1 - 20
Next
Online 1. The absent flesh of law : legal bodies and juridical choreographies [2023]
- Kimmel, Anna Jayne, author.
- [Stanford, California] : [Stanford University], 2023.
- Description
- Book — 1 online resource.
- Summary
-
The Absent Flesh of Law: Legal Bodies and Juridical Choreographies sutures dance studies—kinesthetic awareness, somatic memory, and performative potential—to the field of the legal humanities. This represents an epistemic shift that aims to resist the logocentric norms of knowing that reproduce colonial hierarchies. The dissertation simultaneously recuperates the agency of constituent power, especially in moments of public assembly. Specifically, it frames constitutional text as corporeal by exposing the body's aporetic disappearance: from French legal codes of the 17th century that structured race, to the contemporary elision of the freedom of assembly, to human rights discourse on bodily integrity. Through this disciplinary crossing, it presents a decolonial orientation to the legal subject, animated with all of its humanisms. Separate but entangled images of protesting ballerinas serve as the motivator for each chapter. Their internationally prolific circulation across publics enforces a cross-cultural method of comparison as the events of the image shuttle between the United States, Algeria, and France—nations linked by racial legacies and postcolonial histories. As such, this project draws upon Francophone and Arab material and furthers research in a comparative race studies. Attending to contemporary Black Lives Matter marches in the United States, demonstrations of the Hirak movement in Algeria, and pension protests in Paris—referenced in each photograph—the protesting ballerina grounds this framework to reveal themes of the disciplined body that is both compliant (to a technique) and resistant (to legal expectations). Each of these images begins well into the 21st-century yet points backward in time to more complex histories in which the body was obscured under law: 1) of confederate legacies constructed in the United States that propagate a racialized history, 2) of imperial rule in France during which ballet served state opulence and secured whiteness as property, and 3) of decolonial hope and postcolonial violence in Algeria. The final chapter opens toward international human rights discourse to call for renewed attention to bodily integrity, beyond a priori concepts of dignity that are circumscribed by a Western aesthetic tradition. Theoretically grounded in histories of the archive and the ephemerality of performance, this research draws upon interdisciplinary methods to communicate to scholars in both the legal humanities and in performance and dance studies. I supplement archival material—of police reports, juridical documents, and constitutional revision—with embodied perspectives learned from more than twenty years as a disciplined body in a dance studio. The tension of this pairing allows me to articulate what is lost when the law presumes linguistic form entirely, while the project's geopolitical triangulation reflects a commitment to postcolonial theory, francophone culture, and comparative race studies. As such, it dares to bring together otherwise disparate interlocuters, forcing a reconsideration of entrenched socio-political hierarchies of discipline.
- Also online at
-
Online 2. Abstractions for efficient and reliable serverless computing [2023]
- Li, Qian (Researcher in computer science) author.
- [Stanford, California] : [Stanford University], 2023
- Description
- Book — 1 online resource
- Summary
-
Serverless, also known as function-as-a-service (FaaS), is an increasingly important paradigm in cloud computing. Developers register functions to a managed FaaS platform to serve user requests without the need to maintain their own servers. FaaS abstracts away the complexity of managing infrastructure, offers high availability, and automatically scales. However, today's FaaS platforms are often inefficient and unreliable, leaving developers with several complex application management challenges. Specifically, there are three key challenges: (1) minimizing cost while maintaining performance under varying load, (2) providing strong fault-tolerance guarantees in the presence of failures, and (3) improving debuggability and observability for distributed ephemeral functions. In this dissertation, we describe three new abstractions and build three systems to enhance the cost-efficiency, reliability, and debuggability of FaaS applications. We focus on two important categories of FaaS applications: compute-intensive, such as image recognition services, and data-centric, such as e-commerce web services. First, we address the challenge of cost efficiency for ML inference serving, a growing category of compute-intensive tasks. In particular, we tackle the key question of how to automatically configure and manage resources and models to minimize cost while maintaining high performance under unpredictable loads. Existing platforms usually require developers to manually search through thousands of model-variants, incurring significant costs. Therefore, we propose INFaaS, an automated model-less system where developers can easily specify performance and accuracy requirements without the need to specify a specific model-variant for each query. INFaaS generates model-variants from already trained models and efficiently navigates the large trade-off space of model-variants on behalf of developers to achieve application-specific objectives. By leveraging heterogeneous compute resources and efficient resource sharing, INFaaS guarantees application requirements while minimizing costs. Second, we address the challenge of providing fault tolerance while achieving high performance for data-centric applications. Existing FaaS platforms support these applications poorly because they physically and logically separate application logic, executed in cloud functions, from data management, done in interactive transactions accessing remote databases. Physical separation harms performance, and logical separation complicates efficiently providing fault tolerance. To solve this issue, we propose Apiary, a high-performance database-integrated FaaS platform for deploying and composing fault-tolerant transactional functions. Apiary wraps a distributed database engine and uses it as a unified runtime for function execution, data management, and operational logging. By physically co-locating and logically integrating function execution and data management, Apiary delivers similar or stronger transactional guarantees as comparable systems while significantly improving performance, cost, and observability. Finally, we delve into the challenge of debugging distributed data-centric applications. These applications are hard to debug because they share data across many concurrent requests. Currently, developers need to unravel the complex interactions of thousands of concurrent events to reproduce and fix bugs. To make debugging easier, we extend the tight integration between compute and data in Apiary and explore the synergy between the way people develop and debug their database-backed applications. We propose R^3, a "time travel" tool for data-centric FaaS applications that access shared data through transactions. R^3 allows for faithful replay of past executions in a controlled environment and retroactively execution of modified code on past events, making applications easier to maintain and debug. By recording concurrency information at transaction-level granularity, R^3 enables practical time travel with minimal overhead and supports most production DBMSs. We demonstrate how R^3 simplifies debugging for real, hard-to-reproduce concurrency bugs from popular open-source web applications
- Also online at
-
Online 3. Abstractions for scaling stateful cloud applications [2023]
- Kraft, Peter (Researcher in computer science) author.
- [Stanford, California] : [Stanford University], 2023
- Description
- Book — 1 online resource
- Summary
-
As the scale of both computing and data grows, developers are increasingly building distributed stateful systems in the cloud. However, these systems are challenging to build at scale because they must provide fault tolerance and consistency for stateful computations while managing both compute and data resources. Thus, we need new high-level abstractions that hide the complexity of distributed state management from developers. This dissertation proposes three such abstractions at multiple levels of the stack of a stateful cloud application. The first part of this dissertation targets cloud application developers, proposing Apiary, a database-oriented transactional function-as-a-service (FaaS) platform for stateful cloud applications. FaaS is an increasingly popular programming model because it abstracts away resource management concerns and reduces the complexity of cloud deployment, but existing FaaS platforms struggle to efficiently or reliably serve stateful applications. Apiary solves this problem by tightly integrating function execution with data management, improving FaaS performance on stateful applications by 2-68x while providing fault tolerance and strong transactional guarantees. The second part of this dissertation targets developers of the data management systems on which stateful cloud apps depend, proposing data-parallel actors (DPA), a framework for scaling data management systems. DPA targets an increasingly important class of data management systems called query serving systems, which are characterized by data-parallel, low-latency computations and frequent bulk data updates. DPA allows developers to construct query serving systems from purely single-node components while automatically providing critical properties such as data replication, fault tolerance, and update consistency. We use DPA to build a new query serving system, a simplified data warehouse based on MonetDB, and port existing ones, such as Druid, Solr, and MongoDB, enhancing them with new features such as a novel parallelism-optimizing data placement policy that improves query tail latency by 7-64%. The third part of this dissertation targets application developers utilizing multiple data management systems, proposing Epoxy, a protocol for providing ACID transactions across diverse data stores. Such applications are increasingly common because developers often use multiple data stores to manage heterogeneous data, for example doing transaction processing in Postgres and text search in Elasticsearch while storing image data in a cloud object store like AWS S3. To provide transactional guarantees for these applications, Epoxy adapts multi-version concurrency control to a cross-data store setting. We implement Epoxy for five data stores: Postgres, Elasticsearch, MongoDB, Google Cloud Storage, and MySQL, finding it outperforms existing distributed transaction protocols like XA while providing stronger guarantees and supporting more systems
- Also online at
-
Online 4. Accelerating machine learning algorithms with adaptive sampling [2023]
- Tiwari, Mohit, author.
