submitted to the Faculty of Education in partial fulfilment of the requirements for the degree of Master of Education in the Department of Educational Psychology of the University of Zululand, 1997. / The present study examines the relationship between metacognitive strategy instruction and academic performance among children with learning problems. The metacognitive strategy instruction was based on a metalearning model. Thirty nine pupils with learning problems from grades 4 and 5 participated in the study. Academic performance data on curriculum based history tests and data from the self-report Metacognitive Learning Process Questionnaire (MLPQ) were collected. Baseline pretest data from History Test 1 indicated that pupils attained average percentage scores. Subsequent to training in the use of cognitive and metacognitive strategies pupils produced significantly higher performance scores on History Test 2. No statistically significant differences were found between the pre- and post application of the MLPQ. Increases in the cognitive and metacognitive strategies and test writing abilities sub-tests of the MLPQ were indicated. The results clearly indicated that training in metacognitive strategies coincided with an increase in academic performance. The broader implications and limitations of the study are discussed.

Metalearning aims to obtain knowledge of the relationship between the mechanism of learning and the concrete contexts in which that mechanisms is applicable. As new mechanisms of learning are continually added to the pool of learning algorithms, the chances of encountering behavior similarity among algorithms are increased. Understanding the relationships among algorithms and the interactions between algorithms and tasks help to narrow down the space of algorithms to search for a given learning task. In addition, this process helps to disclose factors contributing to the similar behavior of different algorithms. We first study general characteristics of learning tasks and their correlation with the performance of algorithms, isolating two metafeatures whose values are fairly distinguishable between easy and hard tasks. We then devise a new metafeature that measures the difficulty of a learning task that is independent of the performance of learning algorithms on it. Building on these preliminary results, we then investigate more formally how we might measure the behavior of algorithms at a ner grained level than a simple dichotomy between easy and hard tasks. We prove that, among all many possible candidates, the Classifi er Output Difference (COD) measure is the only one possessing the properties of a metric necessary for further use in our proposed behavior-based clustering of learning algorithms. Finally, we cluster 21 algorithms based on COD and show the value of the clustering in 1) highlighting interesting behavior similarity among algorithms, which leads us to a thorough comparison of Naive Bayes and Radial Basis Function Network learning, and 2) designing more accurate algorithm selection models, by predicting clusters rather than individual algorithms.

Automatic machine learning (AutoML) systems have been shown to perform better when they use metamodels trained offline. Existing offline metalearning approaches treat ML models as black boxes. However, modern ML models often compose multiple ML algorithms into ML pipelines. We expand previous metalearning work on estimating the performance and ranking of ML models by exploiting the metadata about which ML algorithms are used in a given pipeline. We propose a dynamically assembled neural network with the potential to model arbitrary DAG structures. We compare our proposed metamodel against reasonable baselines that exploit varying amounts of pipeline metadata, including metamodels used in existing AutoML systems. We observe that metamodels that fully exploit pipeline metadata are better estimators of pipeline performance. We also find that ranking pipelines based on dataset metafeature similarity outperforms ranking based on performance estimates.

