49755 research outputs found
Sort by
Probabilistic and Optimization Methods for Biomedical and Single-Cell Studies
No full textIn this thesis, we explore probabilistic and optimization methods to address critical challenges in biomedical and single-cell studies. These methods encompass diverse applications, from group testing strategies to advanced computational approaches for single-cell data analysis, highlighting their utility across scales and disciplines.
First, we develop a novel probabilistic framework for one-stage noisy group testing, a technique that efficiently identifies infected individuals within large populations. Motivated by practical needs during the COVID-19 pandemic, we propose a pooling design guided by maximizing pool entropy and a maximum-likelihood recovery algorithm. Our findings reveal the interplay between pooling parameters and randomness in infection vectors, offering a robust and adaptable group testing strategy.
Second, we introduce CellStitch, a 3D cell segmentation method leveraging optimal transport to overcome the challenges of segmenting anisotropic microscopy images. Unlike most existing segmentation approaches, CellStitch circumvents the need for large 3D training datasets by aligning cellular correspondences across imaging layers. Benchmarking on diverse plant microscopy datasets demonstrates that CellStitch outperforms state-of-the-art segmentation methods, particularly on images with high anisotropy, enabling accurate analysis of 3D cellular structures.
Finally, we present RefCM, an algorithm for automating cell-type annotation in single-cell RNA sequencing data based on computing Wasserstein distance between cell populations. RefCM measures gene expression distribution similarities and solves an integer program to align query clusters with reference annotations. Our method achieves superior accuracy in cross-technology, cross-tissue, and cross-species mappings, addressing a critical need for robust annotation in single-cell studies and broadening the applicability of scRNA-seq technologies.
Together, these contributions demonstrate the power of probabilistic modeling and optimization methods in advancing the efficiency and accuracy of biomedical and single-cell analyses, providing tools to tackle challenges in diverse biological contexts
Scalable Silicon Systems for the Accelerator Era
No full textSystem-on-chip (SoC) architectures have evolved rapidly in the past two decades. The early 2000s saw a shift from single-core to multi-core architectures due to technology scaling trends. In the past decade, the limits of multi-core scaling have instead pushed SoCs into an era of specialization. The semiconductor industry has turned to heterogeneous SoCs, which combine general-purpose processors with various specialized hardware units called accelerators, to continue to push the bounds of performance and energy efficiency. As this trend has become pervasive across various domains, we are surely in the Accelerator Era of computing.
Accelerator-rich SoCs, however, are much more difficult to implement than their homogeneous counterparts, as they require the design, integration, and verification of many different components. One solution to lower the growing design costs is to reuse components developed by other teams. However, small academic and industrial teams do not have the budgets to acquire expensive intellectual property. The growing open-source hardware (OSH) movement presents an opportunity to foster truly collaborative engineering. Recently, there has been notable progress towards an OSH ecosystem, thanks to an increased availability of OSH components and tools.
Despite this progress, several key challenges remain. First, integrating many components, designed by different teams, with different languages and tools is no easy task; arriving at silicon implementations, or even just FPGA prototypes, of complete SoCs can be daunting for small teams. Second, the usage of many heterogeneous components can cause contention on key system resources, such as memory, interconnect, and power. Because these components are designed independently, they have no notion of the status of the surrounding system.
Methods to optimize system-level performance must therefore be independent of the design of components themselves. My thesis is that decoupling the design of components, in particular accelerators, from the surrounding SoC architecture is the key to supporting flexible integration and efficient resource management, thereby enabling the scalable design of complex silicon prototypes and the optimization of system-level performance.
Throughout this dissertation, I identify the shortcomings of existing solutions to various computer architecture and integrated-circuit design problems when they are applied to heterogeneous SoCs; the characteristics of these systems necessitate a rethinking of traditional approaches. I propose new solutions that are scalable to large designs and are not only tailored to the diverse characteristics of accelerators but are also decoupled from the design of any particular accelerator. Specifically, these solutions innovate on the management of critical shared resources – such as the memory hierarchy, on-chip interconnect, and power – in many-accelerator SoCs. The proposed solutions are built on top of ESP, an open-source platform for heterogeneous SoC design, to allow for their rigorous evaluation in complete SoC prototypes. On top of FPGA-based prototyping, these solutions are also implemented in fabricated ESP-based chips, thanks to a new agile design methodology for silicon prototyping. I show how this methodology enabled small teams to implement several complex silicon prototypes, such as the EPOCHS-1 SoC. With 14 different types of accelerators, multiple RISC-V cores that coherently boot a Linux operating system, and novel strategies for data orchestration and power management, EPOCHS-1 is arguably the most complex SoC demonstrated in the literature by an academic team.
