Search

Teenagers today are coming of age under unprecedented, pervasive networks of institutional mass surveillance technologies. Surveillance technologies spreading in youth spaces are impacting young people's prospects today and are expected to continue in the future. Additionally, today's teenagers will be tomorrow's decision-makers, tasked with managing and shaping future surveillance systems. Concerns that young people are being normalized to surveillance culture carry serious political implications for just and democratic futures with technology. How do teenagers view these systems? How can educators, researchers, and fellow learners support them to develop their own perspectives and sense of agency with these technologies? This dissertation examines how teenagers learn about surveillance technologies and opportunities for supporting their learning in line with justice-centered goals. Specifically, teenagers' views on institutional privacy and surveillance through an interview study, and learning about a specific institutional surveillance technology in "Drag vs AI" workshops through observational data analyses. "Drag vs AI" is a drag queen-led workshop created by the Algorithmic Justice League which engages teenagers in makeup use to subvert AI-powered facial recognition technologies (FRTs). We organized and studied four workshop enactments with two community partners, Out Boulder County and the Longmont Public Library. Article I, "Expectations vs reality: teenager views of institutional privacy," reveals how teenagers' complex views of institutional surveillance coexist with attitudes of surveillance realism, identifying implications for justice-centered surveillance education. The second article, "Taking Play and Tinkering Seriously in AI Education: Cases from Drag vs AI Teen Workshops," examines tinkering and play as modes of interaction with FRTs in "Drag vs AI" workshops, evidencing an alternative learning methodology termed embodied algorithmic tinkering. Finally, "Drag Pedagogy for CS Educational Work: Reflections from Drag vs AI Workshops," explores how drag pedagogical practices embodied in the design and facilitation of workshops supported playful and critical CS learning around FRTs. In my concluding chapter, I reflect on the relationship between privacy and criticism of surveillance, and examine how the limitations identified in the first study emerged in studies of "Drag vs AI" using a historical action analysis, synthesizing actionable takeaways for practitioners and possible directions for future designs and research. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]

Language is central to education, being the core medium of instruction. Researchers and practitioners have long used manual methods to analyze instructional language like classroom discourse and instructional texts, with the goal of facilitating student-centered instruction. In this dissertation, I offer three studies demonstrating how natural language processing (NLP) can expand the possibilities for student-centered educational research and practice. First, I illustrate how computational methods can help analyze textbook content. Applying lexicons, word embeddings and topic models, I study the depiction of historically marginalized groups in U.S. history textbooks. I find that Latinx individuals are rarely discussed and African Americans are described with low agency and power. The quantification of these inequities can aid researchers, teachers, publishers and other stakeholders make informed decisions when developing and choosing textbooks. Second, I demonstrate how NLP can identify teachers' student-centered talk moves in classroom discourse. I develop an unsupervised machine learning measure for teachers' uptake of student contributions, a high-leverage teaching practice that supports dialogic instruction and makes students feel heard. I extensively validate this measure by analyzing the linguistic phenomena it captures, such as repetition and elaboration, and by demonstrating its correlation with positive educational outcomes across three datasets of student-teacher interaction. Third, I demonstrate how computational research on classroom discourse can be translated into improving educational practice. Building on my uptake measure, I develop a web application that provides teachers with consistent, individualized feedback on their uptake of student contributions. In a large-scale randomized controlled trial in an online computer science course, Code in Place (n=1,136 instructors), I find that the tool improves instructors' uptake of student contributions by 27% and I present suggestive evidence that it also improves students' satisfaction with the course and assignment completion. Together, these studies demonstrate the possibilities for natural language processing methods to measure and to improve key aspects of student-centered instruction. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]

