22484 research outputs found
Sort by
Spurious Numerical Dissipation and Time Accuracy
No full textIn this thesis, we do the numerical analysis of an advection-diffusion-reaction problem in bioseparation and a corrected Smagorinsky model for turbulence. Numerical dissipation due to time discretization schemes often contributes to or causes overdissipation. The goal of this dissertation is to control spurious numerical dissipation and to acquire long-time high-order accuracy.
In the first project, we analyze an advection-diffusion-reaction problem with the non-homogeneous boundary conditions (useful for practical settings) that model the chromatography process, a vital stage in bioseparation. We prove stability and error estimates using finite elements for spatial discretization and the midpoint method for time discretization. These yield a second-order convergence rate and better total mass conservation. The numerical tests validate the theoretical results.
In the second project, we develop a turbulence model named the Corrected Smagorinsky Model (CSM) and analyze it. When the ratio of dissipation of turbulent kinetic energy (TKE) and the production of TKE is equal to , we call it statistical equilibrium. We extend a classical model for turbulence at statistical equilibrium to non-equilibrium turbulence and propose and analyze algorithms for the solution of the extended model. The classical Smagorinsky model's solution is an approximation to a (resolved) mean velocity. Since it is an eddy viscosity model, it cannot represent a flow of energy from unresolved fluctuations to the (resolved) mean velocity. This classical Smagorinsky model was corrected to incorporate this flow and still be well-posed. The computational experiments verify the properties of the algorithms and show that the model captures the non-equilibrium effects.
In the third project, we analyze the one-leg, two-step variable time step methods of Dahlquist, Liniger, and Nevanlinna (DLN) for the time discretization in the Corrected Smagorinsky Model. Turbulent flows strain computational resources in terms of memory usage and CPU (central processing unit) speed. The adaptive DLN methods are second-order accurate and allow large timesteps, hence requiring less memory and fewer FLOPS (floating point operations per second). We demonstrated the method's second-order accuracy, quantified its numerical dissipation, demonstrated error estimates in addition to proving the kinetic energy is bounded for various time steps and illustrated theoretical results by numerical tests
Neuromorphic Decoders: Paving the way towards adaptive, low-power and low-latency Brain Computer Interfaces.
No full textIntracortical Brain-Computer Interfaces (iBCIs) intercept neuronal signals, allowing paralyzed individuals to perform movements and regain daily function. However, these advancements are mostly confined to laboratory settings due to high power consumption and bandwidth requirements for communication and computation in decoders, limiting their portability.
Decoders are often trained offline, requiring significant memory to store neural recordings, and are typically implemented on standard hardware architectures, which increases latency and power consumption. This research aims to develop mobile BCIs with low-power, low-latency decoders that can be used outside labs or hospitals.
Neuromorphic decoders present a promising solution by addressing power and latency constraints. These architectures process spiking data from iBCIs directly, eliminating the need for binning or spike counting and reducing latency and power consumption. Hierarchy of Time-Surfaces (HOTS) is particularly promising for BCI applications, using clustering to analyze data patterns, potentially making HOTS more interpretable than other machine-learning techniques. Online clustering may offer solutions to continual and incremental learning challenges, allowing BCIs to adapt to shifts in neural activity and new tasks without recalibrating or retraining decoders.
However, HOTS presents some challenges: It requires exponential decay kernels that are difficult to implement efficiently on digital hardware, and it shows lower accuracy compared to backpropagation-based spiking neural networks. Additionally, HOTS's current learning rule does not support continual and incremental learning.
This thesis addresses these issues. For hardware, it explores using electrochemical (ECRAM) memristor dynamics to implement exponential decay in HOTS decoders, potentially reducing circuit complexity and energy consumption. For software, it proposes Sup3r, a learning algorithm that improves accuracy and skips uninformative events, enhancing efficiency and stability in online learning. Sup3r demonstrates continual and incremental learning, making it a fundamental advancement for HOTS models, applicable beyond BCI to various fields.
