1983 research outputs found
Sort by
The Roads Not Taken: Model Multiplicity in Machine Learning
No full textIn machine learning, model multiplicity is the existence of multiple models that perform equally well for a given prediction task (also known as the "Rashomon effect" ). The set of near-optimal models is referred to as the "Rashomon set." Predictive multiplicity examines how predictions change over this set of near-optimal models. If model outputs vary significantly across similar models, this information can offer insight into predictive arbitrariness. In this thesis, I introduce frameworks for evaluating and leveraging predictive multiplicity in different settings.
First, I present methods to measure predictive multiplicity in probabilistic classification (predicting the probability of a positive outcome) and develop optimization-based methods to compute these measures efficiently and reliably for convex empirical risk minimization problems. Empirical results show that real-world probabilistic classification tasks can in fact admit competing models that assign substantially different risk estimates. Additionally, I provide insight into how predictive multiplicity arises by analyzing dataset characteristics.
Second, I formulate predictive multiplicity analysis in a resource constrained setting recognizing that predictive allocation tasks are governed by a resource budget. I also extend the multiplicity framing, outlining the concept of multi-target multiplicity for quantifying the impact of choices made in regard to target specification for a given predictive allocation task. With this framework, I demonstrate how to fit separate models that are useful for predicting the three outcomes of interest independently and arriving at a way of ranking patients that results in a more equitable allocation.
Third, I investigate the connections between predictive multiplicity and predictive churn which is the change in predictions pre- and post- model update in response to a change in training data. I present empirical and theoretical results on characterizing churn in terms of the Rashomon set. Results show that churn unstable points overlap by more than 50 percent with ambiguity points. This points to similarities in the two concepts. Theoretical results to characterize predictive churn between two Rashomon sets as well as churn between models within one Rashomon set hinges on the type of Rashomon set.
I focus on predictive multiplicity to advocate for transparency in the prediction model training procedure. These methods to evaluate predictive multiplicity, as well as connections with predictive churn, contribute to a larger effort for machine learning researchers to be accountable to the individuals affected by model predictions. Similar to a person deciding between roads to take while travelling, insight into alternative options (i.e., roads not taken) may provide insight into the significance of the decisions made
Infection Dynamics and Cell Type Specific Responses to Vesicular Stomatitis Virus Infection in the Mouse Brain
No full textViruses are excellent cell biologists. They can be harnessed to interrogate biological systems and co-opted as tools for therapeutic or research purposes. For a tool to operate best, it’s crucial to know how it functions in the environment in which it is utilized. The neuroscience field has adopted neurotropic viruses as transsynaptic tracers to label and modify neural circuits. Although these viruses are used as practical tools in the brain, our understanding of their specific cellular interactions and the brain tissue’s response to them are still limited. We aimed to address these knowledge gaps by developing a viral system to track differential states of infection, better defining tropism in a cell type specific manner, and investigating factors that influence viral infection outcomes. Additionally, we profiled the brain’s response to viral infection and monitored how various cell types contribute to this response over time. Our findings better inform viral tool design for neuroscience and broaden our understanding of the early immune response to viral infections in the brain parenchyma.Medical SciencesMedical Science
Developing an Educational Environmental Game
No full textWiththesteadyimprovementoftechnologicalpowerandaccessibility,games
havebecomewidespreadandpopularamongallages.Educationalgamesespecially
havethepotentialtofosterlearningefficientlyandenjoyably.Thegoalofthisproject
wastocreateanenvironmentalgamethatwassimultaneouslyengagingandrealenough
toinformplayersaboutthenuancesofclimatechange.Thegamewasbuiltcompletely
intheUnitygameengine,andreliesonanenvironmentalmodeltomakeclimate
simulationsbasedonactionstakenbytheplayerin-game.Themodelrunsinthe
backgroundwithinthegame.Theendproductachievedthisgoal,thoughwithalimited
scope.Thegame,calledNP(NaturalPremier)Enviro(mental)Game24,makesuseof
bothactiveandpassiveuseractiontoengageplayers.Thegamewasdesignedtogamify
importantmethodsofclimateactiontogiveplayersinspirationonrealworldclimate
problems.TheEnvironmentalscienceisalsoeasilydigestible,allowingforanenjoyable
experienceregardlessoftheplayer’slevelofbackgroundknowledge.Thoughmuchof
theprojectisthegameitself,thispaperisintendedtoprovidecontextonthepurposeof
thegame,designchoices,andfurtherstepsbeyondthisprojec
States of Risk: policing, security, and abolition in South Africa, 1652-2022
No full textStates of Risk: policing, security, and abolition in South Africa, 1652-2022 offers the first history of policing in South Africa that begins with its inception in 1652 and continues to the present day. Blending twenty-four months of ethnographic research inside private security companies in greater Johannesburg with archival research on colonial state records, missionary accounts, travelogues, newspapers, memoirs, and oral histories, it shows how the institution’s development was driven by successive political economic transformations—the creation of the plantation and the frontier, the abolition of slavery, the advent of industrialization, the invention of apartheid, the dawn of national liberation, and the onset of deindustrialization. Tracing how the latest of these transformations has given rise to the contemporary private security industry—one of the largest for-profit policing sectors in the world, it aims to show how the relationship between social control and racialization has always been globally heterogeneous. More specifically, it argues that the case of present-day South Africa—where a significant portion of the Black working class is not incarcerated but rather enlisted into private police work—troubles foundational assumptions about the relationship between race, social control, and the state in the field of carceral studies. In doing so, it argues that South Africa offers a window into the emerging character of racialization globally and key insights into what movements for police reform and abolition might look like in the Global South