This research explores the emerging field of data science from the scientometric, curricular, and altmetric perspectives and addresses the following six research questions: 1.What are the scientometric features of the data science field? 2.What are the contributing fields to the establishment of data science? 3.What are the major research areas of the data science discipline? 4.What are the salient topics taught in the data science curriculum? 5.What topics appear in the Twitter-sphere regarding data science?6.What can be learned about data science from the scientometric, curricular, and altmetric analyses of the data collected? Using bibliometric data from the Scopus database for 1983-2021, the current study addresses the first three research questions. The fourth research question is answered with curricular data collected from U.S. educational institutions that offer data science programs. Altmetric data was gathered from Twitter for over 20 days to answer the fifth research question. All three sets of data are analyzed quantitatively and qualitatively. The scientometric portion of this study revealed a growing field, expanding beyond the borders of the United States and the United Kingdom into a more global undertaking. Computer Science and Statistics are foundational contributing fields with a host of additional fields contributing data sets for new data scientists to act, including, for example, the Biomedical and Information Science fields. When it comes to the question of salient topics across all three aspects of this research, it was revealed that a large degree of coherence between the three resulted in highlighting thirteen core topics of data science. However, it can be noted that Artificial Intelligence stood out among all the other groups with leading topics such as Machine Learning, Neural Networks, and Natural Language Processing. The findings of this study not only identify the major parameters of the data science field (e.g., leading researchers, the composition of the discipline) but also reveal its underlying intellectual structure and research fronts. They can help researchers to ascertain emerging topics and research fronts in the field. Educational programs in data science can learn from this study about how to update their curriculums and better prepare students for the rapidly growing field. Practitioners and other stakeholders of data science can also benefit from the present research to stay tuned and current in the field. Furthermore, the triple-pronged approach of this research provides a panoramic view of the data science field that no prior study has ever examined and will have a lasting impact on related investigations of an emerging discipline. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]

This dissertation conducted a science classroom intervention using two instructional approaches, computational modeling and paper-based pictorial modeling, in the context of food webs. A series of research papers were written on the impact of the intervention on students' attitudes and learning, and on teachers via professional development and teaching. The first study employed quasi-experimental design, examining the impacts of computational modeling and paper-based pictorial modeling on students' systems thinking and CT skills. A total of 365 seventh-grade students were involved in week-long online learning activities. The students were purposefully assigned to either synchronous computational modeling condition (n = 224) or asynchronous paper-based pictorial modeling condition (n = 141). Students modeled the interrelation between concepts and components of a food web in both conditions. The students took CT and systems thinking-embedded food web assessments before and after the activity and formative assessment after each activity to gather their perceptions of what they thought they learned from the activities. Multilevel modeling and Epistemic Network Analysis were used to analyze the data. The findings indicated a significant increase in students' systems thinking skills regardless of the conditions. Students in computational modeling condition had a substantial increase in their CT scores, while such a result was not detected in the paper-based pictorial modeling condition. ENA results supported this finding, showing that students in computational modeling condition expressed that they learned both science and CT concepts. The second study explored the predictors of students' interests in computationally intensive (or AI-informed) science careers and the impact of computational modeling activity on middle school students' interests in such careers. In parallel, an instrument to measure students' career interests called the Computationally Intensive Science Career Interest (CISCI), was developed and validated. A combination of classical test theory and item response theory was used to validate the CISCI instrument using a sample N = 934. Multiple linear regression and paired-sample "t"-tests were performed to analyze the data. The results showed that the CISCI is a valid and reliable instrument consisting of five constructs. Results from regression tests revealed that students' science attitudes, computer science attitudes, CT skills, and prior computer science-related activities significantly predicted students' career interests. Furthermore, a significant increase in students' perceptions of intensity of discussing computationally intensive careers with their parents was detected after participating in a computational modeling activity. The third study employed a mixed-methods design to investigate the changes in, and sources of, middle school teachers' self-efficacy for teaching science in a computationally rich environment. A total of eleven middle school teachers participated in this study. They took two questionnaires four times--before and after training and teaching--measuring their self-efficacy for teaching science and CT. The teachers were also either interviewed or asked to provide written reflections after taking the questionnaires. Non-parametric Skillings-Mack tests and thematic analysis were used to analyze the data. The results revealed a significant increase in teachers' self-efficacy for teaching science and CT after training and teaching. These changes in self-efficacy were tied to three sources, namely: (1) exposure to computer programming, (2) students' interests and responses to a computationally rich science environment, and (3) teaching repetition and field experience. Results from the three studies highlight the potential benefit of engaging students with learning science in a computationally rich environment to positively influence both their CT and systems thinking skills as well as content knowledge. The study also highlighted the challenges and possibilities for preparing science teachers to teach computational modeling with little or no background in programming or computer science. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]