Submitted by PPG Ci?ncia da Computa??o (ppgcc@pucrs.br) on 2018-10-04T18:59:57Z No. of bitstreams: 1 LUCIANO_ ALVES_DIS.pdf: 1077045 bytes, checksum: ddcbf3be03bec1c7a82f3e07252439a0 (MD5) / Rejected by Sheila Dias (sheila.dias@pucrs.br), reason: Devolvido deviso ? inconsist?ncia de datas no arquivo pdf. Na capa institucional, na ficha catalogr?fica e na folha da banca est? 2016 e na folha de rosto 2018. on 2018-10-05T16:43:09Z (GMT) / Submitted by PPG Ci?ncia da Computa??o (ppgcc@pucrs.br) on 2018-10-08T18:31:55Z No. of bitstreams: 1 LUCIANO_ ALVES_DIS.pdf: 1076874 bytes, checksum: 70823493135f9ec1a577db83eefbd19c (MD5) / Approved for entry into archive by Caroline Xavier (caroline.xavier@pucrs.br) on 2018-10-09T16:36:57Z (GMT) No. of bitstreams: 1 LUCIANO_ ALVES_DIS.pdf: 1076874 bytes, checksum: 70823493135f9ec1a577db83eefbd19c (MD5) / Made available in DSpace on 2018-10-09T16:43:56Z (GMT). No. of bitstreams: 1 LUCIANO_ ALVES_DIS.pdf: 1076874 bytes, checksum: 70823493135f9ec1a577db83eefbd19c (MD5) Previous issue date: 2016-09-23 / Software fault prediction is a significant part of software quality assurance and it is commonly used to detect faulty software modules based on software measurement data. Several machine learning based approaches have been proposed for generating predictive models from collected data, although none has become standard given the specificities of each software project. Hence, we believe that recommending the best algorithm for each project is much more important and useful than developing a single algorithm for being used in any project. For achieving that goal, we propose in this dissertation a novel framework for recommending machine learning algorithms that is capable of automatically identifying the most suitable algorithm according to the software project that is being considered. Our solution, namely FMA-PFS, makes use of the metalearning paradigm in order to learn the best learner for a particular project. Results show that the FMA-PFS framework provides both the best single algorithm recommendation and also the best ranking recommendation for the software fault prediction problem. / A predi??o de falhas de software ? uma parte significativa da garantia de qualidade do software e ? normalmente utilizada para detectar m?dulos propensos a falhar baseados em dados coletados ap?s o processo de desenvolvimento do projeto. Diversas t?cnicas de aprendizado de m?quina t?m sido propostas para gera??o de modelos preditivos a partir da coleta dos dados, por?m nenhuma se tornou a solu??o padr?o devido as especificidades de cada projeto. Por isso, a hip?tese levantada por este trabalho ? que recomendar algoritmos de aprendizado de m?quina para cada projeto ? mais importante e ?til do que o desenvolvimento de um ?nico algoritmo de aprendizado de m?quina a ser utilizado em qualquer projeto. Para alcan?ar este objetivo, prop?e-se nesta disserta??o um framework para recomendar algoritmos de aprendizado de m?quina capaz de identificar automaticamente o algoritmo mais adequado para aquele projeto espec?fico. A solu??o, chamada FMA-PFS, faz uso da t?cnica de meta-aprendizado, a fim de aprender o melhor algoritmo para um projeto em particular. Os resultados mostram que o framework FMA-PFS recomenda tanto o melhor algoritmo, quanto o melhor ranking de algoritmos no contexto de predi??o de falhas de software.

In Lesotho, there are no formal opportunities for professional training of teacher educators. Consequently, the majority of teacher educators have not received a training that could equip them with professional knowledge base that is foundational to any profession. Therefore the question: what are the sources and application of professional knowledge among teacher educators appeared justifiable. Arguably, the teacher educators’ professional knowledge is intricately linked to education practice. Teacher educators have to address the discrepancy between education policy and practice through the training of student teachers who, in turn, have to contribute to the quality of the Lesotho education system. An interpretivist approach was followed in undertaking this study. Data was collected through: narratives, observations of teacher educators and analysis of the curriculum and assessment documents. The unit of analysis was eight teacher educators who are based at the National University of Lesotho’s Faculty of Education. Verification of the extent to which the topic was researchable was through undertaking a pilot study with six teacher educators who were based in the department of Educational Foundations in the same faculty. The analysis of the data revealed an immersion in the teacher educators’ professional landscape provides them ample opportunities to learn from an array of experiences. They accumulated experienced-based professional knowledge relevant to their world of work as they learn to teach, construct, apply and model it in the context that is uniquely teacher education. They have learned to teach teachers mainly from existing education practices which perpetuate what already exists. They face numerous challenges; their teaching is biased towards conventional teaching techniques of a transmissive nature and to a less extent interactive techniques; construction of professional knowledge remains a complex and challenging undertaking. Opportunities to construct own teaching research-based knowledge and supervision of student research are limited. In practice teacher educators have to rethink their pedagogy. Engaging in research adopting a “self-study” approach is unavoidable. Research will enhance their professional development and the quality of the student teachers. / Thesis (PhD)--University of Pretoria, 2011. / Humanities Education / unrestricted

In the last few years, as processing the data became a part of everyday life in different areas of human activity, the automated machine learning systems that are designed to help with the process of data mining, are on the rise. Various metalearning techniques, including recommendation of the right method to use, or the sequence of steps to take, and to find its optimum hyperparameters configuration, are integrated into these systems to help the researchers with the machine learning tasks. In this thesis, we proposed metalearning algorithms and techniques for hyperparameters optimization, narrowing the intervals of hyperparameters, and recommendations of a machine learning method for a never before seen dataset. We designed two AutoML machine learning systems, where these metalearning techniques are implemented. The extensive set of experiments was proposed to evaluate these algorithms, and the results are presented.

