Dissertations / Theses on the topic 'Learning framework'

This thesis discusses the topic of how to improve adaptive and personalized e-learning in order to provide novel learning experiences. A recent literature review revealed that adaptive and personalized e-learning systems are not widely used. There is a lack of interoperability between adaptive systems and learning management systems, in addition to limited collaborative and social features. First of all, this thesis investigates the interoperability issue via two case studies. The first case study focuses on how to achieve interoperability between adaptive systems and learning management systems using e-learning standards and the second case study focuses on how to augment e-learning standards with adaptive features. Secondly, this thesis proposes a new social framework for personalized e-learning, in order to provide adaptive and personalized e-learning platforms with new social features. This is not just about creating learning content, but also about developing new ways of learning. For instance, in the presented vision, adaptive learning does not refer to individuals only, but also to groups. Furthermore, the boundaries between authors and learners become less distinct in the Web 2.0 context. Finally, a new social personalized prototype is introduced based on the new social framework for personalized e-learning in order to test and evaluate this framework. The implementation and evaluation of the new system were carried out through a number of case studies.

We propose a developmental framework that enables the robot to learn affordances through interaction with the environment in an unsupervised way and to use these affordances at different levels of robot control, ranging from reactive response to planning. Inspired from Developmental Psychology, the robot&rsquo
s discovery of action possibilities is realized in two sequential phases. In the first phase, the robot that initially possesses a limited number of basic actions and reflexes discovers new behavior primitives by exercising these actions and by monitoring the changes created in its initially crude perception system. In the second phase, the robot explores a more complicated environment by executing the discovered behavior primitives and using more advanced perception to learn further action possibilities. For this purpose, first, the robot discovers commonalities in action-effect experiences by finding effect categories, and then builds predictors for each behavior to map object features and behavior parameters into effect categories. After learning affordances through self-interaction and self-observation, the robot can make plans to achieve desired goals, emulate end states of demonstrated actions, monitor the plan execution and take corrective actions using the perceptual structures employed or discovered during learning. Mobile and manipulator robots were used to realize the proposed framework. Similar to infants, these robots were able to form behavior repertoires, learn affordances, and gain prediction capabilities. The learned affordances were shown to be relative to the robots, provide perceptual economy and encode general relations. Additionally, the affordance-based planning ability was verified in various tasks such as table cleaning and object transportation.

While many Marshallese learners thrive in school environments, far more have struggled to find academic success, both at home and abroad. While this has been documented by educational researchers for decades, there is a dearth of research about how Marshallese students learn most effectively. Examining culturally-sustaining educational models that have resulted in successful student outcomes in other indigenous groups can inform strategies to improve educational experiences for Marshallese students. Understanding how recognized Marshallese experts in a range of fields have successfully learned and passed on knowledge and skills is important to understanding how formal school environments can be shaped to most effectively support Marshallese student learning.

This study examines the learning and teaching experiences of recognized Marshallese holders of traditional and contemporary knowledge and skills, in order to document a Marshallese indigenous learning framework. This research used bwebwenato (talk story) as a research method, to learn from the experiences of ten Marshallese experts in knowledge and skills ranging from sewing to linguistics and from canoe-making to business.

Key findings include the four key components of a Marshallese indigenous learning framework: • Relationships • Motivation for Learning • Teaching Strategies • Extending Networks Teaching strategies are comprised of the commonalities among the way Marshallese have learned and mastered both traditional and contemporary skills. Chief among these are: introducing the topic at a young age, scaffolding, demonstrating and observing, learning through relevant practice, and correcting learners constructively. To a lesser extent, and in a context in which the learner and teacher are not related in a familial way, learning and teaching occurs through visual aids and asking instructor for assistance.

We propose that for all situated agents, the process of task learning has many elements in common. A better understanding of these elements would be beneficial to both engineers attempting to design new agents for task learning and completion, and also to scientists seeking to better understand natural task learning. Therefore, this dissertation sets out our characterisation of agent-independent task learning, and explores its grounding in nature and utility in practise. We achieve this chiefly through the construction and demonstration of two novel task learning systems. Cross-Channel Observation and Imitation Learning (COIL; Wood and Bryson, 2007a,b) is our adaptation of Deb Roy’s Cross-Channel Early Lexical Learning System (CELL; Roy, 1999; Roy and Pentland, 2002) for agent-independent task learning by imitation. The General Task Learning Framework (GTLF) is built upon many of the principles learned through the development of COIL, and can additionally facilitate multi-modal, lifelong learning of complex skills and skill hierarchies. Both systems are validated through experiments conducted in the virtual reality-style game domain of Unreal Tournament (Digital Extremes, 1999). By applying agent-independent learning processes to virtual agents of this kind, we hope that researchers will be more inclined to consider them on a par with robots as tools for learning research.

