In this thesis an architecture, similar to subsumption architectures, is presented which uses low level behaviour modules, based on combinations of machine learning techniques, to create teams of autonomous agents cooperating via shared plans for interaction. The purpose of this is to perform effective single plan execution within multiple scenarios, using a modern team based first person shooter video game as the domain and visualiser. The main focus is showing that through basic machine learning mechanisms, applied in a multi-agent setting on sparse data, plans can be executed on game levels of varying size and shape without sacrificing team goals. It is also shown how different team members can perform locally sub-optimal operations which contribute to a globally better strategy by adding exploration data to the machine learning mechanisms. This contributes to the reinforcement learning problem of exploration versus exploitation, from a multi-agent perspective.

This case study examined the spread of ideas in a Gr. 5/6 classroom in which the teacher was attempting to foster a knowledge building community. The goal of the research was to explore the relationship between the social network of the classroom (in terms of face-to-face and computer-mediated interactions), the teacher’s role, and the spread of ideas. Further, the thesis examined how social network tools may help teachers better understand the pedagogical implications of Scardamalia and Bereiter’s (1991) Teacher A, B, C models. Analyses of videotaped lessons revealed the teacher used a complex mix of traditional instructional methods and knowledge building strategies while trying to shift the locus of control of learning to students. Critical teacher-driven processes included the class-wide adoption of knowledge building vocabulary and practices, and efforts to foster higher levels of student-student discourse. Analyses of online interactions provided strong evidence of highly interconnected student-student online networks, with the note reading network being especially dense. Longitudinal studies revealed these network established themselves early in the unit, and persisted during the course of the inquiry. There was evidence that idea improvement was present in addition to idea spread. In face-to-face classroom communication, the teacher’s role was more central, particularly in "Knowledge Building Talk" sessions. However, here too, the teacher made efforts to shift the locus of control. Overall the analyses suggest that social network tools are potentially useful for helping teachers make the difficult transition from "Teacher A" and "Teacher B" strategies, in which the locus of control is with the teacher, to "Teacher C" strategies, in which strategic cognitive processes are turned over to students. This dissertation proposes that movement toward Teacher C practices may be illustrated, in part, by a shift in classroom network topologies from that of a star-shaped network, centered on the teacher, to a highly interconnected student-student network. Finally, the thesis recounts a number of ways in which the use of social network tools uncovered discourse patterns of which the teacher was unaware, including gender differences in reading, building-on, and contribution patterns.

To date, the self-revealing interface has been the elusive holy grail of the user interface community. This research advocates the use of early language learning as a model for early user interface learning. This model can be used to reason about how users learn through exploration, and gain ideas as to how to design the implicit, online help needed to make a user interface self-revealing. The idea for this model came from a strong analogy between user interfaces and language. This analogy is based on fundamental similarities, and strengthened both by observations in a case study, and the general user interface literature. A case study of early exploratory user interface learning was done in the hopes of finding similarities between the learning of languages and interfaces. Although the study did reveal many similarities, which support the model, what was most interesting was their differences. Most notably, motherese, an important form of supportive feedback that is universally present in language learning, was missing in the user interface learning. Motherese is a distinct speech variant that is used by experienced language users in conversing with children. It helps to guide children towards an understanding of correct behaviours through acknowledgment, repetition, and correction of their utterances. An experiment was devised to evaluate an analogous type of instruction in the bootstrap learning of a novel user interface technique. The experiment validated the instruction's ability to shorten the initial learning period and ingrain new techniques better than un-aided exploratory learning. Motherese-style instruction meets the requirements for instruction that is self-revealing, and is firmly grounded by the strong analogy between language and user interfaces. The application of it to user interface learning is online and integrated within the actual context of the application. It is also demonstrative and non-verbal, giving users implicit instruction, and therefore does not suffer from the terminology or contextual switching issues that written instruction does. <br /><br /> Although a number of questions remain to be answered about the general applicability of motherese-inspired user interface instruction, the model presented has yielded the first empirically-based idea for designing self-revealing instruction. It is anticipated that future research using this model will help researchers to reason about both self-revealing instruction and new user behaviour.