- [Stanford, California] : [Stanford University], 2023
- Description
- Book — 1 online resource
- Summary
-
The era of huge data necessitates highly efficient machine learning algorithms. Many common machine learning algorithms, however, rely on computationally intensive subroutines that are prohibitively expensive on large datasets. Oftentimes, existing techniques subsample the data or use other methods to improve computational efficiency, at the expense of incurring some approximation error. This thesis demonstrates that it is often sufficient, instead, to substitute computationally intensive subroutines with a special kind of randomized counterparts that results in almost no degradation in quality. The results in this thesis are based on techniques from the adaptive sampling literature. Chapter 1 begins with an introduction to a specific adaptive sampling problem: that of best-arm identification in multi-armed bandits. We first provide a formal description of the setting and the best-arm identification problem. We then present a general algorithm, called successive elimination, for solving the best-arm identification problem. The techniques developed in Chapter 1 will be applied to different problems in 2, 3, and 4. In Chapter 2, we discuss an how the k-medoids clustering problem can be reduced to a sequence of best-arm identification problems. We use this observation to present a new algorithm, based on successive elimination, that matches the prior state-of-the-art in clustering quality but reaches the same solutions much faster. Our algorithm achieves an n/logn reduction in sample complexity over prior state-of-the-art, where n is the size of the dataset, under general assumptions over the data generating distribution. In Chapter 3, we analyze the problem of training tree-based models. The majority of the training time for such models is in splitting each node of the tree, i.e., determining the feature and corresponding threshold at which to split each node. We show that the node-splitting subroutine can be reduced to a best-arm identification problem and present a state-of-the-art algorithm for training trees. Our algorithm depends only on the relative quality of each possible split, rather than explicitly depending on the size of the training dataset, and reduces the explicit dependence on dataset size n from O(n), for the most commonly-used prior algorithm, to O(1). Our algorithm applies generally to many tree-based models, such as Random Forests and XGBoost. In Chapter 4, we study the Maximum Inner Product Search problem. We observe that, as with the k-medoids and node-splitting problems, the Maximum Inner Product Search problem can be reduced to a best-arm identification problem. Armed with this observation, we present a novel algorithm for the Maximum Inner Product Search problem in high dimensions. Our algorithm reduces the explicit scaling with d, the dimensionality of the dataset, O(sqrt(d)) to O(1) under reasonable assumptions on the data. Our algorithm has several advantages: it requires no preprocessing of the data, naturally deals with the addition or removal of new datapoints, and includes a hyperparameter to trade off accuracy and efficiency. Chapter 5 concludes this thesis with a summary of its contributions and possible directions for future work
- Also online at
-
- Liu, Jingxiao, author.
- [Stanford, California] : [Stanford University], 2023
- Description
- Book — 1 online resource
- Summary
-
The objective of this research is to achieve accurate and scalable bridge health monitoring (BHM) by learning, integrating, and generalizing the monitoring models derived from drive-by vehicle vibrations. Early diagnosis of bridge damage through BHM is crucial for preventing more severe damage and collapses that could lead to significant economic and human losses. Conventional BHM approaches require installing sensors directly on bridges, which are expensive, inefficient, and difficult to scale up. To address these limitations, this research uses vehicle vibration data when the vehicle passes over the bridge to infer bridge conditions. This drive-by BHM approach builds on the intuition that the recorded vehicle vibrations carry information about the vehicle-bridge interaction (VBI) and thus can indirectly inform us of the dynamic characteristics of the bridge. Advantages of this approach include the ability for each vehicle to monitor multiple bridges economically and eliminating the need for on-site maintenance of sensors and equipment on bridges. Though the drive-by BHM approach has the above benefits, implementing it in practice presents challenges due to its indirect measurement nature. In particular, this research tackles three key challenges: 1) Complex vehicle-bridge interaction. The VBI system is a complex interaction system, making mathematical modeling difficult. The analysis of vehicle vibration data to extract the desired bridge information is challenging because the data have complex noise conditions as well as many uncertainties involved. 2) Limited temporal information. The drive-by vehicle vibration data contains limited temporal information at each coordinate on the bridge, which consequently restricts the drive-by BHM's capacity to deliver fine-grained spatiotemporal assessments of the bridge's condition. 3) Heterogeneous bridge properties. The damage diagnostic model learned from vehicle vibration data collected from one bridge is hard to generalize to other bridges because bridge properties are heterogeneous. Moreover, the multi-task nature of damage diagnosis, such as detection, localization, and quantification, exacerbates the system heterogeneity issue. To address the complex vehicle-bridge interaction challenge, this research learns the BHM model through non-linear dimensionality reduction based on the insights we gained by formulating the VBI system. Many existing physics-based formulations make assumptions (e.g., ignoring non-linear dynamic terms) to simplify the drive-by BHM problem, which is inaccurate for damage diagnosis in practice. Data-driven approaches are recently introduced, but they use black-box models, which lack physical interpretation and require lots of labeled data for model training. To this end, I first characterize the non-linear relationship between bridge damage and vehicle vibrations through a new VBI formulation. This new formulation provides us with key insights to model the vehicle vibration features in a non-linear way and consider the high-frequency interactions between the bridge and vehicle dynamics. Moreover, analyzing the high-dimensional vehicle vibration response is difficult and computationally expensive because of the curse of dimensionality. Hence, I develop an algorithm to learn the low-dimensional feature embedding, also referred to as manifold, of vehicle vibration data through a non-linear and non-convex dimensionality reduction technique called stacked autoencoders. This approach provides informative features for achieving damage estimation with limited labeled data. To address the limited temporal information challenge, this research integrates multiple sensing modalities to provide complementary information about bridge health. The approach utilizes vibrations collected from both drive-by vehicles and pre-existing telecommunication (telecom) fiber-optic cables running through the bridge. In particular, my approach uses telecom fiber-optic cables as distributed acoustic sensors to continuously collect bridge dynamic strain responses at fixed locations. In addition, drive-by vehicle vibrations capture the input loading information to the bridge with a high spatial resolution. Due to extensively installed telecom fiber cables on bridges, the telecom cable-based approach also does not require on-site sensor installation and maintenance. A physics-informed system identification method is developed to estimate the bridge's natural frequencies, strain and displacement mode shapes using telecom cable responses. This method models strain mode shapes based on parametric mode shape functions derived from theoretical bridge dynamics. Moreover, I am developing a sensor fusion approach that reconstructs the dynamic responses of the bridge by modeling the vehicle-bridge-fiber interaction system that considers the drive-by vehicle and telecommunication fiber vibrations as the system input and output, respectively. To address the heterogeneous bridge properties challenge, this research generalizes the monitoring model for one bridge to monitor other bridges through a hierarchical model transfer approach. This approach learns a new manifold (or feature space) of vehicle vibration data collected from multiple bridges so that the features transferred to such manifold are sensitive to damage and invariant across multiple bridges. Specifically, the feature is modeled through domain adversarial learning that simultaneously maximizes the damage diagnosis performance for the bridge with available labeled data while minimizing the performance of classifying which bridge (including those with and without labeled data) the data came from. Moreover, to learn multiple diagnostic tasks (including damage detection, localization, and quantification) that have distinct learning difficulties, the framework formulates a feature hierarchy that allocates more learning resources to learn tasks that are hard to learn, in order to improve learning performance with limited data. A new generalization risk bound is derived to provide the theoretical foundation and insights for developing the learning algorithm and efficient optimization strategy. This approach allows a multi-task damage diagnosis model developed using labeled data from one bridge to be used for other bridges without requiring training data labels from those bridges. Overall, this research offers a new approach that can achieve accurate and scalable BHM by learning, integrating, and generalizing monitoring models learned from drive-by vehicle vibrations. The approach enables low-cost and efficient diagnosis of bridge damage before it poses a threat to the public, which could avoid the enormous loss of human lives and property
- Also online at
-
Online 6. Achieving order with two-photon lithography : colloidal self-assembly and direct laser writing [2023]
- Doan, David, author.