Diese Arbeit untersucht prominente Erfassungsmethoden metakognitiver Aktivitäten die während des Lernprozesses zum Einsatz kommen (online) auf deren Güte und Reaktivität. Im Fokus stehen die Methoden Laut-Denken, Fragebogenmethode sowie die Erfassung von Lernleistungsurteilen. Lernaufgaben werden durch komplexe Textlernaufgaben sowie Problemlöseaufgaben in deren Rahmen abduktive Schlüsse gefordert sind repräsentiert. In Studie 1 wurden metakognitive Aktivitäten die mittels retrospektiv eingesetzten Fragebögen sowie mittels Laut-Denken erfasst wurden gegenübergestellt. Dabei wurden die Fragebogenitems parallel zum polytomen Kategoriensystem mit dessen Hilfe die Daten aus der Laut-Denken-Methode ausgewertet wurden konstruiert. Im Rahmen der Auswertung der Laut-Denken Daten war die Übereinstimmung zweiter unabhängiger, gut geschulter Urteiler unbefriedigend. Die Übereinstimmungsunterschiede zwischen den Kategorien sowie zwischen den Probanden waren erheblich. Dies weist darauf hin, dass das Kategoriensystem nicht zur Auswertung der Laut-Denken Daten geeignet ist. Zudem scheinen große Unterschiede in der Nutzung metakognitiver Aktivitäten zwischen den Probanden zu bestehen. Zwischen Fragebogendaten und Laut-Denken-Daten besteht ein geringer nicht signifikanter negativer Zusammenhang. In Studie 2 wurde die Reaktivität der Laut-Denken-Methode und der Aufzeichnung von Lernleistungsurteilen während des Bearbeitens einer Textlese- sowie Problemlöseaufgabe untersucht. Die Ergebnisse dieser experimentellen Studie mit 2x2 Design legen nahe, dass von Laut-Denken im Rahmen von Problemlöseaufgaben reaktive Effekte zu erwarten sind. Von Lernleistungsurteilen hingegen sind reaktive Effekte lediglich im Rahmen von komplexen Textleseaufgaben zu erwarten. Auch im Rahmen dieser Erhebung mittels Laut-Denken konnte lediglich eine unbefriedigende Reliabilität der Messung berichtet werden obgleich in dieser Studie 11 unabhängige Urteiler zum Einsatz kamen. Auch hier wurde keine erwähnenswerte Korrelation zwischen Fragebogendaten und Laut-Denken Erhebung ermittelt. In Studie 3 wurden metakognitive Aktivitäten zu mehreren Messzeitpunkten im Kontext einer komplexen Gruppenlernaufgabe erhoben. Die Ergebnisse weisen auf einen individuellen Einsatz metakognitiver Aktivitäten unabhängig von der Lernsituation hin. Insgesamt lassen die Ergebnisse der Studien darauf schließen, dass Laut-Denken zumindest dann keine valide Erfassung metakognitiver Aktivitäten während des Lernens ermöglicht, wenn polytome Kategoriensysteme mit einer hohen Anzahl an Kategorien zum Einsatz kommen. Außerdem ist in spezifischen Lernsituationen von potentiellen reaktiven Effekte der Erhebung auszugehen. Dies gilt auch für die Erfassung des Monitoring- und Überwachungs-/Regulierungsverhaltens mittels Lernleistungsurteilen.