Concentrating Alkalinity for Direct Air Capture of Carbon Dioxide: Using Osmotic Pressure for Concentration and Separation
No full textClimate change has resulted in the pressing need to globally manage carbon sources and sinks. Reducing emissions as much and as fast as possible must be prioritized to avoid the worst harms from global warming, which will be felt most by vulnerable populations. However, if warming is to be halted this century, it has now become apparent that removing atmospheric CO2, on the scale of gigatonnes of CO2 per year, will also be necessary to offset the hardest-to-avoid emissions. Direct air capture technologies, which are industrial processes for chemically removing atmospheric CO2, have been introduced and are being deployed as one such approach for carbon removal. This work specifically contributes to the understanding of aqueous dissolved inorganic carbon-based direct air capture technologies in the following ways:
First, we propose a new approach for removing atmospheric CO2, the alkalinity concentration swing (ACS), driven by concentrating and diluting aqueous alkaline solution, which increases and decreases the partial pressure of CO2 of the solution, respectively. Second, we improve on the ACS process by introducing a selectivity step, which separates bicarbonate ions from carbonate ions. A theoretical investigation reveals that bicarbonate-enrichment allows for reaching higher cycle capacity, higher CO2 partial pressure, and improved absorption rates. Third, nanofiltration is experimentally studied, and confirmed as a mechanism to enrich bicarbonate ions, reaching bicarbonate-carbonate selectivity factors above 30. Fourth, we experimentally demonstrate the ability to use reverse osmosis, a membrane-based separation process driven by applied pressure, as a method for concentrating alkalinity. We find that our proposed approach has significant limitations, but it has certain promising pathways for improvement and feasible deployment, if absorption rate shortcomings are able to be sufficiently addressed through promoters and if energy recovery steps are implemented. Fifth, a generalized theory for acid-base concentration swings is developed, introducing a new conceptual framework for controlling pH through osmotic drive. Sixth, absorption and outgassing rates are experimentally studied for alkaline and weak acid solutions, with and without CO2 promoters. And finally, seventh, a broader process perspective on dissolved inorganic carbon-based DAC is introduced to examine the trade-off between the three key figures of merit: volumetric cycle capacity, absorption flux, and cycle energy
The Imagined Past
No full textWhat is the difference between remembering the past and imagining it? We can remember things we did from an imagined point of view, like when I recall nervously giving a speech as though I were seated in the audience. We can remember past experiences through a generic representation, like when you picture your childhood walk to school. And we can remember feeling one thing through the emotional lens of another, such as past hopefulness in light of a recent disappointment. My dissertation is about the role of imagination in memory, and what its involvement means for the nature of our knowledge of the past
A longitudinal randomized trial of a sustained content literacy intervention from first to second grade: Transfer effects on students’ reading comprehension.
No full textWe developed a sustained content literacy intervention that emphasized building domain and topic knowledge from Grade 1 to Grade 2 and evaluated transfer effects on students’ reading comprehension outcomes. The Model of Reading Engagement (MORE) intervention emphasizes thematic lessons that provide an intellectual framework for helping students connect new learning to a general schema (i.e., how scientists study past events). A total of 30 elementary schools (N = 2,952 students; N = 144 teachers) were randomly assigned to a treatment or control group. Over 12 months, the treatment group students participated in (a) spring Grade 1 thematic content literacy lessons in science and social studies followed by wide reading of thematically related informational texts during summer, and (b) fall to spring Grade 2 thematic content literacy lessons in science. After implementation of Grade 1 thematic content literacy lessons and summer support for reading, treatment group students experienced smaller summer losses on a domain-general measure of reading than control group students. Following the sustained implementation of thematic content literacy lessons in science through Grade 2, treatment group students also outperformed their control group peers on a science content reading comprehension outcome (ES = .18). Furthermore, we found transfer effects on science content reading comprehension that varied by passage-item type (near-, mid-, and far-transfer passages determined by the inclusion and number of directly taught words in passages). A sustained content literacy intervention that aligns content and instruction across grades can help students transfer knowledge to novel reading comprehension tasks.Accepted Manuscrip
Wiring and connectivity enabling CO2 sensitization in the Aedes aegypti mosquito
No full textThe mosquito Aedes aegypti’s human host-seeking behavior depends upon the integration of multiple sensory cues. One of these cues, carbon dioxide (CO2), primes and activates host-seeking behavior, but the neuronal circuits underlying the transduction and processing of CO2 information remain only vaguely understood. In this thesis, I leverage data from synapse-resolution electron microscopy to investigate the central cells and synapses processing CO2 information in the Ae. aegypti brain, uncovering connectivity motifs and subcellular features that may form a neural substrate for robust behavioral and functional response to CO2.