This dissertation presents a novel theory of event memory along with an associated computational model that embodies the claims of view which is integrated within a cognitive architecture. Event memory is a general-purpose storage for personal past experience. Literature on event memory reveals that people can remember events by both the successful retrieval of specific representations from memory and the reconstruction of events via schematic representations. Prominent philosophical theories of event memory, i.e., causal and simulationist theories, fail to capture both capabilities because of their reliance on a single representational format. Consequently, they also struggle with accounting for the full range of human event memory phenomena. In response, we propose a novel view that remedies these issues by unifying the representational commitments of the causal and simulation theories, thus making it a hybrid theory. We also describe an associated computational implementation of the proposed theory and conduct experiments showing the remembering capabilities of our system and its coverage of event memory phenomena. Lastly, we discuss our initial efforts to integrate our implemented event memory system into a cognitive architecture, and situate a tool-building agent with this extended architecture in the Minecraft domain in preparation for future event memory research. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]

Machine learning has achieved state-of-the-art performance in many areas, including image recognition and natural language processing. However, there are still many challenges and mysteries attracting numerous researchers. This dissertation comprises a series of works concerning problems at the intersection of computer science theory, adversarial robustness, generalization theory, and social science. The first part is to understand adversarial robustness from two perspectives: the efficiency of popular defense mechanisms against adversarial attacks and to understand how adversarial training trade-offs prediction accuracy to robustness. Specifically, in the first perspective, we investigate the popular defense mechanism that formulates adversarial training as robust optimization and trains with projected gradient descent. We study the non-concave landscape of the adversarial loss of a two-layer neural network. Our main result proves that the projected gradient ascent finds a local maximum of this non-concave problem in a polynomial number of iterations with high probability. In the second perspective, we introduce the Adversarial Influence Function (AIF) as a tool to investigate the solution produced by robust optimization. The proposed AIF enjoys a closed-form and can be calculated efficiently and is useful in quantifying the trade-off between accuracy and robustness. The second part is devoted to understanding generalization in modern learning. We investigate how a popular data augmentation scheme helps generalization and also develop a new notion of stability tailored for modern machine learning. Specifically, for data augmentation, we provide theoretical analysis to demonstrate how using Mixup in training helps model robustness and generalization. For robustness, we show that minimizing the Mixup loss corresponds to approximately minimizing an upper bound of the adversarial loss. This explains why models obtained by Mixup training exhibit robustness to several kinds of adversarial attacks such as the Fast Gradient Sign Method (FGSM). For generalization, we prove that Mixup augmentation corresponds to a specific type of data-adaptive regularization that reduces overfitting. Our analysis provides new insights and a framework to understand Mixup. For a new type of stability, we propose "locally elastic stability" as a weaker and distribution-dependent stability notion, which still yields exponential generalization bounds. We further demonstrate that locally elastic stability implies tighter generalization bounds than those derived based on uniform stability in many situations by revisiting the examples of bounded support vector machines (SVM), regularized least square regressions, and stochastic gradient descent (SGD). The final part is about fairness, where we initiate the study of the construction of "scaffolding sets," a small collection "C" of sets with the property that multi-calibration with respect to "C" ensures recovering predictors of individual probabilities accurately, and not just calibration, of the predictor. Our approach is inspired by the folk wisdom that the intermediate layers of a neural net learn a highly structured and useful data representation. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]

Constraint satisfaction problems are a powerful way to abstract and represent academic and real-world problems from both artificial intelligence and operations research. A constraint satisfaction problem is typically addressed by a sequential constraint solver running on a single processor. Rather than construct a new, parallel solver, this work develops convenient parallelization methods for pre-existing solvers, and thereby benefits from both state-of-the-art research and increasingly available computational resources. This work proposes and evaluates several approaches that exploit scheduling and partitioning. It formulates parallel algorithm portfolio construction as an integer- programming problem, and solves the problem with algorithms that combine case-based reasoning with either a greedy algorithm or a complete integer-programming solver. This work also develops a hybrid adaptive paradigm that learns critical information to support search and workload splitting while it solves the problem. Extensive experiments show that these approaches improve the performance of the underlying sequential solvers. The hybrid paradigm solves many difficult problems left open after recent solver competitions. Although the empirical results are mainly on constraint satisfaction problems, this work also generalizes the reasoning component of the parallel scheduler to a new, case-based paradigm, and successfully applies it to protein-ligand docking. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]