The hope is that these combined solutions will pave the way for low-power, adaptive neuromorphic decoders, enabling patients to regain autonomy outside the laboratory setting
Essays in Political Economy
No full textThis dissertation consists of three essays that contribute to the fields of political economy and public economics. Chapter 1 studies the effect of Virginia’s 2014 voter ID law on political participation rates. I gathered data on registered voters who lack DMV records and constructed a novel dataset on election precinct changes over time to identify areas of the state where more people are likely to be impacted by the voter ID law. I find a significant and durable decline in voter turnout in these areas that is driven by a decline in the number of individuals who register to vote. I also consider the role of countermobilization against new voting restrictions and find evidence that more Democratic parts of the state saw smaller effects from the voter ID law. Chapter 2 studies the impact of the 2013 Supreme Court decision, Shelby County v. Holder, on ballot access for Black and Hispanic voters. We use a rich dataset on voter behavior for the universe of registered voters combined with Census block-level sociodemographic attributes to document a decrease in turnout for Black, relative to white, individuals. These effects are concentrated in counties with larger Black and Hispanic populations, consistent with strategic targeting of voter suppression. In Chapter 3, I further study the impact of the Shelby County v. Holder decision on state public finance. I document that counties that were previously covered by federal oversight under the VRA have experienced a decrease in intergovernmental transfers from state governments. Using a triple-difference design, I then show that this decrease in transfers is concentrated among cities with larger Black populations
From Variance-Reduced Initialization to Knowledge Distillation-Inspired Pruning at Initialization: Embedding Efficiency Right from the Onset of Neural Network Training
No full textThe metaphor of Artificial Intelligence (AI) as “the new electricity” aptly describes its evolution into a ubiquitous tool, but this progress has come at a steep price due to the
increasing complexity of Deep Neural Network (DNN) architectures, which present formidable training challenges. This dissertation explores solutions to address some of the training associated primary challenges and embeds efficiency right from the outset of training. As a remedy to the exploding and vanishing gradient problem (EVGP), and a highly irregular optimization landscape hindering learning, the study introduces a universally applicable Variance-Reduced initialization technique that initializes weights as Gaussian random matrices, with parameters of the distribution derived using a Gaussian integral. Subsequently, the weight matrices are “Variance-Reduced” through a carefully designed process dependent on the network architecture. Theoretically, we demonstrate that our technique positions initial parameters closer to the optimum and facilitates faster convergence. Experimentally, we showcase that our approach offers better generalization, promotes a more stable learning process, and substantiates superior test performance. Furthermore, this thesis addresses overparameterization, yet another challenge, presenting a paradigm shift in pruning-at-initialization
with the Knowledge Distillation-based Lottery Ticket Search (KD-LTS). This framework efficiently extracts heterogeneous information from an ensemble of teacher networks. By employing a series of deterministic relaxations to address a Mixed Integer Optimization problem for training binary masks, the technique transfers the distilled information into a dense, randomly initialized student network, thus facilitating the identification of subnetworks at initialization. This work is believed to be the first in the literature to achieve state-of-the-art results across a Pareto optimal boundary, including sparsity, test performance, and computational complexity in identifying subnetworks at initialization
American Cinema’s Prison Industrial Complex: Carceral and Anti-Carceral Moving Image Media in the 20th Century
No full textThis dissertation examines the carceral and anti-carceral uses of cinema and moving image media in the American 20th century. The carceral managerial class, comprised of various bureaucrats and officiants within the prison industries, asserted influence over the American film industry’s depiction of prisons. Their arrangements with Hollywood, though fractured and multifaceted, allowed enterprising individuals within the studio system to extract great exploitation value from carceral images, architecture, and labor. Moreover, these penal managers sought out opportunities to imagine, produce, and exhibit their own cinematic visions in the form of industrial, training, educational, and publicity films. Yet moving images also proved to be powerful tools for independent and often radical filmmakers who objected to the encroachment of the prison into public life, particularly Black and of color life. Fictional narrative, experimental, and documentary modes of independent moving image production by those whom the carceral state threatened reveal that cinema can be a weapon for or against the carceral state. Prison officials and anti-prison filmmakers both intuited this simultaneous carceral and anti-carceral potential of the moving image. It motivated their discourses, projects, and aesthetic approaches.
To make this argument, I first draw on archives and trade journals from the cinema and prison industries by branching outward from their points of imbrication: movie reviews in the American Journal of Correction, prison newspapers tucked away in studio files, letters from corrections officials to studio executives, film promotional materials in the scrapbooks of wardens, and other points of contact between a spectacularly visible industry and a practically invisible one.