In South Africa today a constant stream of demands characterise higher education. The global employment market expects graduating students to be flexible, adoptable and prepared to take responsibility for their own learning and their own continuous professional development. The pace of technological change in health sciences and the volume of available information highlight the need to develop students’ critical reflective thinking. A paradigm shift is required in health science education from one of providing instruction to one of promoting effective and lifelong learning. Educators in health sciences need to revisit, rethink and evaluate criteria for health practice. The challenge of this research is to investigate the integration of critical reflection as a learning strategy in the outcomes of learning programmes within a transformative education approach; the integration of theory and practice through a reflective learning approach; the development and implementation of different learning tools to facilitate effective learning; the importance of establishing an understanding of the link between the learning styles of students and critical reflection as a learning strategy; and the value of the integration of critical reflection to promote lifelong learning. A mixed methods research approach was primarily utilised to monitor facilitation of learning initiatives and appropriate activities for strengthening the learning-centred approach, through reflective journals and reflective learning groups. A quantitative and qualitative study was used in which a pilot study questionnaire, observations, structured interviews and questionnaires were conducted and completed. The findings of this investigation indicate that critical reflection adds value to the effectiveness of learning. The investigation also revealed the value of sharing learning experiences in a small learning group and proved that the learning environment for radiography education allows enough opportunities to integrate an alternative learning strategy such as critical reflection. All role players in health science education need to build their own skills to function effectively as whole brain thinkers with a view to maximizing learning effectiveness. Reflective practice enhances lifelong learning and can also be utilised as a tool for quality control of the learning programme. / Thesis (PhD)--University of Pretoria, 2008. / Curriculum Studies / unrestricted

Diese Arbeit untersucht prominente Erfassungsmethoden metakognitiver Aktivitäten die während des Lernprozesses zum Einsatz kommen (online) auf deren Güte und Reaktivität. Im Fokus stehen die Methoden Laut-Denken, Fragebogenmethode sowie die Erfassung von Lernleistungsurteilen. Lernaufgaben werden durch komplexe Textlernaufgaben sowie Problemlöseaufgaben in deren Rahmen abduktive Schlüsse gefordert sind repräsentiert. In Studie 1 wurden metakognitive Aktivitäten die mittels retrospektiv eingesetzten Fragebögen sowie mittels Laut-Denken erfasst wurden gegenübergestellt. Dabei wurden die Fragebogenitems parallel zum polytomen Kategoriensystem mit dessen Hilfe die Daten aus der Laut-Denken-Methode ausgewertet wurden konstruiert. Im Rahmen der Auswertung der Laut-Denken Daten war die Übereinstimmung zweiter unabhängiger, gut geschulter Urteiler unbefriedigend. Die Übereinstimmungsunterschiede zwischen den Kategorien sowie zwischen den Probanden waren erheblich. Dies weist darauf hin, dass das Kategoriensystem nicht zur Auswertung der Laut-Denken Daten geeignet ist. Zudem scheinen große Unterschiede in der Nutzung metakognitiver Aktivitäten zwischen den Probanden zu bestehen. Zwischen Fragebogendaten und Laut-Denken-Daten besteht ein geringer nicht signifikanter negativer Zusammenhang. In Studie 2 wurde die Reaktivität der Laut-Denken-Methode und der Aufzeichnung von Lernleistungsurteilen während des Bearbeitens einer Textlese- sowie Problemlöseaufgabe untersucht. Die Ergebnisse dieser experimentellen Studie mit 2x2 Design legen nahe, dass von Laut-Denken im Rahmen von Problemlöseaufgaben reaktive Effekte zu erwarten sind. Von Lernleistungsurteilen hingegen sind reaktive Effekte lediglich im Rahmen von komplexen