The contributions of my dissertation also have broader impact on both research and teaching. I have contributed all of my work back into the open-source release of ESP, thereby strengthening its offering to the OSH community. In fact, in just the past few years, ESP has been used for research at over 20 institutions. I have also helped bring ESP into the classroom at Columbia University to teach collaborative SoC design to the next generation of engineers
“Windows Turned Sideways Make Bridges”: Rooting a Pedagogy of Connectedness in the Secondary ELA Classroom
No full textIn this dissertation, I examined how three female secondary English Language Arts (ELA) teachers understand what it means to teach diverse literature and how they navigated particular challenges and tensions that arise in their New York City public high school. Through narrative inquiry, this study examined when select secondary ELA female teachers encountered ruptures in teaching diverse literature within their constructed curriculum. More specifically, the narratives of three female educators’ experience provided insights into the processes of how the teachers developed literacy practices for their students. As a result, qualitative data (including individual interview data, audio-transcribed meeting data, and student-generated work) were collected.
This study utilized a combination of conceptual and theoretical frameworks of silences and Spivak’s notions of marginality that are grounded, respectively, in power structures, cultural norms, and dominant ideologies to interpret the data. These frameworks provided critical lenses for understanding the pedagogical decisions and instructional strategies that teachers enacted in their secondary ELA classrooms. The concept of silences can take multiple forms including institutional silences (omitted narratives in the curriculum) and classroom silences (teachers hesitating to engage in discussions about race, gender, or oppression). This conceptual framework of silences helped examine what is left unsaid, who is permitted to speak, and how power dynamics shape discourse around diverse literature in the classroom. Applying Spivak’s theories, this study considered how teachers engage with marginalized voices in literature, how they amplify or silence those perspectives, and how they challenge or reinforce dominant ideologies in their pedagogy. This theoretical framework also interrogated whether teachers feel constrained by department and school policies when selecting and teaching diverse texts.
The data revealed that teachers’ interpretations of “diverse literature” and their pedagogical choices were shaped by multiple intersecting factors. Primarily, their personal lived experiences, school demographics during their early teaching years, and broader sociopolitical contexts throughout their professional careers played significant roles in influencing how they approached diverse literature in the classroom. These factors collectively shaped how teachers navigated tensions around diverse literature—whether in choosing texts, framing discussions, or responding to students during classroom discussions.
Understanding these influences helps illuminate the complexities of teaching diverse literature in secondary ELA classrooms, particularly how personal and professional contexts intersect with larger societal forces.
Implications of the study included:
• the importance of advocating a pedagogical and curricular design practice toward teaching diverse literature through intentional pedagogies that integrate intersectionality and fosters connections across multiple marginalized groups, rather than treating diversity as separate categories;
• promoting increased and continuous dialogue between secondary ELA teachers, students, and researchers to encourage further interrogation of what it means to teach diverse texts in a meaningful way that allows teachers and students to navigate through differences and such tensions in a more mindful manner; and
• investing in teacher education development programs and courses for preservice and current teachers on how to have crucial conversations that allow them to navigate with students on difficult topics like race and gender.
With these implications, we hope to push dialogue toward understanding each other and of ourselves more deeply while acknowledging the sociopolitical contexts in which we are all embedded in
Essays in Urban and Spatial Economics
No full textThe essays in this dissertation concern the distribution of people and economic activity across space at the national, regional, and urban levels.
The first chapter studies the consistent power law observed in city-size distributions across countries, often referred to as Zipf’s law. To study this phenomenon, we analyze a general spatial equilibrium framework with heterogeneous locations, where trade and the locational productivities and amenities are subject to random variation in geographic and climatic characteristics. We prove how population is distributed spatially due to such random variation across space, demonstrating how population is lognormally distributed within countries and that the largest locations, i.e. cities, within countries appear to follow a power law.