- [Stanford, California] : [Stanford University], 2023
- Description
- Book — 1 online resource
- Summary
-
Structural or spatial order at the nanometer/micron regime is an avenue to improve material properties. The field of photonics and metamaterials have shown that size-effects at these regimes, in combination with purposefully designed architected structures, can enhance mechanical and optical performance. A common approach to achieve these types of ordered structures is through colloidal self-assembly or direct laser writing of 3D structures. In this work, I propose using direct laser writing to fabricate colloidal particles and to fabricate complex 3D structures that have enhanced mechanical properties. In the first part of my work, I focus on colloidal self-assembly as a method to achieve order. Due to the limited chemistries and shapes of colloids available to self-assemble, a large majority of self-assembled structures remain elusive. I propose using two-photon lithography to fabricate micron-sized particles and experimentally study the effect of shape (both concave and convex) on the final self-assembled structure. This method allows for highly monodisperse fabrication of colloidal particles which can then be imaged using optical techniques due to their micron size. I fabricate colloidal conical shapes that self-assemble under entropic conditions (depletants) and tune the degree of assembly by changing the effective driving force through size. I then use a custom machine learning framework to identify these assembled structures (columnar grains) and recover self-assembly trends in which larger particles show a higher degree of self-assembly. Building upon this work, convex particles, specifically the Archimedean truncated tetrahedron, are also fabricated using the same framework and studied under another entropic condition (hard-particle interaction). These particles assemble in a six-fold symmetry upon interaction with an interface and transition to a three-fold symmetry upon application of a potential field. Analytical and computational results show that the six-fold symmetry state is a quasi-stable state and upon additional energy input, a transition occurs to achieve the lower energy state. In the second part of my work, I use two-photon lithography in conjunction with nanoclusters to enhance the direct laser writing process and improve the mechanical properties. I fabricate lattices with micron sized features and test them mechanically. The resulting nanocomposite lattices shows high stiffness and best-of-class energy absorbance by suppressing layer by layer collapse that is commonly seen with these types of structures
- Also online at
-
Online 7. Actions speak louder than words : a series of pilot studies developing novel approaches to measuring implicit attitudes [2023]
- Woodford, Benjamin Sky, author.
- [Stanford, California] : [Stanford University], 2023.
- Description
- Book — 1 online resource.
- Summary
-
The following pages will present a review of three papers that examine the measurement of bias in cognitive processing. These studies build upon one another to propose various methodological approaches for understanding the relationships between endorsed identities, learned familiarity with the ideas we interact with each day, and the personally endorsed attitudes formed in social contexts. Historically, behaviorist theories dominated education research, proponents such as Thorndike and Skinner argued that the mind was an enigmatic "black box" and that the only reliable measure of intention was its outward behavioral manifestations. This perspective was driven in part by the belief that "seeing is believing," for much of the sighted human population. Sloganized turns of phrase like this serve as an example of a cognitive shortcut using language to create a mental bias for preferring one choice over another; that simple idea ends up being the focus of this dissertation by exploring the measurement of cognitive biases rooted in near-automatic responses to stimuli. The integration of behaviorism and cognitivism in psychological assessment is a current topic of interest for researchers. This investigation aims to bridge the gap between these two fields by improving the current approaches to measuring attitudes. Understanding of human experiences has expanded leading to various learning theories emerging to connect physical behavior and internal mental experiences. While the importance of observable behavior in understanding our internal processes cannot be denied, it is also important to consider the influence of internal experiences on observable behavior. Endpoint analysis, a term I use here to refer to the study of final decision outcomes while seeking to understand internal processes, is common in educational and psychological research. Examples of endpoint analysis include the use of brain imaging to assess the effectiveness of interventions (e.g. Iuculano et al., 2015), self-report questionnaires to measure psychological indicators (e.g. Gosling, Rentfrow & Swann, 2003), and reaction time comparisons to assess bias (e.g. Greenwald, McGhee & Schwartz, 1998). However, these approaches often only allow for limited examination of the facets that more current methods can readily detect such as hesitation, ease of cognitive activation, and increased cognitive complexity during decision-making processes. Methods using endpoint analysis approaches have numerous appropriate applications and have contributed significantly to a range of scientific inquiries. In contrast, process tracing as described by Collier (2011), involves tracking the development of a phenomenon using markers that are collected during the process leading to final outcomes. The present dissertation samples from these existing fields to present three papers that explore different approaches to understanding complex psychological and socially grounded processes, with methods that can be implemented at scale within existing learning environments. In paper one, the predictive power of self-identity and connections to math attitudes are examined. In paper two, the Implicit Association Test methodology used in paper one is adapted to develop a process tracing analogy to the original Implicit Association Test procedure. The methods-focused pilot study uses Finger Tracking: a method for collecting data about implicit cognitive processes coded by Ethan Roy in Pavlovia to allow the collection of implicit cognitive process data on touchscreen devices. In paper three, a process-tracing method called cognitive surveys are introduced to measure math mindsets, combining elements of self-report surveys and IAT categorical comparisons while collecting data in a Finger Tracking environment; a method that shows promise for repeated assessments of mindsets and other related psychological constructs of interest to researchers. Each paper builds on the last in a very specific way. Paper one uses a combination of self-report and implicit measures to show how self-identification as a "math person" is associated with both explicit math anxiety self-report data and implicit math/art IAT comparisons. Paper two examines the IAT as a tool, providing evidence and methods to interpret IAT procedures administered in the Finger Tracking environment. The evidence shows how an existing implicit social cognition metric can be adapted and extended through the use of movement tracking to infer cognitive processing stages. In the third paper, a novel procedure is introduced to bridge the gap between implicit and explicit survey methods. Each prompt responded to in paper three's data set used a bank of learning endorsement statements to understand group dispositions; derived from an existing math mindset survey and the identity statement shown in paper one to have predictive power for math-learning attitudes. The jump between IAT methodology and Finger Tracking is gleaned from paper two where the analogies to reaction time with Finger Tracking provide evidence showing how convergent metrics from tracking movement serve to bolster implicit valenced attitude data. Paper three uses the lessons from the previous two papers to propose the Cognitive Survey paradigm, a new method of measuring attitude endorsement which is accessible to the average researcher willing to incorporate new research methods. Techniques such as functional magnetic resonance imaging and electroencephalography offer valuable insights into the location and time scale processes in the brain respectively, and costs are coming down, but both are often too expensive for use in educational settings due to the infrastructure and staff requirements needed for both measurement and analysis stages. In light of the growing interest in neurological information and the prohibitive expense of many existing neurological probing methods, alternative approaches such as facial recognition, eye tracking, and body tracking have become viable options for expanding research avenues in recent years. The forthcoming investigation explores implicit cognition, first using implicit association test methodology, then a form of movement tracking called Finger Tracking, a form of movement tracking as a tool