Textleseaufgaben zu erwarten. Auch im Rahmen dieser Erhebung mittels Laut-Denken konnte lediglich eine unbefriedigende Reliabilität der Messung berichtet werden obgleich in dieser Studie 11 unabhängige Urteiler zum Einsatz kamen. Auch hier wurde keine erwähnenswerte Korrelation zwischen Fragebogendaten und Laut-Denken Erhebung ermittelt. In Studie 3 wurden metakognitive Aktivitäten zu mehreren Messzeitpunkten im Kontext einer komplexen Gruppenlernaufgabe erhoben. Die Ergebnisse weisen auf einen individuellen Einsatz metakognitiver Aktivitäten unabhängig von der Lernsituation hin. Insgesamt lassen die Ergebnisse der Studien darauf schließen, dass Laut-Denken zumindest dann keine valide Erfassung metakognitiver Aktivitäten während des Lernens ermöglicht, wenn polytome Kategoriensysteme mit einer hohen Anzahl an Kategorien zum Einsatz kommen. Außerdem ist in spezifischen Lernsituationen von potentiellen reaktiven Effekte der Erhebung auszugehen. Dies gilt auch für die Erfassung des Monitoring- und Überwachungs-/Regulierungsverhaltens mittels Lernleistungsurteilen.

Cette thèse a pour objectif la conception d’une architecture de contrôle bio-inspirée permettant à un robot autonome de naviguer sur de grandes distances. Le modèle développé permet également d’améliorer la compréhension des mécanismes biologiques impliqués. De précédents travaux ont montré qu’un modèle de cellules de lieu, enregistrées chez le rat, permettait à un robot mobile d’apprendre des comportements de navigation robustes, tels qu’une ronde ou un retour au nid, à partir d’associations entre lieu et action. La reconnaissance d’un lieu ne reposait alors que sur des informations visuelles. L’ambiguïté de certaines situations (e.g. un long couloir) ne permettait pas de naviguer dans de grands environnements. L’ajout d’autres modalités constitue une solution efficace pour augmenter la robustesse dans des environnements complexes. Cette solution nous a permis d’identifier les briques minimales nécessaires à la fusion d’informations multimodales, d’abord par le biais d’un conditionnement simple entre 2 modalités sensorielles, puis par la formalisation d’un modèle, plus générique, de prédictions inter-modales. C’est un mécanisme bas niveau qui permet de générer une cohérence perceptive : l’ensemble des modalités sensorielles s’entraident pour ne renvoyer qu’une perception claire et cohérente aux mécanismes décisionnels de plus haut niveau. Les modalités les plus corrélées sont ainsi capables de combler les informations manquantes d’une modalité défaillante (cas pathologique). Ce modèle implique la mise en place d’un système de prédiction et donc une capacité à détecter de la nouveauté dans ses perceptions. Ainsi, le modèle est également capable de détecter une situation inattendue ou anormale et possède donc une capacité d’auto-évaluation : l’évaluation de ses propres perceptions. Nous nous sommes ensuite mis à la recherche des propriétés fondamentales à tout système d'auto-évaluation.La première propriété essentielle a été de constater qu’évaluer un comportement sensorimoteur revient à reconnaître une dynamique entre sensation et action, plutôt que la simple reconnaissance d’une forme sensorielle. La première brique encapsule donc un modèle interne minimaliste des interactions du robot avec son environnement, qui est la base sur laquelle le système fera des prédictions.La seconde propriété essentielle est la capacité à extraire l’information pertinente par le biais de calculs statistiques. Il est nécessaire que le robot apprenne à capturer les invariants statistiques en supprimant l’information incohérente. Nous avons donc montré qu’il était possible d’estimer une densité de probabilité par le biais d’un simple conditionnement. Cet apprentissage permet de réaliser l’équivalent d’une inférence bayésienne. Le système estime la probabilité de reconnaître un comportement à partir de la reconnaissance d’informations statistiques apprises. C’est donc par la mise en cascade de simples conditionnements que le système peut apprendre à estimer les moments statistiques d’une dynamique (moyenne, variance, asymétrie, etc...). La non-reconnaissance de cette dynamique lui permet de détecter