This research was a triangulated study of student attitudes towards instructors' rapport-building traits and their preferences amongst instructors' rapport-building practices in online learning environments. Participants were undergraduate and graduate students enrolled in courses within an educational technology program at a central Texas university. The study employed a mixed-methods approach involving the Likert-item assessment of learners' attitudes, the identification and prioritization of learner preferences through pairwise comparisons, and semi-structured interviews that provided richer, more detailed information. Findings indicated a strong preference for instructor-based traits and practices over pedagogically-based ones. These traits and practices loaded into the components of social presence, enjoyable interaction, and personal connection.

Higher education is undergoing rapid changes brought about by the ongoing financial crisis, globalization, and the rapid advancement of information technology. This scholarly personal narrative will apply assemblage theory and system dynamics to analyze the financial, cultural, and political constraints hampering change processes at traditional institutions of higher learning. Using this analysis as a starting point, the author will describe an open learning organization that addresses these issues, and how these principles have been applied to create Oplerno, LLC.–a new kind of higher educational institution.

Doctor of Philosophy / Department of Curriculum and Instruction / Dean A. Zollman / The work described here represents an effort to design, construct, and test an interactive online multimedia learning environment that can provide physics instruction to students in their homes. The system was designed with one-on-one human tutoring in mind as the mode of instruction. The system uses an original combination of a video-based tutor that incorporates natural language processing video-centered lessons and additional illustrative multimedia. Our Synthetic Interview (SI) tutor provides pre-recorded video answers from expert physics instructors in response to students’ typed natural language questions. Our lessons cover Newton’s laws and provide a context for the tutoring interaction to occur, connect physics ideas to real-world behavior of mechanical systems, and allow for quantitative testing of physics. Additional multimedia can be used to supplement the SI tutors’ explanations and illustrate the physics of interest. The system is targeted at students of algebra-based and concept-based physics at the college and high school level. The system logs queries to the SI tutor, responses to lesson questions and several other interactions with the system, tagging those interactions with a username and timestamp. We have provided several groups of students with access to our system under several different conditions ranging from the controlled conditions of our interview facility to the naturalistic conditions of use at home. In total nearly two-hundred students have accessed the system. To gain insight into the ways students might use the system and understand the utility of its various components we analyzed qualitative interview data collected with 22 algebra-based physics students who worked with our system in our interview facility. We also performed a descriptive analysis of data from the system’s log of user interactions. Finally we explored the use of machine learning to explore the possibility of using automated assessment to augment the interactive capabilities of the system as well as to identify productive and unproductive use patterns. This work establishes a proof-of-concept level demonstration of the feasibility of deploying this type of system. The impact of this work and the possibility of future research efforts are discussed in the context of Internet technologies that are changing rapidly.

Approximately 50% of doctoral students in social science, humanities, and educational doctoral programs fail to earn the Ph.D. This number is 10% to 15% higher for students enrolled in online or limited-residency programs. Using in-depth interviewing and qualitative data analysis techniques, this study examined participants' recollections of their experience as students in a limited-residency doctoral program and their reasons for withdrawal. The study addresses the following question "What is the nature of the participants' experiences of doctoral attrition in a limited-residency doctoral program?" The use of a grounded theory analysis helped identify obstacles that ultimately cause students to withdraw from limited-residency programs. The elucidation of these barriers led to the development of a theoretical model comprised of three components; each clarified relationships between attrition and a support issue (i.e., advisor support, dissertation support and program support). These components were then combined into a single theoretical model that identified the nature of participants' experience of attrition. The theoretical model helps identify steps faculty and administration could take in order to reduce attrition. The study's findings are presented in a discussion of themes found throughout the participant's narratives. Recommendations for effective doctoral education practices from existing literature are supported in the findings of this study. The limited-residency doctoral program may consider offering several forms of support to improve doctoral retention. Additionally, the program should give close attention to the relationship between the advisor and the student. Recommendations were made regarding significant program factors, accountability measures for dissertation committees and chairperson, improved monitoring of attrition, and improving the overall communication with the dissertation students. The concluding chapter includes implications of the findings and recommendations for further research regarding doctoral student attrition.