for understanding complex psychological and social processes. Finger tracking is a method that utilizes an assumption of a brain-body connection to examine the relationship between movement and cognitive processing. Based on the idea that changes in the direction and velocity of finger movements reflect the mental effort involved during the task. The approach builds on body tracking research adapted to touchscreens in a coding framework developed by Ethan Roy for use in Pavlovia. The recombination allows a touchscreen interface to conduct Finger Tracking experiments and record the data, which can then be analyzed using an R studio package called mousetrap (Kieslich & Henninger, 2017). One benefit of Finger Tracking is that it can provide a wealth of process data that can be analyzed using traditional statistical methods and offers opportunities to probe previously unexplored relationships connected to cognitive processes. In paper three short statements elicit responses while making fast choice selections allows probing identity-based questions in Finger Tracking studies that facilitate the collection of both implicit and explicit responses within a single sitting. The method shows promise to provide a comprehensive understanding of an individual's experiences and cultural context. The effectiveness of Finger Tracking as a tool for analyzing the brain-body connection and cognitive processing is still emerging research and has not been fully established in the psychological sciences. Further empirical evidence and analysis are needed to fully understand the potential of this approach through a better understanding of how to interpret interactions and trends provided by the numerous available data streams. A small step in that direction is undertaken here. The development of trace process methods for this inquiry aims to advance the assessment of cognitive bias in educational settings and suggests tools that can be utilized in the broader landscape of psychological research. Finger tracking, which involves data collection on touchscreen devices and allows the automation of analysis based on quantitative metrics collected in flight, has the potential to bridge the gap between cutting-edge neuroscience and behavioral measures for education research. Finger Tracking allows for the accurate inference of cognitive processing stages at the trial level of granularity, providing greater specificity than most available methods currently in popular use. The first paper included in this inquiry utilizes an Implicit Association Test to examine the relationship between mathematical bias and self-reported math anxiety. Results show how math identity is proportionally related to cognitive bias through a series of self-report and implicit math attitude instruments. The data demonstrates the known relationship between math anxiety and implicit math attitudes measures evidenced in previous research, providing data to show how important math identity is in relationship to math anxiety both implicitly and explicitly. The discussion explores the bias of the crowd model presented by Payne, Vuletich & Lundberg (2017) to explain how social norms present in classroom situations can create the cultural inputs that structure internal narratives important for the acquisition of foundational knowledge in cognitively demanding contexts. The second paper introduces a revised version of the Implicit Association Test administered on tablet computers, which allows for the derivation of new metrics using multiple cognitively relevant processing markers. This revised tool has the potential for transformative
insights into existing measures of implicit bias; via the introduction of multiple D-score analogies based on implicit metrics relevant to processing, adapted from the well-known IAT. The analysis provides evidence supporting the discriminant value of the adapted Implicit Association Test via breaking down the data by block condition to perform deeper analysis than traditionally available with reaction time methods. Additionally, the second paper does a spatio-temporal analysis to show how cognitive bias develops along each stage of responding. The third paper adapts mindset survey questions to the Finger Tracking environment and introduces the Cognitive Survey. The paradigm aims for the simultaneous measurement of implicit and explicit mathematical mindsets in 5th or 7th grade school-age children in an educational setting. This study examines the agreement and subconscious familiarity of attitudes during a fast-reaction forced-response two-choice survey paradigm. Findings suggest that Finger Tracking data can provide new measures for the assessment of math mindsets. Finger tracking evidence shows how a sample of school children who regularly attend a school focused on reinforcing growth mindset messages, have internalized a positive learning disposition towards math as evidenced by the group-level behavioral data presented in paper three. The task was administered on touch screen devices in a classroom setting and reflects context-specific valence information about attitudes since students did the exercise in their authentic learning environment. Overall the three studies presented and the various inquiries propose approaches that can be applied to traditional research programs in the search for a more complete understanding of the assessment of cognitive bias in educational settings. Further research is needed to confirm the validity and utility of the methods undertaken in these three pilot studies. That said, my claim is that the culminating evidence in paper three presenting the cognitive survey represents a step forward in the assessment of attitudes in a variety of social science fields.
- Also online at
-
- Mohamed, Amr Mohamed Sayed Ahmed, author.
- [Stanford, California] : [Stanford University], 2023.
- Description
- Book — 1 online resource.
- Summary
-
Since the first assembly of the human genome in 2003, artificial intelligence and biology have both been improving at an astonishing rate, often synergistically. High-throughput technologies have created an opportunity to re-envision biology and medicine using novel computational techniques applied to vast datasets. Already, deep learning is beginning to impact biological research due to its ability to learn arbitrarily complex relationships from large-scale data. However, these advances create new challenges: How do we resolve and incorporate model predictions within existing knowledge and paradigms? How do we enable the use of neural network models as in silico oracles to assess hypotheses and guide experiments? How do we safely deploy deep learning systems and establish trust with researchers and practitioners, who require guarantees and a rationale for decision making? This thesis attempts to address these questions in two parts. In the first part, we focus on deep learning models of transcription factor (TF) binding, which have had striking successes modeling in vivo binding at nucleotide resolution. We present AffinityDistillation which leverages neural network models to perform novel in silico marginalization experiments at large scale to extract thermodynamic affinities of TF-DNA interactions, thereby generating quantitative predictions that can be tested in follow-up in vitro experiments. In addition to providing biophysical interpretations of neural network predictions, AffinityDistillation enables the use of neural network models as in silico biophysical oracles to assess how/whether certain in vitro phenomena manifest themselves in vivo. The second part of this thesis is focused on the safe deployment of deep learning systems to ensure they are adaptable to distributional shifts, particularly label shift. Label shift refers to the phenomenon where the prior class probability p(y) changes between the training and test distributions, while the conditional probability p(x|y) stays fixed. Label shift arises in biomedical settings, where a classifier trained to predict disease given symptoms must be adapted to scenarios where the baseline prevalence of the disease is different. Here we (1) show that combining maximum likelihood with a type of calibration called bias-corrected calibration outperforms previous methods across diverse datasets and distribution shifts, (2) prove that the maximum likelihood objective is concave, and (3) introduce a principled strategy for estimating source-domain priors that improves robustness to poor calibration. Furthermore, by using calibrated probabilities as a proxy for the true class labels, we can estimate the change in any arbitrary metric due to abstentions. Leveraging this, we present a general framework for abstention that can be applied to optimize any metric of interest, that is adaptable to label shift at test time, and that works out-of-the-box with any classifier that can be calibrated. Altogether, the computational approaches developed in this thesis can be of some use in the endeavor to understand the genome and better human health.
- Also online at
-
Online 9. Adapting expansion microscopy to imaging mass spectrometry : multiplexed interrogation of pathology samples at high resolution [2023]
- Bai, Yunhao, author.