In Chapter 1, I describe the collection of, and connectivity analyses within, a transmission electron microscopy (TEM) dataset of the female Ae. aegypti palp glomeruli; this dataset includes the glomerulus selectively responsive to CO2. While olfactory sensory neurons (OSNs) innervating the CO2 glomerulus do not make more feedforward synapses onto one projection neuron compared to surrounding maxillary palp glomeruli, CO2 OSNs make a relatively high number of recurrent connections among themselves. In Chapter 2, I investigate unique ribbon-synapse-like structures that occur at a subset of these contacts between CO2 OSNs, which may reinforce the recurrent network. In Chapter 3, I describe comparative analyses between the CO2 glomerulus in Ae. aegypti and relevant glomeruli in the well-studied dietary generalist Drosophila melanogaster, finding that the connectivity of the Ae. aegypti CO2 glomerulus is more consistent with a strongly attractive glomerulus (DM1) than the state-dependent aversive CO2 glomerulus (V). Together, this work suggests that attractive glomerular circuits may share feedforward connectivity principles across species and elevated levels of recurrent connectivity and ribbon-like structures may help to support mosquito attraction and sensitization to CO2
Structuring Information: Printed Tables as Organizing Tools in Early Modern Europe
No full textIn this dissertation, I study the organization of information in an era long before Wikipedia and digital spreadsheets. More specifically, I investigate the methods by which qualitative and quantitative tables were printed and used in Europe between 1450 and 1700. I use editions produced across the European continent during this period, although the bulk of my sources come from Germany, the Low Countries, and Italy. My detailed analysis of visual forms and technical processes related to printed tables provides a new perspective on the intellectual and cultural priorities of the early modern period. Printed tables will not explain the Protestant Reformation or the rise of absolutism, but they do stand witness to the creativity of authors and printers seeking to effectively communicate information as simple as basic multiplication and as complicated as world history.
My research grows from a burgeoning discourse on diagrams at the intersection of book history, art history, and history of science. Accordingly, by comparing editions in astronomy and finance as well as linguistics and surveying (to name a few), my project builds on scholarship that has typically distinguished between, rather than synthesized, various genres. The introduction establishes the place occupied by tables within the rich diagrammatic landscape of early modern printing. In the first chapter, I use the calendar as a highly recognizable form of table to see how we can make sense of ubiquity and continuity when it is usually easier for the historian to find unique examples of change over time. The second chapter, on historical tables, demonstrates how and why this diagrammatic tradition survived when a parallel prose tradition of history writing was available and popular during the same period. The third chapter is about long series of quantitative tables, and in it, I investigate how these large collections of information were usable at all by printers and readers alike. The fourth chapter focuses on “unstable tabular forms,” including errors and upside-down presentation, to test the physical and conceptual limits of the table. Finally, the epilogue considers how tabular-adjacent printing, especially accounting ledgers, can inform our understanding of the construction and use of tables proper
Subsistence and Ritual: Animal Economy in the Bronze Age Qaidam Basin in Northwestern China
No full textUnderstanding the diversity and complexity of human-animal interactions is crucial to examining human adaptation to specific environmental conditions and for exploring the development of ancient societies. The latter half of the second millennium BC witnessed the rise of the Shang Dynasty in northern China, marking the beginning of an era characterized by various economic, cultural, and political innovations. While previous zooarchaeological research has shed light on the significant roles that animals played in Shang society, our understanding of contemporary communities outside the Shang territories remains limited. This dissertation investigates the exploitation of animal resources in Bronze Age northwestern China during the latter half of the second millennium BC, with a specific focus on the Nuomuhong Culture located in the high-altitude Qaidam Basin of the northeastern Qinghai-Xizang Plateau. Ongoing excavations at the large site of Xia’eryamakebu in present-day central Qinghai Province, have yielded a significant quantity of animal remains from both burial and non-burial contexts. Through the analysis of this extensive collection, this dissertation seeks to understand the animal-related subsistence strategies employed by the Nuomuhong people and the ritualistic mortuary practices unique to their society, which differ from those observed in other contemporary cultures. The wealth of zooarchaeological data from Xia’eryamakebu not only facilitates cross-cultural and cross-regional comparative research on animal use in Bronze Age China but also allows for a better understanding of the intricate interplay between subsistence practices and ritual traditions in ancient societies