The second chapter studies changes in the distribution of people regionally in the United States between 1940 and 2000. The “Sun Belt" is a region of the southern United States notable for significant population growth in recent decades, and an intuitive explanation is that the quality of life in the region rose due to increased valuation of weather-related amenities. Existing amenity value estimates for the Sun Belt from canonical spatial equilibrium models, however, are low and are estimated to have fallen in recent decades. I investigate two hypotheses using the canonical Rosen-Roback equilibrium framework: that amenity valuations have fallen in the Sun Belt during the 20th century, and whether geography and climate were an important channel affecting quality of life and population growth. I find that between 1940 and 2000, amenity values in the Sun Belt rose, in contrast to previous claims in the literature. However, using instrumental variables analy- sis, I find that weather’s effect on quality of life is not a significant driver of Sun Belt migration in canonical spatial equilibrium models.
The third and final chapter studies patterns in racial segregation in American cities between 1970 and 2000. We build a general equilibrium discrete choice model with spatial spillovers on racial preferences, with the model itself nesting canonical theories from the economics literature about the cross-section and dynamics of segregation. We simulate the model and test its predictions on U.S. Census data, finding that racial segregation is an inherently spatial phenomenon. In particular and in contrast to much of the existing literature, we find that people of the same race cluster in nearby neighborhoods, that clusters are common features of U.S. cities across time, and that racial change occurs at the boundaries of clusters
How Community College Students Choose a Program of Study: Faculty and Staff Perspectives
No full textBased on faculty and staff interviews at four colleges, this brief describes how students experience the program choice process, including barriers they face, and discusses practices for improving support for program and career exploration
Optimizing Memory and Storage Performance in Cloud Datacenters
No full textData-intensive applications increasingly dominate modern datacenters. This surge is propelled by a number of applications including AI/ML training and inference, HPC, data lakes, and large-scale cloud storage systems. However, software overhead remains a persistent challenge, with studies showing nearly half of cloud computing cycles wasted. As storage speeds outpace traditional system architectures, operating system overhead has emerged as a dominant bottleneck constraining overall system performance.
This thesis addresses these critical inefficiencies through a three-pronged approach. First, we revisit extensible operating systems as an effective way to optimize the storage and memory stack in cloud datacenters. Second, we uncover new sources of memory usage inefficiencies in ML accelerators, by constructing novel performance profiling tools. Third, we revisit and optimize traditional distributed protocols.
This thesis addresses these critical inefficiencies through a three-pronged approach that targets key sources of storage and memory overhead. First, I focus on developing efficient storage and memory stacks through OS extensions, enhancing system efficiency through bypassing OS layers and by enabling flexible policies for key OS components.
Second, I target suboptimal memory utilization in ML accelerators by developing novel performance debugging tools to analyze the low-level interaction of model code with the micro-architecture. This unlocks new insights into model inefficiencies and how to address them.
Third, I address storage efficiency through revisiting traditional storage protocols like replication. More specifically, I enhance async replication with strong staleness guarantees and fast failover, enabling better application performance and storage utilization
A Three-Study Examination of Emotion Regulation: Addressing Challenges in Assessment
No full textEmotion regulation, or individuals’ attempts to influence emotional experience or expression, is a transdiagnostic process related to the development, maintenance, and treatment of multiple mental health conditions. Despite a proliferation of research over the past 20 years, emotion regulation remains difficult to assess with accuracy and precision, limiting opportunities for study comparison and hampering targeted treatment development. Across three studies, this dissertation addresses key challenges to emotion regulation assessment.
Study 1 introduces and examines the psychometric properties of a novel self-report measure of behavioral dysregulation across two samples. Study 2 examines relationships among momentary negative affect and emotion regulation, and later worry and rumination, in a sample of individuals with generalized anxiety disorder and healthy controls. The use of ecological momentary assessment (EMA) methods in this study further highlights the importance of capturing the functional nature of regulatory processes as they unfold in real time.