v
I then turn to the more diffuse and lived archives of independent film practice, drawing on close readings, oral histories, abolitionist theory, and digital humanistic mapping practices to connect carceral conditions to cinema production. Intervening in scholarship on media and governance, Black media, and the carceral state, “American’ Cinema’s Prison Industrial Complex” not only reveals the connective tissues between prison and screen, it develops a historical and theoretical heuristic for understanding the power of moving images to both produce and reimagine our carceral world
A Multi-Analytic Examination of Race, Sex, and Cigarette Craving among Smokers
No full textThe current research investigates the relationships among race, sex, and in vivo cigarette craving, aiming to elucidate potential differences across various sociodemographic groups. Leveraging the largest sample size to date in studies utilizing in vivo cigarette cue exposure, this research employed two distinct statistical approaches to assess craving dynamics. Findings revealed that Black female and male smokers exhibited heightened baseline urge ratings and were more prone to experiencing maximal peak-provoked craving during cue exposure compared to White female and male smokers. Additionally, White female smokers demonstrated heightened cue reactivity relative to White male smokers. Methodologically, this study provides compelling evidence for a comprehensive statistical approach to understanding experimentally induced cigarette craving, highlighting the importance of considering intersecting identity factors relationship with varied craving outcome measures. Clinically, these results underscore the necessity of understanding and addressing racial and sex disparities in smoking behavior and health outcomes
From sensors to stories: Enabling community-driven actionable data collection for air quality advocacy
No full textEnvironmental justice communities often face resource constraints and high levels of environmental pollution. They have limited opportunities to document and collectively understand their environment and may lack the tools and resources needed to advocate for healthier living conditions. My dissertation addresses the problem of producing reliable, accessible multi-modal data and sustainably maintaining socio-technical infrastructure for community-driven air quality monitoring in environmental justice areas. I collaborate with local community organizations to implement citizen science environmental monitoring projects in Pittsburgh. My approach integrates a low-cost physical distributed sensor network with a social network of community scientists to ensure long-term data collection, avenues to share context that isn’t captured by sensors, and a way for residents to access and interpret this data. Residents interact with the physical sensor network to help overcome the technical challenges that come with the devices, and sensing devices collect long-term environmental data that is used by residents to better understand and validate their lived experiences.
I use mixed-methods to analyze the collected environmental data, participant feedback, and reflections using qualitative and quantitative methods to identify how to ensure reliable, actionable data through participatory research approaches. I use the frameworks of Engaged Scholarship and Participatory Action Research to build sustainable partnerships between institutions and residents, and to find ways of making community science data actionable through data literacy and storytelling workshops. By making residents the owners of socio-technical infrastructure, and supporting them with technical aspects (installing devices, troubleshooting, accessing data) we build community stewardship of the system. My dissertation contributes to the fields of citizen science, science communication, and socio-technical systems by developing a holistic participatory approach to infrastructure development. It demonstrates that inclusive collaborations and real-time data can empower residents to make informed decisions about their health and drive community initiatives to address environmental concerns. This work advances environmental justice by providing a platform for data-driven solutions to air pollution challenges, fostering equitable and impactful change through collaborative sensing initiatives
Equitable Admissions in a Physician Assistant Program: Redefining Merit for the 21st Century Medical Provider
No full textIncreasing the diversity of the physician assistant (PA) profession can mitigate the effects of systemic health inequities, but PA programs face challenges in recruiting and matriculating cohorts with social identity diversity. As the Program Director for the University of Pittsburgh Physician Assistant Studies Residential Program, I aim to “matriculate a class, who will, as [physician assistants], address the diverse healthcare workforce needs of the United States” (Fancher et al., 2022, 21:14). This inquiry examined the effects of removing a higher-level chemistry course and a higher-level psychology course as program requirements through the following questions: (1) To what extent did dropping the higher-level chemistry and psychology courses change the social identity diversity in the application pool? (2) To what extent did dropping the higher-level chemistry and psychology courses affect the science and overall GPA of applicants who identify with one or more historically URiM groups? (3) To what extent did dropping the higher-level chemistry and psychology courses influence applicants who identify with one or more URiM groups in their decision to apply to the University of Pittsburgh Physician Assistant Studies Residential Program?