qu’une situation est anormale.Mais détecter un comportement inhabituel ne nous renseigne pas pour autant sur son inefficacité. Le système doit également surveiller l’évolution de cette anomalie dans le temps pour pouvoir juger de la pertinence du comportement. Nous montrons comment un contrôleur émotionnel peut faire usage de cette détection de nouveauté pour réguler le comportement et ainsi permettre au robot d’utiliser la stratégie la plus adaptée à la situation rencontrée. Pour finir, nous avons mis en place une procédure de frustration permettant au robot de lancer un appel à l’aide lorsqu’il détecte qu’il se retrouve dans une impasse. Ce réseau de neurones permet au robot d’identifier les situations qu’il ne maîtrise pas dans le but d’affiner son apprentissage, à l’instar de certains processus développementaux. / The goal of this thesis is to build a bio-inspired architecture allowing a robot to autonomouslynavigate over large distances. In a cognitive science point of view, the model also aim at improv-ing the understanding of the underlying biological mechanisms. Previous works showed thata computational model of hippocampal place cells, based on neurobiological studies made onrodent, allows a robot to learn robust navigation behaviors. The robot can learn a round or ahoming behavior from a few associations between places and actions. The learning and recog-nition of a place were only defined by visual information and shows limitations for navigatinglarge environments.Adding other sensorial modalities is an effective solution for improving the robustness of placesrecognition in complex environments. This solution led us to the elementary blocks requiredwhen trying to perform multimodal information merging. Such merging has been done, first,by a simple conditioning between 2 modalities and next improved by a more generic model ofinter-modal prediction. In this model, each modality learns to predict the others in usual situa-tions, in order to be able to detect abnormal situations and to compensate missing informationof the others. Such a low level mechanism allows to keep a coherent perception even if onemodality is wrong. Moreover, the model can detect unexpected situations and thus exhibit someself-assessment capabilities: the assessment of its own perception. Following this model of self-assessment, we focus on the fundamental properties of a system for evaluating its behaviors.The first fundamental property that pops out is the statement that evaluating a behavior is anability to recognize a dynamics between sensations and actions, rather than recognizing a sim-ple sensorial pattern. A first step was thus to take into account the sensation/action couplingand build an internal minimalist model of the interaction between the agent and its environment.Such of model defines the basis on which the system will build predictions and expectations.The second fundamental property of self-assessment is the ability to extract relevant informa-tion by the use of statistical processes to perform predictions. We show how a neural networkcan estimate probability density functions through a simple conditioning rule. This probabilis-tic learning allows to achieve bayesian inferences since the system estimates the probability ofobserving a particular behavior from statistical information it recognizes about this behavior.The robot estimates the different statistical momentums (mean, variance, skewness, etc...) of abehavior dynamics by cascading few simple conditioning. Then, the non-recognition of such adynamics is interpreted as an abnormal behavior.But detecting an abnormal behavior is not sufficient to conclude to its inefficiency. The systemmust also monitor the temporal evolution of such an abnormality to judge the relevance of thebehavior. We show how an emotional meta-controller can use this novelty detection to regu-late behaviors and so select the best appropriate strategy in a given context. Finally, we showhow a simple frustration mechanism allows the robot to call for help when it detects potentialdeadlocks. Such a mechanism highlights situations where a skills improvement is possible, soas some developmental processes.