- [Stanford, California] : [Stanford University], 2023
- Description
- Book — 1 online resource
- Summary
-
The recent development of Expansion Microscopy (ExM) and related techniques physically enlarge samples for enhanced spatial resolution while largely retaining native biomolecular content and coordinates, but almost exclusive to fluorescent imaging as the readout modality, which restricts its multiplex ability, and the range of biomolecule targets that can be studied. In contrast, Imaging Mass Spectrometry (IMS) has emerged as a powerful tool for acquiring spatially multiplexed or omics information, particularly in the label-free profiling of biomolecule targets. IMS excels at investigating cellular positioning in tissues but currently faces challenges in achieving high-resolution biomolecular features due to costly instrumental modifications and limitations imposed by physics. Integrating ExM with tissue IMS technologies would enable comprehensive and multiscale studies of tissue biology. While this concept seems straightforward, it is counterintuitive since the expanded ExM hydrogel resembles a sponge filled with water, whereas most IMS methodologies typically require water-free samples for optimal resolution and to meet sample handling requirements. This dissertation proposes an ExM framework that not only enables complete removal of water from hydrogels while preserving their lateral magnification, but also preserves the biomolecules inside expanded archival clinical samples. The processed hydrogel can be seamlessly integrated into existing tissue staining protocols and IMS instrumentation with minimal modifications. By combining the strengths of ExM and IMS, this research opens up new possibilities for investigating tissue biology at multiple scales, facilitating a deeper understanding of complex biological systems. Chapter 1 provides an overview of the methodology background and concepts, along with the motivation of this thesis work. I start with the introduction of the current stage of ExM, with a inspection of two mainstream workflows of ExM, and the underlying principle of the expansion process. The chapter then delves into the current stage of IMS, encompassing both Secondary Ion Mass Spectrometer-IMS (SIMS-IMS) and Matrix Assisted Laser Desorption/Ionization-IMS (MALDI-IMS) is introduced. This comprehensive picture will set the stage for discussing the motivation behind this thesis work, focusing on the limitations that could been overcome and the possibilities that have emerged with the advent of the ExM hydrogel platform to other imaging modalities, particularly IMS. Then, an close inspection on the obstacles, the fragility and high-water content of the fully expanded ExM hydrogel is introduced. Chapter 2 provides detailed insights into the selection and optimization of protocols to combine ExM with antibody mass-tag reporters and SIMS-based multiplex imaging methods such as Multiplexed Ion Beam Imaging (MIBI) and Imaging Mass Cytometry (IMC). This chapter presents a solution to address the challenges posed by the fragility and high-water content of the fully expanded ExM hydrogel. A controlled dehydration protocol with substrate adhesion is introduced for fully expanded samples to facilitate the formation of vacuum-compatible hydrogels for IMS instruments. Distortion test, thickness measurement and expansion fold assessment are performed to characterize the hydrogel. Thus, the chapter establishes a pipeline that allows archival human tissue sections to be expanded to 3.7 times of their original size, stained with Lanthanide-conjugated antibody cocktails, completely dehydrated to be accommodated in vacuum or desiccated chambers of IMS instruments, then imaged by MIBI or IMC using the same parameters as normal tissue sections, which I termed as Expand and comPRESS hydrOgels (ExPRESSO). This pipeline has been applied to archival human lymphoid and brain tissue sections to resolve orchestrated features of tissue architecture, particularly that of the Blood-Brain Barrier (BBB). With further antibody titration, this pipeline holds promise to interrogate archival tissue section with more than 40 channels down to 100 nm resolution for deep tissue profiling. In addition, this pipeline has been extended to include nucleotide targets, allowing for genus-level differentiation of mouse gut bacteria and their interaction with host cells. Chapter 3 further explores the possibility of combining ExPRESSO and label-free IMS imaging methodology, e.g. MALDI-IMS. With ExPRESSO, the majority of proteomics and associated N-Glycans can be anchored and expanded up to 4 times of its original size, and adapted to MALDI-IMS as a readout modality. With a modified protocol that compatible with MALDI-IMS, this chapter demonstrates that N-Glycan signals can be preserved through the ExPRESSO protocol, then be profiled with MALDI-IMS with a enhanced resolution compared with non-expanded sample. These results pave the way for routine N-Glycans profiling of archival tissues at resolutions as high as the single-cell level, enabling deeper interrogation and understanding of heterogeneity of pathological states in humans
- Also online at
-
Online 10. Adaptive algorithms for data science and computational genomics [2023]
- Baharav, Tavor Zvi, author.
- [Stanford, California] : [Stanford University], 2023.
- Description
- Book — 1 online resource.
- Summary
-
Recent years have seen a sustained exponential growth in the volume of data generated within the domains of data science and biology. This dramatic pace of data generation has outstripped the growth in computational power predicted by Moore's Law, creating a computational bottleneck. Existing algorithmic techniques are computationally insufficient, and cannot scale to meet this demand. My research has focused on developing new algorithmic techniques to handle this massive scale of data, using probabilistic modeling to design algorithms that adapt to the input dataset's structure. In this thesis, I present my work on data-adaptive algorithms, highlighting their application in computational genomics. The first part of this thesis focuses on how we can utilize methods from multi-armed bandits to construct computationally adaptive methods for data-scientific problems. We show that adaptivity can be used to design algorithms that are more efficient than their brute-force counterparts, while still providing theoretical guarantees discarded by common heuristic approaches. These methods leverage the fact that practical problems have structured data that is generated by nature, so classical worst-case analyses are overly pessimistic. To this end, we develop a general framework for the use of adaptivity in a class of computational tasks arising in data-science, including in clustering and randomized numerical linear algebra. Tackling each separate problem highlights the modularity of this framework, which we show can be further improved to exploit the structure and constraints of the specific task, which sometimes yield new multi-armed bandit settings requiring the development of new bandit algorithms. The second part of this thesis focuses on the application of these ideas to computational genomics. Nowhere is the computational bottleneck of exploding data more apparent than in genomics, where the rapidly decreasing cost of sequencing has fallen orders of magnitude faster than Moore's Law. The first problem I discuss is that of estimating pairwise sequence similarity, a common first step in genome assembly, for which we develop a novel statistical measure of similarity which adapts to the overall sequence similarity in the dataset, and a computationally efficient adaptive SVD to estimate it. The second problem I highlight is that of reference-free inference, analyzing sequencing data without the use of a reference genome. For this task we develop a novel method for analyzing sequencing data without a reference genome, and a novel finite-sample valid test on contingency tables. In both of these problems, we highlight the importance of improved probabilistic modeling and show how it can be used to develop new algorithms that adapt to structure in the data and are both statistically and computationally efficient.
- Also online at
-
Online 11. Addressing discontinuous root-finding for subsequent differentiability in machine learning, inverse problems, and control [2023]
- Johnson, Daniel Scott, author.
- [Stanford, California] : [Stanford University], 2023
- Description
- Book — 1 online resource
- Summary
-
There are many physical processes that have inherent discontinuities in their mathematical formulations. This thesis is motivated by the specific case of collisions between two rigid or deformable bodies and the intrinsic nature of that discontinuity. The impulse response to a collision is discontinuous with the lack of any response when no collision occurs, which causes difficulties for numerical approaches that require differentiability which are typical in machine learning, inverse problems, and control. We theoretically and numerically demonstrate that the derivative of the collision time with respect to the parameters becomes infinite as one approaches the barrier separating colliding from not colliding, and use lifting to complexify the solution space so that solutions on the other side of the barrier are directly attainable as precise values. Subsequently, we mollify the barrier posed by the unbounded derivatives, so that one can tunnel back and forth in a smooth and reliable fashion facilitating the use of standard numerical approaches. Moreover, we illustrate that standard approaches fail in numerous ways mostly due to a lack of understanding of the mathematical nature of the problem (e.g. typical backpropagation utilizes many rules of differentiation, but ignores L'Hopital's rule)
- Also online at
-
Online 12. Advanced methods for photon attenuation and scatter correction for combined positron emission tomography and magnetic resonance imaging (PET/MRI) [2023]
- Anaya, Emily Alexandra, author.
- [Stanford, California] : [Stanford University], 2023.
- Description
- Book — 1 online resource.