Finally, Study 3 also uses EMA to examine associations between momentary emotion regulation and recall bias in self-reported negative and positive emotion in a transdiagnostic clinical sample. The dissertation concludes with a general discussion regarding the implications of these three studies for improving emotion regulation assessment, including limitations and avenues for future research
Understanding Regional Aerosol Variability and Trends Through Satellite Constraints on Source Region Drivers and Arctic Processes
No full textGlobally, aerosols exert a strong influence on Earth’s energy balance by absorbing or reflecting solar radiation, and modifying cloud radiative characteristics. However, both global and regional radiative effects from aerosol remain highly uncertain, both in the present climate and under future emissions scenarios. In the present climate, some of that uncertainty arises from poor constraints on the variability of aerosol loadings in space and time.
This is especially true in remote regions like the Arctic, where observations are sparse. As aerosol radiative effects depend on factors such as solar insolation, temperature, surface albedo, and moisture availability---all of which also vary across regions and seasons---improving constraints on the spatiotemporal variability of aerosol within remote regions, and the processes governing that variability, is an important step for understanding the energy budget in such locations.
The Arctic in particular is both highly sensitive to variations in radiative forcing, and also shapes important feedbacks that affect the rest of the global climate. Hence, constraints on the processes governing aerosol variability in this region are especially important for understanding both the regional energy balance and the long term effects of Arctic warming on the broader climate system. In populated, industrial regions, observations are less limited, but uncertainty surrounding future aerosol impacts arises from challenges disentangling natural and anthropogenic signals, as well as scenario uncertainty stemming from the inherent unpredictability of human activities. Recent air pollution regulations in many countries have produced multi-year declines in anthropogenic emissions.
At the same time, increasing summertime emissions from wildfires have changed the chemical and seasonal distribution of global and regional aerosol burdens. In the near-future, anticipated further declines in industrial emissions are expected to unmask additional warming from greenhouse gases. However, such declines are likely to be regionally inhomogenous, and the extent to which political and social changes influence local emissions remains difficult to predict. Lockdown periods during the COVID-19 pandemic provided an opportunity to examine the effects of lifestyle changes on aerosol burdens, globally and in different source regions, and to disentangle the effects of societal changes from long-term trends and natural sources of variability.
Satellite observations of aerosol optical depth (AOD) are widely used for assessing variability and trends in global and regional aerosol burdens, providing high resolution, long-term coverage across much of the globe. Indeed, these data products play a central role in this dissertation. However, over the course of my research I found that many satellite and reanalysis AOD products exhibited a seasonal cycle opposite to that found in ground-based station measures and satellite lidar products, over the mid to high latitudes. This discrepancy suggests that biases in retrieval processing may be distorting representations of seasonality in these regions. Understanding the causes of these biases and assessing the magnitude of their effects is an important step toward improving aerosol characterization in future research, while also offering an opportunity to address fundamental questions in remote sensing.
In Chapters 2 and 3 of this dissertation, I use satellite observations of aerosol optical depth (AOD) to examine the processes governing aerosol spatiotemporal variability in the Arctic, and to disentangle the effects of long-term trends, natural variability, and pandemic-related lifestyle changes on source region aerosol burdens over the first year of the pandemic. In Chapter 4, I assess the extent to which sampling biases, data quality, retrieval geometry, and lidar retrieval artifacts contribute to the biases in seasonality described above.
To address the first question, I applied a K-Means clustering algorithm to monthly median (2007-2021) Arctic AODs, finding four distinct aerosol seasonality regimes. The subregions corresponding to each regime exhibited distinct topographic, ecological, and meteorological characteristics that likely affect transport, emissions, and deposition. This chapter constrains aerosol variability in the Arctic and identifies potential drivers of subregional variations in seasonality.To understand the role of lifestyle changes on regional aerosol burdens, I compared the effects of the pandemic lockdowns with long-term trends and variations in AOD from natural aerosol sources, such as dust and smoke, in four major Northern Hemisphere source regions.
I found that in most regions, the lockdown-signature was smaller than the effects of long-term trends. In one region (India), natural variability from dust emissions eclipsed the lockdown-signature by an order of magnitude, while smoke emissions from wildfires in the Western United States (US) masked pandemic-signatures in both the US and Europe in the later part of the year. These findings suggest that emissions changes resulting from individual lifestyle changes may play a relatively minor role in shaping global and regional aerosol burdens, and that policy-driven changes, climate change feedbacks, and natural variability in aerosol emissions may exert greater influence on future aerosol trends and their associated climate impacts.