I employed a pragmatic parallel mixed methods design to explore my inquiry questions using pre-existing quantitative data from the Centralized Application for Physician Assistants (CASPA) and qualitative data from semi-structured interviews of current physician assistant students. The key findings show that removing these prerequisite courses did not significantly increase the social identity diversity of the applicants, nor the prerequisite science grade point
average pre- and post-intervention, highlight that the path to PA school is challenging and expensive, and identify that this PA program’s lower threshold for eligibility to apply and program goals that center diversity, equity, and inclusion are key factors influencing URiM students’ decision to apply. This inquiry argues that higher-level chemistry prerequisites are harming URiM prospective PA students without significant academic benefit, that removing higher-level chemistry courses will likely only achieve positive outcomes years after their removal, and that a combination of removing barriers, like prerequisites, and creating invitations, like diversity, equity, and inclusion programming, work synergistically to increase the number of URiM applicants
The Divided Selves of America: Examining the Psychological Underpinnings of Inconsistent Self-Perceptions Among American Partisans
No full textDrawing from philosophical perspectives and social psychological theories, this work explores self-knowledge and questions a profound ambiguity: How well do we know ourselves? Contemporary research on U.S. political conflict typically investigates how inaccurate perceptions of outgroup partisans aggravate interparty hostility. However, understudied is the role of inaccurate self-views—how people believe one thing about themselves but behave differently. This dissertation examines inconsistencies in self-perceptions within contemporary American sociopolitical discourse. Data were collected on Prolific from 2020 to 2024, spanning two U.S. presidencies. In Study 1, I developed and validated the Relativist-Absolutist Mindset (RAM) scale, which measures self-reported relativism and absolutism. The RAM scale demonstrated strong psychometric properties, including discriminant validity, predictive validity, and internal consistency. This study found that while Democrats and Republicans exhibited self-view consistency in apolitical contexts, significant inconsistencies emerged in political contexts. Democrats were more hostile than expected, whereas Republicans were less hostile. Study 2 replicated these findings using additional political and apolitical topics. Study 3 extended the findings by exploring additional political topics and examining identity relevance and moral convictions as potential mechanisms. Across various policy issues, moral convictions consistently mediated the relationship between party affiliation and self-view inconsistency. Democrats, reporting higher moral convictions, exhibited greater hostile inconsistencies in reactions to disagreement compared to Republicans. The findings reveal that while Democrats and Republicans are consistent in their self-views on apolitical topics, inconsistencies occur in political contexts, making the former more hostile in disagreements. Mechanistically, moral convictions, rather than identity relevance, appeared to drive these inconsistencies. Increased hostility among Democrats may be partly due to the current U.S. political climate. Events like gun violence, threats to reproductive rights, climate change, and attacks on LGBTQ+ rights are more threatening for Democrats, while Republicans may see a return to traditionalism as a gain. This research highlights the importance of understanding partisan blind spots in self-views. Practical implications include informing interventions to reduce polarization. This dissertation enhances understanding of self-knowledge and political discourse by showing how self-view inconsistencies influence partisan reactions, highlighting the role of moral convictions in shaping political attitudes and behaviors
The Art of Transformation: Performance Pedagogy, Embodied Cognition, and Metamorphosis as Method
No full textDecades of converging evidence from Neuroimaging, Behavioral Studies, and Cognitive Science point towards the embodied, extended, embedded, and enactive affordances of cognition, challenging traditional models of Mind/Body dualism. In the context of performance, studies from Embodied Cognition (EC) suggest that kinesthetic alterity induces transposed states of being, availing actors to dramatic possibilities otherwise not apparent when rendered through their daily neurophysiological selves. Despite these advancements, BA and BFA curricula in the United States demonstrate stubbornly persistent Cartesian divides, segregated along classical fault lines at levels of institutional conceptualization, categorization, and instruction. Taking scholars Rick Kemp and Vladimir Mirodan’s definition of “transformative acting” as a recursive process of dynamic self-repatterning that stimulates cognitive, perceptual, and affective modulation, this thesis seeks to investigate how a sustained conversation between EC and Performance Pedagogy may enrich and augment actor training via integrated somatic techniques, as undertaken by the creation of a 15-week Liberal Arts undergraduate BA course entitled “The Art of Transformation” (AOT) at the University of Pittsburgh. Drawing from Philosophy, Neuroscience, EC, Phenomenology, Gesture Studies, Affect Theory, Linguistics, Metaphor Theory, and Cultural Criticism, I identify key concepts and takeaways for “transformative acting,” and demonstrate how these principles are embedded in the praxis of Rudolph Laban, Jacques Lecoq, Michael Chekhov, and Richard Schechner’s Rasabox exercises. Weaving these practitioners into a semester-long learning scaffold, I outline the implementation, progression, and learnings from the course progression, concluding with suggestions for future iterations of AOT towards more integrative curricula. By inviting the student performer to explore diversionary and dynamic self-patterns as aesthetic epistemology, as well as investigating how embodied research and its schematization may be variously applied to devising, monologues, and partnered scenes, AOT endeavors to equip students with a tacit archive of kinesthetic schema, enhanced neurosomatic competencies, and procedural know-how to enact their own feats of transformation, onstage and off