This dissertation transforms a set of system complexity reduction problems to feature selection problems. Three systems are considered: classification based on association rules, network structure learning, and time series classification. Furthermore, two variable importance measures are proposed to reduce the feature selection bias in tree models. Associative classifiers can achieve high accuracy, but the combination of many rules is difficult to interpret. Rule condition subset selection (RCSS) methods for associative classification are considered. RCSS aims to prune the rule conditions into a subset via feature selection. The subset then can be summarized into rule-based classifiers. Experiments show that classifiers after RCSS can substantially improve the classification interpretability without loss of accuracy. An ensemble feature selection method is proposed to learn Markov blankets for either discrete or continuous networks (without linear, Gaussian assumptions). The method is compared to a Bayesian local structure learning algorithm and to alternative feature selection methods in the causal structure learning problem. Feature selection is also used to enhance the interpretability of time series classification. Existing time series classification algorithms (such as nearest-neighbor with dynamic time warping measures) are accurate but difficult to interpret. This research leverages the time-ordering of the data to extract features, and generates an effective and efficient classifier referred to as a time series forest (TSF). The computational complexity of TSF is only linear in the length of time series, and interpretable features can be extracted. These features can be further reduced, and summarized for even better interpretability. Lastly, two variable importance measures are proposed to reduce the feature selection bias in tree-based ensemble models. It is well known that bias can occur when predictor attributes have different numbers of values. Two methods are proposed to solve the bias problem. One uses an out-of-bag sampling method called OOBForest, and the other, based on the new concept of a partial permutation test, is called a pForest. Experimental results show the existing methods are not always reliable for multi-valued predictors, while the proposed methods have advantages. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]

This thesis aims to define and extend a line of computational models for text comprehension that are humanly plausible. Since natural language is human by nature, computational models of human language will always be just that--models. To the degree that they miss out on information that humans would tap into, they may be improved by considering the human process of language processing in a linguistic, psychological, and cognitive light. Approaches to constructing vectorial semantic spaces often begin with the distributional hypothesis, i.e., that words can be judged "by the company they keep." Typically, words that occur in the same documents are similar, and will have similar vectorial meaning representations. However, this does not in itself provide a way for two distinct meanings to be composed, and it ignores syntactic context. Both of these problems are solved in "Structured Vectorial Semantics" (SVS), a new framework that fully unifies vectorial semantics with syntactic parsing. Most approaches that try to combine syntactic and semantic information will either lack a cohesive semantic component or a full-fledged parser, but SVS integrates both. Thus, in the SVS framework, interpretation is interactive, considering both syntax and semantics simultaneously. Cognitively-plausible language models would also be incremental, support linear-time inference, and operate in only a bounded store of short-term memory. Each of these characteristics is supported by right-corner Hierarchical Hidden Markov Model (HHMM) parsing; therefore, SVS will be transformed into right-corner form and mapped to an HHMM parser. The resulting representation will then encode a psycholinguistically plausible incremental SVS language model. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]

The research presented in this dissertation focuses on the organizational and consumer need for knowledge based support in unstructured domains, by developing a measurement scale for dynamic interaction. Addressing this need is approached and evaluated from two different perspectives. The first approach is the development of Knowledge Based Systems (KBS) designed to operate in unstructured domains. By meshing the system attributes of KBS and Decision Support Systems (DSS), researchers and developers have begun designing KBS for unstructured domains that tract with the user's iterative decision process and allow the user to evaluate alternative solutions. The second approach to providing knowledge support in unstructured domains is through knowledge synthesization. The latest trend in Web 2.0 research focuses on mashup applications that are designed to synthesize knowledge by semantically connecting disjointed information and knowledge sources (Blake and Nowlan, 2008). The focus of this dissertation is on developing and empirically evaluating a new IS construct that can be used by researchers to quantify dynamic interaction. Dynamic interaction is empirically evaluated in two IS domains: KBS and eCommerce mashups. The nomological net of dynamic interaction is initially quantified in the KBS domain, and then is expanded from evaluating this new construct in the mashup domain. This dissertation consists of two parts and four chapters. The first part is titled "Knowledge Based System's Support of Unstructured Domains Through Dynamic Interaction" and has two chapters that cover KBS designed for unstructured domains by including dynamic interaction. The 1st chapter is titled "Knowledge Based Systems to Support Unstructured Decisions: A Literature Review" and provides a literature review of recent KBS, of which are designed for unstructured domains and include an iterative user interface. The 2nd chapter is titled "Dynamic Interaction in Knowledge Based Systems: An Exploratory Investigation and Empirical Evaluation". This study empirically investigates the KBS discussed in the previous chapter, which are designed for unstructured domains and include dynamic interaction. The second part of the dissertation is titled "Decision Support Mashups--Knowledge Synthesization Through Dynamic Interaction" and is composed of two chapters that discuss knowledge synthesization in unstructured domains through dynamic interaction in mashups. The 3rd chapter is titled "Mashups: A Literature Review and Classification Framework". It presents a mashup literature review, a mashup classification framework, and provides a literary foundation for the final chapter of the dissertation. The final and 4th chapter is titled "Decision Support Mashups in Unstructured Domains: An Empirical Evaluation of Dynamic Interaction" and performs an empirical evaluation of dynamic interaction in decision support mashups. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]