- Summary
-
Combined positron emission tomography and magnetic resonance (PET/MR) imaging combines the molecular information from PET with the structural information from MR, which allows the regions showing biological pathways of disease from PET to be localized with respect to anatomical structures in the body from MR. In order to obtain qualitatively and quantitatively accurate PET images, it is necessary to correct PET data for photon attenuation due to both the patient tissue and hardware components in the field-of-view FOV of the PET/MR system. Unlike in combined PET/CT, this correction is challenging to perform in PET/MR because the MR signal is not directly related to the attenuation properties of the tissues. Traditional methods to correct for photon attenuation include atlas and segmentation methods, however these methods have limitations, such as requiring additional scan time or not being patient specific. This thesis presents alternative methods of attenuation correction using machine learning and traditional image processing. In the former case, scatter correction also results as a natural consequence of the machine learning methodology. Previous groups have shown promising results generating synthetic CT images from MR data for photon attenuation correction (AC) using a U-shaped convolutional neural network (U-Net) or conditional generative adversarial networks (cGANs). The synthetic CT images can be used to correct for attenuation during image reconstruction in the same manner as done for PET/CT. Other groups have shown the ability to generate CT images from uncorrected PET data. In this thesis, we build upon this work and conclude that the most accurate results for synthetic CT image generation are obtained when one incorporates both uncorrected PET data and MR data in the input to a cGAN model. Our proposed attenuation and scatter corrected (ASC) solution produced PET images with superior image quality compared to the commercially-available atlas method, with an average structural similarity index (SSIM) value of 0.941 ± 0.004 vs. 0.911 ± 0.006 and an average peak signal-to-noise ratio (PSNR) value of 47.3 ± 0.4 vs. 44.3 ± 0.3 for a model that uses image data from two MR pulse sequences as well as uncorrected PET data as input. Our second proposed method directly generates corrected ASC PET images from non-attenuation and non-scatter corrected (NASC) PET data and/or MR data, avoiding the conventional approach of generating a pseudo-CT followed by image reconstruction. The mean pixel value difference between the generated ASC PET and the gold standard ASC PET obtained from a PET/CT scan of the same patient for the four different models we trained were 1.5% ± 0.8% (MR Dixon Water input), 2% ± 1% (MR Diffusion Weighted Imaging input), 1.0% ° ± 0.8% (NASC PET input) and 0.9% ± 0.6% (MR Dixon Water, MR Dixon Fat, and NASC PET input). We also describe the development and results of three novel alternative GAN methods, including a vision transformer GAN (ViT-GAN), a shifted window GAN (Swin-GAN) and an attention-gated Pix2Pix (AG-Pix2Pix) to directly generate ASC PET images from single or multi-modality image data inputs. The average PSNR and SSIM values for the multi-modal Swin-GAN input were 39.1 ± 5.5 and 0.98 (IQR 0.98-1.00), respectively, and these values for the single-modal Swin-GAN input were 39.3 ± 5.6 and 0.99 (IQR 0.98-0.99). Lastly, we discuss a novel method to perform AC of the radio-frequency (RF) coil located in the FOV of PET/MR systems. This method involves locating markers placed on an RF coil using a setup in which several cameras are placed just outside the PET/MR system FOV. We also present synthetic results demonstrating the degradation of the PET image due to inaccurate position estimation of a flexible RF coil. This method is capable of reducing the PET signal error due to the presence of a flexible RF coil by roughly 14%.
- Also online at
-
Online 13. Advancements in RNA 2'-hydroxyl covalent modification : novel technologies and chemical insights [2023]
- Chatterjee, Sayantan, author.
- [Stanford, California] : [Stanford University], 2023.
- Description
- Book — 1 online resource.
- Summary
-
The unusual nucleophilicity of 2′-OH groups on RNA molecules has proven broadly useful for investigating and conjugating this biomolecule. Carefully designed electrophilic small molecules can selectively modify RNA 2′-OH, creating adducts that may often be difficult to generate using other techniques like solid-phase oligonucleotide synthesis or enzymatic RNA synthesis. By targeting 2′-OH groups instead of nucleobase structures, this modification strategy is sequence-independent and RNA-selective (over DNA) by design. The reagents employed in this approach are also generally sensitive to 2′-OH environment in RNA: preferentially reacting at non-base paired nucleotides over base-paired locations. The importance of this technology is demonstrated by salient applications in the investigation of RNA secondary structure inside live cells, control of RNA function, and enhancement of messenger RNA stability, among many others. Although highly useful, until recently RNA 2′-OH chemical modification technology has had several limitations. The small molecule electrophiles employed in this approach have classically been highly reactive acylating reagents, with short half-lives in water limiting their utility particularly in cellular studies. These reagents are also generally difficult to purify and store, and require >10% DMSO composition (by volume) in reactions for preparative levels of RNA conversion. The possibility of generating novel covalent linkages at RNA 2′-OH beyond just the ester and carbonate moieties created by these reagents has also never been explored. In addition, strategies to direct 2′-OH acylation towards base-paired regions of RNA, or to achieve site-selective RNA modification using a catalytic approach, have not yet been developed. Finally, the effect of chirality in small molecule acylating agents have also never been investigated. This work herein presents the development of novel technologies for 2′-OH-selective RNA chemical modification to overcome many of the current limitations, and reports discoveries of new chemical insights that may broadly inform future work on RNA chemical modification. Chapter 1 describes the synthesis and characterization of a series of acylating reagents possessing fluorescence properties. The design of these molecules aims to target and selectively modify base-paired regions of RNA through intercalative scaffolds and/or electrostatic interactions. It presents the optimization of RNA reaction conditions to enhance 2′-OH acylation yields and investigates the potential of using these molecules as innovative reporters of RNA secondary structure through reverse transcription-stop (RT-stop) analysis. The ability of these reagents to demonstrate a fluorescent light-up signal in presence of RNA is also tested. The data and results summarized in this chapter demonstrate that for the reagents tested, the presence of nucleic acid-binding scaffolds and fluorogenic motifs do not lead to selective reactivity at base-paired regions or a significant fluorescence light-up upon RNA modification. Chapter 2 presents the synthesis and characterization of a series of DNA oligonucleotides modified with nucleophilic transfer catalysts. The design of these modified oligonucleotides aims to achieve covalent modification at a specific 2′-OH in an RNA sequence, by utilizing sequence complementarity to the target RNA, without the need for superstoichiometric quantities of complementary DNA by leveraging nucleophilic catalysis. This chapter presents the exploration of RNA reaction conditions with a variety of RNA 2′-OH modifying reagents, and characterization of modification by using RT-stop analysis (via PAGE gel electrophoresis). The early data obtained for this study, presented in this chapter, indicate that all the tested combinations of catalyst-containing oligonucleotides and RNA modifying small molecule reagents do not achieve site-selective modification in a test substrate single-stranded RNA. Chapter 3 describes RNA Sulfonylation, a novel chemical technology for 2′-OH selective RNA covalent modification. It reports that many activated small-molecule sulfonyl species can exhibit extended lifetimes in water and retain 2′-OH reactivity. The data establish favorable aqueous solubility for selected reagents and successful RNA-selective reactions at stoichiometric and superstoichiometric yields, particularly for aryl sulfonyltriazole species. Sulfonyltriazoles are more stable than most prior carbon electrophiles by orders of magnitude in aqueous environments, and tolerate silica chromatography. Furthermore, an azide-substituted sulfonyltriazole reagent is described to introduce labels into RNA via click chemistry. Like acylation, sulfonylation occurs with selectivity for unpaired nucleotides over those in the duplex structure, and a sulfonate adduct causes reverse transcriptase stops, suggesting future use in RNA structure analysis. Probing of rRNA is demonstrated in human cells, indicating possible cell permeability. Chapter 4 outlines the investigation of a series of chiral acylimidazole reagents with varying steric hindrance near the electrophilic center. It describes experiments that aim to determine yields and diastereoselectivity in RNA reaction for the tested compounds. Timecourse analysis of RNA acylation for a subset of enantiomeric reagent pairs reveals new insights into preferred structural features for high reactivity and diastereoselectivity. This study establishes the importance of chirality in small molecule RNA modifying reagents, with insights that may inform future design of RNA targeting molecules. Chapter 5 introduces RNA Arylation, another novel chemical technology for 2′-OH selective RNA covalent modification. It presents the investigation of the RNA reactivity of a series of electrophilic aromatic reagents with varying structural scaffolds. It describes screening experiments that aim to determine yields and 2′-OH selectivity in RNA reaction for the tested compounds. Importantly, many of these compounds are highly water-soluble, and for the first time RNA modification reactions with >99% yields are reported in total absence of any organic co-solvent which may be highly useful for future biological applications. Many of the reported molecules are also highly stable in water, with half-lives of days, while possessing a surprising ability to modify RNA with preparative yields within hours. The facile synthesis, purification, and applications in RNA conjugation using Cu-free click chemistry are described. This study introduces RNA 2′-OH Arylation as a promising technology that overcomes the limitations of prior RNA modification techniques.