Finally, in Chapter 4 I use colocated retrievals from lidar and a passive sensor instrument to show that seasonality biases between data products arise from the interplay between passive sensor retrieval quality and retrieval geometry. Specifically, passive sensor retrievals flagged as "low-quality" declined relative to lidar measures as the solar zenith angle (SZA) increased, while those flagged as "high-quality" remained stable. In the winter in the mid to high latitudes, "low-quality" retrievals predominate and the SZA is high, resulting in systematically lower average passive sensor AODs. In contrast, the NH summer is characterized primarily by “high-quality” retrievals at most latitudes, such that aggregate products maintain a constant high bias relative to lidar.
I further demonstrate that the drivers of globally low biases in lidar products do not vary with solar geometry, and that the effects of sampling biases on seasonality are small compared to the combined influence of retrieval quality and the SZA on passive sensor retrievals. The findings described in Chapter 4 will help data users interpret measurements of Arctic and midlatitude aerosol in different seasons, while also contributing to improved constraints on satellite retrievals of reflectance and AOD under complex viewing conditions
The Universal Postal Union, Parcel Post, and Postal Policy in the United States of America, 1894–1912
No full textThis essay traces the influence of the Universal Postal Union (UPU) on the public debate that took place in United States between 1894 and 1912 over the expansion of the mandate of the US Post Office Department to admit merchandise into the mail. This debate can be said to have begun in 1894 with the publication of an audacious tract by Cowles in which he proposed the establishment of a nationwide government-owned merchandise-delivery channel to be operated by the US Post Office Department. The merchandise-delivery issue would remain on the public agenda until 1912, when the US Congress enacted the first comprehensive parcel post law. This new law would go into effect on New Year’s Day in 1913; among its most prominent advocates was Oregon Senator Jonathan Bourne Jr. Though the UPU influenced this public
debate in various ways, it had little effect on either the timing of the 1912 law or its rationale. Far more important was the shifting balance of power in the US Congress, and, in particular, the emergence after 1910 of a bloc of lawmakers sympathetic to big-city mail-order houses and unbeholden to the powerful phalanx of country shopkeepers who had for over a decade mounted a powerful challenge to any legislation designed to reconfigure the channels of trade
Illuminating Subicular Dynamics through Multiphoton Holography
No full textThe subiculum has a poorly-defined, but necessary, role in the neurobiology of memory, despite a rich tradition of research investigating hippocampal contributions to memory. While traditional models of memory retrieval highlight the reactivation of a specific engram formed during learning, emerging evidence suggests the subiculum is necessary for the retrieval of trace fear memories without being necessary for their initial encoding.
This presents a central paradox: how can a brain region be critical for retrieving a memory it wasn't necessary to form? Given the subiculum’s unique, recurrent architecture, its diverse array of projections, and a broad implication in the etiology of epilepsy, we hypothesized that the subiculum may be facilitating engram reactivation through the amplification of CA1 output. To test this hypothesis, we conducted multiphoton holographic calcium imaging of the subiculum and subicular-hippocampal border region during an ethologically-grounded, trace-fear conditioning task. In this task, mice learned to associate one of two tones (CS+, CS-) with an aversive stimulus despite being separated by a 10-second stimulus-free trace period. Conditioned fear expression took the form of burrowing: head-fixed mice retract a lightweight “burrow” that slides along a frictionless rail.
Surprisingly, we observed robust activity in response to both the CS+ and CS-. We observed rapid, cue-onset evoked responses specific to both the CS+ and CS-. In contrast, subicular activity exhibited reciprocal changes at cue-offset. Unexpectedly, the offset of CS- tone triggered an unexpected, immediate burst of activity. Meanwhile, we observed a substantial reduction of activity in specific neurons during CS+ trials, which peaked near the anticipated US delivery. Our result suggests that the subiculum behaves less like a conveyor belt for a canonical replay of a fixed engram and more like a signal processing hub that projects dense, high-dimensional information from which downstream regions can flexibly extract information without requiring substantial network reconfiguration