The study of this dissertation provides a solution to a very specific problem instance in the area of data mining, data warehousing, and service-oriented architecture in publishing and newspaper industries. The research question focuses on the integration of data mining and data warehousing. The research problem focuses on the development of intelligent information retrieval and web mining technique using SOA to accelerate the processing time at lower cost. The findings of the research are proved by designing and developing a prototype. The solution can be applied to resolve similar practical problems in various domains, including media stations, magazines, military, and medicine. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]

Commenting source code has long been a common practice in software development. This thesis, consisting of three pieces of work, made novel use of the code comments written in natural language to improve software reliability. Our solution combines Natural Language Processing (NLP), Machine Learning, Statistics, and Program Analysis techniques to achieve this goal. First, innovations from multiple directions have been proposed to improve software reliability. Unfortunately, many of the innovations are not fully exploited by programmers. To bridge the gap, we propose a new approach, cComment, to "listen" to thousands of programmers by studying their programming comments. Since comments express programmers' assumptions and intention, comments can reveal programmers' needs. These programmers' needs provide guidance (1) for language/tool designers on where they should develop new techniques or enhance the usability of existing ones, and (2) for programmers on what problems are most pervasive and important so that they should take initiatives to adopt some existing tools or language extensions. We studied 1050 comments randomly sampled from the latest versions of Linux, FreeBSD, and OpenSolaris at the time of writing. We found that 52.6% of these comments could be leveraged by existing or to-be-proposed tools for improving reliability. Our findings include: (1) many comments describe code relationships, code evolutions, or the usage and meaning of integers and integer macros, (2) a significant amount of comments could be expressed by existing annotation languages, and (3) many comments express synchronization related concerns but are not well supported by annotation languages. Second, compared to source code, comments are more direct, descriptive and easy-to-understand. Comments and source code provide relatively redundant and independent information regarding a program's semantic behavior. As software evolves, they can easily grow out-of-sync, indicating two problems:(1) bugs--the source code does not follow the assumptions and requirements specified by correct program comments; (2) bad comments--comments that are inconsistent with correct code, which can mislead programmers to introduce bugs in subsequent versions. Unfortunately, as most comments are written in natural language, no solution has been proposed to automatically analyze comments and detect inconsistencies between comments and source code. iComment took the first step in automatically analyzing comments written in natural language to extract implicit program rules and use these rules to automatically detect inconsistencies between comments and source code, indicating either bugs or bad comments. We evaluate iComment on four large code bases: Linux, Mozilla, Wine and Apache. Our experimental results show that iComment automatically extracts 1832 rules from comments with 90.8-100% accuracy and detects 60 comment-code inconsistencies, 33 new bugs and 27 bad comments, in the latest versions of the four programs when the study was conducted. Nineteen of them (12 bugs and 7 bad comments) have already been confirmed by the corresponding developers while the others are currently being analyzed by the developers. Lastly, we proposed and implemented aComment to detect operating system concurrency bugs and handle the complex interaction between interrupts and lock. Specifically, we designed a new type of interrupt related annotations, and semi-automatically generated 96,821 such annotations for the Linux kernel. These annotations were automatically propagated from 246 seed annotations, directly inferred from comments and code assertions. By extracting annotations from both comments and code, we are able to extract more annotations than using a single source as only a small number (6) of the annotations can be extracted from both sources. These annotations were used to check against source code to detect software bugs, and 9 bugs were detected from the latest version of the Linux kernel at the time of writing. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]