Thesis (Ph.D.)--Massachusetts Institute of Technology, Dept. of Brain and Cognitive Sciences, 1999.
Includes bibliographical references (p. 297-314).
This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections.
Human concept learning presents a version of the classic problem of induction, which is made particularly difficult by the combination of two requirements: the need to learn from a rich (i.e. nested and overlapping) vocabulary of possible concepts and the need to be able to generalize concepts reasonably from only a few positive examples. I begin this thesis by considering a simple number concept game as a concrete illustration of this ability. On this task, human learners can with reasonable confidence lock in on one out of a billion billion billion logically possible concepts, after seeing only four positive examples of the concept, and can generalize informatively after seeing just a single example. Neither of the two classic approaches to inductive inference hypothesis testing in a constrained space of possible rules and computing similarity to the observed examples can provide a complete picture of how people generalize concepts in even this simple setting. This thesis proposes a new computational framework for understanding how people learn concepts from examples, based on the principles of Bayesian inference. By imposing the constraints of a probabilistic model of the learning situation, the Bayesian learner can draw out much more information about a concept's extension from a given set of observed examples than either rule-based or similarity-based approaches do, and can use this information in a rational way to infer the probability that any new object is also an instance of the concept. There are three components of the Bayesian framework: a prior probability distribution over a hypothesis space of possible concepts; a likelihood function, which scores each hypothesis according to its probability of generating the observed examples; and the principle of hypothesis averaging, under which the learner computes the probability of generalizing a concept to new objects by averaging the predictions of all hypotheses weighted by their posterior probability (proportional to the product of their priors and likelihoods). The likelihood, under the assumption of randomly sampled positive examples, embodies the size principle for scoring hypotheses: smaller consistent hypotheses are more likely than larger hypotheses, and they become exponentially more likely as the number of observed examples increases. The principle of hypothesis averaging allows the Bayesian framework to accommodate both rule-like and similarity-like generalization behavior, depending on how peaked the posterior probability is. Together, the size principle plus hypothesis averaging predict a convergence from similarity-like generalization (due to a broad posterior distribution) after very few examples are observed to rule-like generalization (due to a sharply peaked posterior distribution) after sufficiently many examples have been observed. The main contributions of this thesis are as follows. First and foremost, I show how it is possible for people to learn and generalize concepts from just one or a few positive examples (Chapter 2). Building on that understanding, I then present a series of case studies of simple concept learning situations where the Bayesian framework yields both qualitative and quantitative insights into the real behavior of human learners (Chapters 3-5). These cases each focus on a different learning domain. Chapter 3 looks at generalization in continuous feature spaces, a typical representation of objects in psychology and machine learning with the virtues of being analytically tractable and empirically accessible, but the downside of being highly abstract and artificial. Chapter 4 moves to the more natural domain of learning words for categories of objects and shows the relevance of the same phenomena and explanatory principles introduced in the more abstract setting of Chapters 1-3 for real-world learning tasks like this one. In each of these domains, both similarity-like and rule-like generalization emerge as special cases of the Bayesian framework in the limits of very few or very many examples, respectively. However, the transition from similarity to rules occurs much faster in the word learning domain than in the continuous feature space domain. I propose a Bayesian explanation of this difference in learning curves that places crucial importance on the density or sparsity of overlapping hypotheses in the learner's hypothesis space. To test this proposal, a third case study (Chapter 5) returns to the domain of number concepts, in which human learners possess a more complex body of prior knowledge that leads to a hypothesis space with both sparse and densely overlapping components. Here, the Bayesian theory predicts and human learners produce either rule-based or similarity-based generalization from a few examples, depending on the precise examples observed. I also discusses how several classic reasoning heuristics may be used to approximate the much more elaborate computations of Bayesian inference that this domain requires. In each of these case studies, I confront some of the classic questions of concept learning and induction: Is the acquisition of concepts driven mainly by pre-existing knowledge or the statistical force of our observations? Is generalization based primarily on abstract rules or similarity to exemplars? I argue that in almost all instances, the only reasonable answer to such questions is, Both. More importantly, I show how the Bayesian framework allows us to answer much more penetrating versions of these questions: How does prior knowledge interact with the observed examples to guide generalization? Why does generalization appear rule-based in some cases and similarity-based in others? Finally, Chapter 6 summarizes the major contributions in more detailed form and discusses how this work ts into the larger picture of contemporary research on human learning, thinking, and reasoning.
by Joshua B. Tenenbaum.
Ph.D.