- Also online at
-
Online 14. Advancements in wireless communication : AI-driven error correcting code design and multiuser system optimization [2023]
- Liao, Yun, author.
- [Stanford, California] : [Stanford University], 2023.
- Description
- Book — 1 online resource.
- Summary
-
The rapid expansion of wireless technology in recent years has ushered in an era where an increasing number of devices are interconnected. These devices, from the Internet of Things sensors to industrial automation systems, place stringent demands on wireless networks in terms of data rates and latency. As the wireless landscape evolves towards 5G and next-generation networks, the need for efficient and reliable communication systems becomes paramount. This dissertation presents a comprehensive exploration of advancements in wireless communication design, focusing on two pivotal areas: AI-based polar code design and multi-user system optimization. The first part of the dissertation ventures into the realm of AI-based polar code design. Polar codes, known for their error correction capabilities, play a pivotal role in ensuring reliable data transmission. This part introduces two novel techniques that harness the power of reinforcement learning, transforming polar code design into maze-traversing game and graph evolution process, respectively. Both learning-based approaches successfully find polar codes that outperform the existing ones, and the graph-based method offers scalability and adaptability that breaks through bottlenecks seen in other learning-based methods for code design. The dissertation's second part turns its focus to multi-user system optimization within next-generation networks, particularly in cases where the number of users exceeds the wireless environment's capacity. This part conducts an in-depth analysis of optimization algorithms meticulously crafted to maximize system capacity. Additionally, it highlights the practical implementation of newly-developed, readily-available software solutions. In an age characterized by the intricate interplay of connected devices and users, these optimization techniques assume a central role in shaping the seamless, interconnected experiences that next-generation networks are expected to deliver. This dissertation's findings underscore the transformative potential of interdisciplinary research in wireless communication. By addressing the complex challenges posed by evolving wireless technologies, it paves the way for innovative solutions that can push network capacities to their limits.
- Also online at
-
Online 15. Advancing ambient intelligence in healthcare : granularity, efficiency, and privacy [2023]
- Luo, Zelun, author.
- [Stanford, California] : [Stanford University], 2023.
- Description
- Book — 1 online resource.
- Summary
-
Artificial Intelligence (AI) is revolutionizing the healthcare industry by providing cutting-edge solutions that can significantly enhance patient care and medical research. However, this wave of technological transformation has largely overlooked the physical environment of clinical care delivery. This dissertation illuminates the significant impact of integrating video activity recognition, powered by advanced machine learning and intelligent sensors, into ICU and residential care, demonstrating its potential to alleviate healthcare burdens and improve patient outcomes. We first detail how healthcare spaces can be equipped with ambient intelligence using smart sensors and machine learning algorithms. We then introduce three crucial technical foundations that enable this application. Firstly, we instantiate a novel task, benchmark, and model for discerning complex human activities, offering a hierarchical comprehension of actor roles, object attributes, and their relationships. Secondly, we unveil a suite of data and label-efficient algorithms tailored to address overcome the scarcity of annotated data, a challenge amplified by privacy constraints, proprietary considerations, and the substantial costs associated with domain-specific annotation. Finally, we embark on a comprehensive dialogue on ensuring trustworthy machine learning from ethical and privacy perspectives. We introduce a scalable, differentially private algorithm, specifically designed for large-scale video activity recognition tasks. To demonstrate the real-world impact of our ambient intelligence system, we conclude the dissertation by showcasing two clinical deployments - one within the hospital environment and one in daily living spaces.
- Also online at
-
Online 16. Advancing generative models for real-world applications [2023]
- Choi, Eun Young, author.
- [Stanford, California] : [Stanford University], 2023.
- Description
- Book — 1 online resource.
- Summary
-
While generative models hold thrilling potential, their limited usability presents substantial challenges for their widespread adoption in real-world applications. Specifically, existing methods tend to amplify harmful societal biases ingrained in their training data and often fail to accurately reflect subjective user specifications such as style in the generated outputs. Furthermore, a notable performance gap exists when handling data distributions with unique structures, such as periodicity, restricting their applicability beyond image and text data. This dissertation considers all such facets to help construct safe, reliable generative AI systems for practical integration and deployment. First, we present a methodological framework for tackling the challenges of bias mitigation and controllability. Building on the classical approach of density ratio estimation (DRE), we develop techniques to correct a learned model distribution such that it exhibits characteristics that are more closely aligned with an alternative target distribution of interest. Together, these contributions not only yield a new theoretical framework for DRE, but also improve performance on a diverse set of downstream tasks such as domain adaptation, data augmentation, and mutual information estimation. Next, we present two real-world applications of these methods for societal applications. We demonstrate that: (a) our reweighted generative modeling framework successfully mitigates dataset bias, and (b) more controllable models can better customize AI-generated music to individual preferences and assist the creative process. Finally, we conclude by developing new learning algorithms that incorporate domain-specific inductive biases into generative models for wireless communications, as well as for discrete data distributions.
- Also online at
-
Online 17. Advancing resource recovery following anaerobic secondary treatment of domestic wastewater [2023]
- Kim, Andrew Hyunwoo, author.
- [Stanford, California] : [Stanford University], 2023
- Description
- Book — 1 online resource
- Summary
-
Proper treatment of domestic wastewater is crucial for protecting human health and the environment. However, conventional wastewater treatment processes are often high-cost, energy-intensive, and insufficient for recovering resources. Furthermore, water infrastructure in the United States is nearing the end of its intended design lifespan, posing a key opportunity for reinvention. Anaerobic secondary treatment is a promising example of next-generation water infrastructure that prioritizes resource recovery through the production of methane energy. However, anaerobic secondary effluent requires further attention due to the presence of dissolved methane, sulfide, nitrogen, and phosphorus. This dissertation explores post-treatment of anaerobic secondary effluent to maximize resource recovery from domestic wastewater. Specifically, a life cycle assessment was performed to evaluate tradeoffs between physical/chemical processes and biological processes for dissolved methane, sulfide, nitrogen, and phosphorus removal. Additionally, a membrane-aerated biofilm reactor was tested to treat anaerobic secondary effluent with high concentrations of ammonium-nitrogen and sulfide. Lastly, the use of wastewater-derived struvite as a novel fire retardant was explored to improve the profitability of phosphorus-recovery technologies. These studies serve to direct future efforts in developing complete water treatment trains with anaerobic secondary treatment
- Also online at
-
- Tikenogullari, Oguz Ziya, author.
- [Stanford, California] : [Stanford University], 2024.
- Description
- Book — 1 online resource.
