 1500 CE, embraced a wide and dizzying range of practitioners, practices, and conceptual underpinnings. Regardless of who was trying to heal whom, medieval medicine, at least as practiced in Christian Europe, was largely ineffective. It may have had a placebo effect; the academic rhetoric that came to embellish it might have convinced some of its efficacy; but it did little to affect pathology or to stop the spread of diseases in a society, the majority of whose members subsisted in grossly unhealthful conditions. Nonetheless, the belief in and demand for therapeutic services of all types from a vast array of healers—secular and clerical, university trained and lay—remained strong through and beyond the Black Death in 1348. In light of this persistent faith in the possibility of healing, it is no surprise that the figure of the doctor—the arzet or arzatinne—appears so frequently in Middle High German texts of the period.
This dissertation examines the perceptions of medieval medicine as reflected in a variety of such texts across multiple genres. It explores the variety of “attitudes” evinced towards the medical practitioner, ranging from the adulation of the wundarzet in epic poetry and of the saintly healer in hagiography; to the skepticism voiced by the preacher of the people, Berthold von Regensburg; to the ambiguous portrayal of secular healing in works such as Hartmann von Aue’s Erec or Der arme Heinrich. Alternatively, the doctor is fiendish in Reinhart Fuchs, a fraudster in Der Pfaffe Amis, and a fool in the merchant scenes of the Easter Plays. Medieval mystics who chose to rely on Christ’s healing power alone welcomed illness and disdained secular healing. And all the while, university-trained professionals attempted to ground explanations for the course and outcome of the illnesses they encountered, for their own treatment successes and failures, in Galenic science—a system often indistinguishable from magic. Notably absent from medieval German literature is any instance of arrant contempt for the doctor. Such expressions of contempt would have to wait for the modern era—a time in which paradoxically, although medicine has become more reliable, respect for and fear of medical authority has waned.Germanic Languages and Literature
Rikke and the Frost-Star Cat
No full textRikke and the Frost-Star Cat is a young adult polar fantasy novel. Rikke, the heroine, sets out through the icy landscape, which is both unforgiving and (secretly) magical, in search of life-saving medicine for her grandfather Mikkel and answers to questions about her family history. Rikke’s childhood has been filled with love and friendship, but dangers loom all around. The villagers have struggled for the past generation, since the ruling matriarchy was overthrown by a vicious group called The Blades. Rikke has only known life under the strict rule of the Blades, and her days are marked by the contrast between the cozy hygge inside her family home and the oppressive governing of these armed men. Unbeknownst to Rikke, a rebel group has been quietly plotting to return the matriarchy to power. As she traverses the frozen forests, Rikke will encounter friends and potential foes as she uncovers the mystery of her heritage, including the unknown legacy and hidden gifts of the women of her family.Extension Studie