In this thesis, learning refers to the intelligent computational extraction of knowledge from data. Supervised learning tasks require data to be annotated with labels, whereas for unsupervised learning, data is not labelled. Semi-supervised learning deals with data sets that are partially labelled. A major issue with supervised and semi-supervised learning of data streams is late-arriving or missing class labels. Assuming that correctly labelled data will always be available and timely is often unfeasible, and, as such, supervised methods are not directly applicable in the real world. Therefore, real-world problems usually require the use of semi-supervised or unsupervised learning techniques. For instance, when considering a spam detection task, it is not reasonable to assume that all spam will be identified (correctly labelled) prior to learning. Additionally, in semi-supervised learning, "the instances having the highest [predictive] confidence are not necessarily the most useful ones" [41]. We investigate how self-training performs without its selective heuristic in a streaming setting. This leads us to our contributions. We extend an existing concept drift detector to operate without any labelled data, by using a sliding window of our ensemble's prediction confidence, instead of a boolean indicating whether the ensemble's predictions are correct. We also extend selective self-training, a semi-supervised learning method, by using all predictions, and not only those with high predictive confidence. Finally, we introduce a novel windowing type for ensembles, as sliding windows are very time consuming and regular tumbling windows are not a suitable replacement. Our windowing technique can be considered a hybrid of the two: we train each sub-classifier in the ensemble with tumbling windows, but delay training in such a way that only one sub-classifier can update its model per iteration. We found, through statistical significance tests, that our framework is (roughly 160 times) faster than current state of the art techniques, and achieves comparable predictive accuracy. That being said, more research is needed to further reduce the quantity of labelled data used for training, while also increasing its predictive accuracy.

abstract: This study aims to gain an understanding of higher education interventions taking place in refugee camps around the world that implement hybrid online and on-site models. Through an archival, database study, this uncovers the most salient characteristics of 8 international interventions (Australian Catholic University, Borderless Higher Education for Refugees, Jesuit Worldwide Learning: Higher Education at the Margins, InZone, Kepler, Mosaik, Global Border Studies, and Education for Humanity) in regard to logistics, academics, technology, and pedagogy. The study found multiple ways in which these programs seek to increase inclusion and success of refugee learners. These techniques include (1) free tuition, (2) nutrition, security, and transportation accommodations, (3) gender equity provisions, (4) course accreditation, (5) preparatory courses, (6) student support and development, (7) durable solutions related to employment, (8) tailored curricula, (9) flexibility of course structure, (10) critical thinking & reflection, (11) hybrid, adaptable, and portable course delivery, (12) on-site technology support, and (13) accommodations related to electricity and internet connectivity. / Dissertation/Thesis / Masters Thesis Justice Studies 2019

Thesis advisor: Marilyn Cochran-Smith / Current institutional and technological innovations are challenging face-to-face, college- and university-based teacher preparation programs as never before. Among those innovations are two emerging phenomena: New graduate schools of education (nGSEs) and fully online teacher preparation programs. nGSEs are new independent graduate schools that are not university-based but are state-authorized and approved as institutions of higher education to prepare teachers, endorse them for initial teacher certification, and grant master’s degrees (Cochran-Smith et al., 2019). Fully online teacher preparation programs are programs that relocate teacher preparation from the physical environments of the brick-and-mortar university to the digital environments of the internet and provide prospective teachers with flexible alternatives to face-to-face pathways. While both fully online teacher preparation programs and nGSEs have garnered enthusiastic media attention and critique, there is a very limited amount of in-depth knowledge about fully online teacher preparation programs and virtually no independent research on nGSEs. This dissertation helps to address those gaps in research. The central purpose of this dissertation was to examine the intersection of fully online teacher preparation and the phenomenon of teacher preparation at nGSEs by investigating teacher preparation at TEACH-NOW Graduate School of Education, a fully online, for-profit, nGSE headquartered in Washington, D.C. and rapidly expanding as a provider of initial teacher education. Intended to be descriptive and interpretive, this qualitative case study sought to understand the phenomenon of teacher preparation at TEACH-NOW from the perspectives of its participants. Based on qualitative analysis of multiple sources of evidence, the main argument of this dissertation is that TEACH-NOW operated at the nexus of a complex tension between the push to be innovative and the pull to be legitimate. Findings suggest that TEACH-NOW skillfully navigated that tension by establishing tight coherence around three key indicators of innovation (business model, technology, program structure) and by achieving major accepted markers of credibility within the larger teacher education organizational field. This dissertation also argues that TEACH-NOW’s approach to teacher preparation necessitated that teacher candidates self-manage their program experiences in accordance with their individual needs, circumstances, and preferences. The dissertation concludes with discussion of important themes and specific research, practice, and policy implications. / Thesis (PhD) — Boston College, 2019. / Submitted to: Boston College. Lynch School of Education. / Discipline: Teacher Education, Special Education, Curriculum and Instruction.