- Summary
-
Cardiovascular diseases (CVDs) are the leading cause of death globally, accounting for an estimated 17.9 million deaths in 2019. CVDs are a group of disorders that affect the heart and blood vessels, and include coronary heart disease, stroke, heart failure, and arrhythmias. Computational models of the heart offer a powerful tool to understand the mechanisms of CVDs and develop new diagnostic and therapeu- tic strategies. Finite element method (FEM) is a widely used numerical method for solving complex engineering problems, and has been successfully applied to model the heart. In this thesis, we used FEM to develop computational models of the heart to understand disease mechanisms, study model parameters to understand their ef- fects on the simulation results, and develop a novel constitutive material model to improve accuracy of the simulations. We first studied the effects of viscoelastic model parameters on the simulation results of healthy heart model, in order to understand the trade-off between increased model complexity and increased accuracy of the sim- ulations. We also developed a patient-specific computational model of a diseased pediatric heart. We used this model to study the mechanisms playing role in the progress of this disease and to evaluate possible treatment options. Finally, we devel- oped a novel constitutive material model for the heart muscle. This model is more accurate than existing models, and can better capture the complex behavior of the heart muscle. Our work demonstrates the potential of computational modeling to improve our understanding of CVDs and develop new diagnostic and therapeutic strategies. Given the scale of the CVD epidemic, it is essential to continue to invest in research on computational modeling of the heart.
Online 19. Advancing the use of InSAR measurements for groundwater management and science in California's San Joaquin Valley [2023]
- Lees, Matthew Edward, author.
- [Stanford, California] : [Stanford University], 2023.
- Description
- Book — 1 online resource.
- Summary
-
In California's San Joaquin Valley, groundwater extraction has caused widespread land subsidence during the past 15 years. Many locations have seen sinking at 30 cm/yr, but just a few cm/yr of subsidence is sufficient to cause adverse impacts. Damage has therefore been extensive, with loss of capacity in major aqueducts a particular concern. In recent years, two significant events have transformed the way subsidence is understood and managed. First is the widespread availability of interferometric synthetic aperture radar data (InSAR), enabling accurate, high-resolution monitoring of subsidence across the Valley. The second was the passing in 2014 of the Sustainable Groundwater Management Act (SGMA), putting a legal requirement on groundwater agencies in the Valley to devise and implement groundwater management plans to avoid the "significant and unreasonable" results of subsidence. The central tenet of this thesis is to use the opportunity presented by InSAR data to meet the need encapsulated by SGMA. My objective is to use InSAR data to advance our scientific understanding of subsidence, with an emphasis on the development of innovative management practices that aid the implementation of SGMA. In the first Chapter, I focus on a numerical tool for simulating subsidence known as a 1D compaction model. A 1D compaction model simulates the delayed drainage and subsequent compaction of clays in response to head declines in coarser-grained materials, and these models are the primary numerical tool used to simulate subsidence in the Valley. Our ability to make insights with existing 1D compaction models has been restricted by the models' short temporal coverage and treatment of the subsurface as a single layer. Through careful data preparation in collaboration with local partners, I was able to develop a 1D compaction model that covered a 65-year period from 1952-2017 and resolved the subsurface into three layers. The model, which was calibrated using InSAR data, enabled me to identify that the majority of subsidence originates as deep compaction, and that timescales of residual compaction are in the decades-to-centuries range, as opposed to the 1--3-year timescales suggested in some literature. I then expanded this work by applying the 1D compaction model to a new location and extending it to project subsidence until 2080. I found that current groundwater management plans will lead to very high subsidence rates, but also concluded that simplifications in 1D compaction modelling mean there are substantial limitations in our ability to make predictive models. Overall, the first Chapter of this thesis demonstrates the ability, and limitations, of 1D compaction models to improve our understanding and management approach to subsidence. In the second Chapter, I consider the link between subsidence and storage. Monitoring storage changes is essential for sustainable groundwater management. Storage changes can be quantified by considering the two main components through which they are expressed: saturation changes and deformation of aquifer materials. I quantified these components in the Tule and Kaweah subbasins of the San Joaquin Valley. To quantify the component expressed through saturation changes, I followed existing observational approaches by identifying head measurements from shallow wells and scaling by specific yield. When considering the deformation component, I noted that standard approaches in the Valley are to ignore it or approximate it with a simple linear relation to measured head. However, by considering head and deformation measurements made at extensometers, I found that assuming a linear relationship between deformation and head might provide a poor estimate. Instead, I used InSAR data to quantify the deformation component of storage changes. I showed that the two components -- saturation and deformation -- accounted for storage declines of equal magnitude over 2015-2021, suggesting that the practice of ignoring the deformation component can result in large errors when estimating storage changes in regions with subsidence. Summing the two calculated components gave a new estimate of the total storage change that captured the major trends seen in independent estimates, while better accounting for the deformation component. An additional benefit is that this method accounts for the deformation component in the unconfined aquifer. This method to quantify total storage change can be a practical and effective tool to support groundwater management. In the final Chapter, I take a big-picture view of Valley-scale subsidence. Subsidence in the Valley has occurred in two periods, 1925-1970 ("the historic period") and during a series of droughts from 2007-present, causing adverse impacts during both periods. Our subsidence record during the 2007-present period is incomplete due to a 2011-15 gap in Valley-wide observations, making it difficult to develop an appropriate management response. To meet this need, I used satellite geodesy to obtain measurements of subsidence and quantify the Valley-wide subsidence volume during the 15-year period 2007-present. I found a total subsidence volume of 14 km3, comparable to the 19 km3 that occurred during the 55-year historic period. Considering the 2007-present subsidence rate and projected future overdraft reductions, I conclude that halting Valley-wide subsidence is not a realistic goal. Instead, I recommend focusing on localities where subsidence impacts are greatest and reducing overdraft from the deeper regions of the aquifer system where subsidence originates.
- Also online at
-
Online 20. Adversity roadmap : improving psychological well-being and resilience through a values-affirmation approach [2023]
- Tay, Isabelle Qi Yin, author.
- [Stanford, California] : [Stanford University], 2023.
- Description
- Book — 1 online resource.
- Summary
-
In this dissertation, I created a psychological roadmap for navigating adversity by grounding people in their core values. Research trends and global events have underscored the enduring significance of well-being and mental health. Central to this study is the self-affirmation theory, which proposes that humans possess an intrinsic need to perceive themselves as morally and adaptively capable, even when confronted with challenges. My objective was to tailor values-affirmation-based interventions to specific contexts and target groups, with the aim of enhancing well-being and motivation-related outcomes in the general population (Study 1), elite athletes (Study 2), and college students (Study 3). Study 1 involved a three-stage longitudinal randomized experiment conducted during the initial months of the COVID-19 pandemic. The study evaluated the effectiveness of two versions of the values-affirmation intervention in mitigating pandemic-related psychological burdens, particularly the feelings of social disconnection and loneliness. Notably, the interventions proved most beneficial for men, who exhibited the sharpest decline in social belonging and mental health. In Study 2, I delved into the world of high-performance athletics. I identified two root causes of compromised psychological well-being among professional Dancesport athletes -- (1) high psychological stress, and (2) outcome-contingent self-worth. To address these challenges, I developed a novel "Resilience Training + Affirmation" (RTA) intervention. Results demonstrated promising advantages of RTA for athletes' mental health and self-efficacy. Furthermore, once the competitive stress had subsided, RTA bolstered athletes' sense of social belonging and reinforced their dance identities. The findings also indicated the RTA's potential to enhance resilience among Dancesport athletes, even in the face of unsatisfactory competition results, thereby mitigating the risks associated with outcome-dependent self-worth. Study 3 extended the insights from Study 1 to an educational context. By integrating established values-affirmation techniques with actionable plans, I aimed to assist students in establishing tangible links between their values and everyday experiences. I explore unexpected correlations between well-being outcomes and pertinent covariates, such as minority race, gender, social class, and first-generation status. My aspiration is that the discoveries outlined in this dissertation, along with future research endeavors, will pave the way for the development of finely tuned strategies capable of enhancing various facets of the human experience across diverse domains.
- Also online at
-