Content is generated on the web at an increasing rate. The type of content varies from text on a traditional webpage to text on social media portals (e.g., social network sites and microblogs). One such example of social media is the microblogging site Twitter. Twitter is known for its high level of activity during live events, natural disasters, and events of global importance. Challenges with the data in the Twitter universe include the limit of 140 characters on the text length. Because of this limitation, the vocabulary in the Twitter universe includes short abbreviations of sentences, emojis, hashtags, and other non-standard usage. Consequently, traditional text classification techniques are not very effective on tweets. Fortunately, sophisticated text processing techniques like cleaning, lemmatizing, and removal of stop words and special characters will give us clean text which can be further processed to derive richer word semantic and syntactic relationships using state of the art feature selection techniques like Word2Vec. Machine learning techniques, using word features that capture semantic and context relationships, can be of benefit regarding classification accuracy. Improving text classification results on Twitter data would pave the way to categorize tweets relative to human defined real world events. This would allow diverse stakeholder communities to interactively collect, organize, browse, visualize, analyze, summarize, and explore content and sources related to crises, disasters, human rights, inequality, population growth, resiliency, shootings, sustainability, violence, etc. Having the events classified into different categories would help us study causality and correlations among real world events. To check the efficacy of our classifier, we would compare our experimental results with an Association Rules (AR) classifier. This classifier composes its rules around the most discriminating words in the training data. The hierarchy of rules, along with an ability to tune to a support threshold, makes it an effective classifier for scenarios where short text is involved. Traditionally, developing classification systems for these purposes requires a great degree of human intervention. Constantly monitoring new events, and curating training and validation sets, is tedious and time intensive. Significant human capital is required for such annotation endeavors. Also, involved efforts are required to tune the classifier for best performance. Developing and tuning classifiers manually using human intervention would not be a viable option if we are to monitor events and trends in real-time. We want to build a framework that would require very little human intervention to build and choose the best among the available performing classification techniques in our system. Another challenge with classification systems is related to their performance with unseen data. For the classification of tweets, we are continually faced with a situation where a given event contains a certain keyword that is closely related to it. If a classifier, built for a particular event, due to overfitting to what is a biased sample with limited generality, is faced with new tweets with different keywords, accuracy may be reduced. We propose building a system that will use very little training data in the initial iteration and will be augmented with automatically labelled training data from a collection that stores all the incoming tweets. A system that is trained on incoming tweets that are labelled using sophisticated techniques based on rich word vector representation would perform better than a system that is trained on only the initial set of tweets. We also propose to use sophisticated deep learning techniques like Convolutional Neural Networks (CNN) that can capture the combination of the words using an n-gram feature representation. Such sophisticated feature representation could account for the instances when the words occur together. We divide our case studies into two phases: preliminary and final case studies. The preliminary case studies focus on selecting the best feature representation and classification methodology out of the AR and the Word2Vec based Logistic Regression classification techniques. The final case studies focus on developing the augmented semi-supervised training methodology and the framework to develop a large collection learning optimizer to generate a highly performant classifier. For our preliminary case studies, we are able to achieve an F1 score of 0.96 that is based on Word2Vec and Logistic Regression. The AR classifier achieved an F1 score of 0.90 on the same data. For our final case studies, we are able to show improvements of F1 score from 0.58 to 0.94 in certain cases based on our augmented training methodology. Overall, we see improvement in using the augmented training methodology on all datasets.
Master of Science

Thesis (Ph.D.)--Indiana University, Dept. of Psychological and Brain Sciences the Dept. of Cognitive Science, 2009.
Title from PDF t.p. (viewed on Jul 19, 2010). Source: Dissertation Abstracts International, Volume: 70-12, Section: B, page: 7870. Advisers: John K. Kruschke; Jerome R. Busemeyer.

This research aimed at developing a framework for measuring and enhancing the learning capability of construction contractors. Construction contractors' learning relates to how they imbibe knowledge and other stimuli from their internal and external business environments and how the acquired knowledge is applied to meet the challenges of current and future business environments. The general study of learning antecedents for construction contractors has mainly focused on training of employees. Equally, a lack of a methodology for measuring the learning capability of a company has been one of the main problems for implementing organisational learning within companies. However, this research is the first attempt to provide the antecedents for learning of construction contractors as entities. The outcome of which is a learning framework for auditing learning capabilities of construction contractors as one of the significant contribution to this research. The learning framework should provide construction executives with the means for measuring the extent to which learning takes place in their corporate establishments. This should promote proactive interventions for continuous improvement of their business processes. The developed learning framework maps ten core learning processes i.e. learning dimensions that influence the learning of construction contractors and addresses improvement through: individual learning; the use of teams; internal sharing of knowledge; learning from reviews; integrating work and learning; undertaking investigations within or with others; learning from or with others; continuous renewal of business processes; search for new development; and acquiring a capability to identify and respond to future possible business processes. Parallel to these learning dimensions, this research has identified eight factors that are required if a construction contractor is to achieve double loop or generative learning. The factors are aimed at providing senior construction executives with proactive intervention strategies to overcome specific barriers to learning within their own organisations. Such factors include: objective progress on learning demonstrated through the measurement of business processes; climate of openness; committed leadership to learning; rewarding innovations; shared vision; systems thinking; personal mastery; and mental modelling. Traditionally, measures of performance have heavily relied on financial indictors. However, such measures often only indicate the level of performance rather than explain the contributing factors. Consequently, the learning framework should provide a composite measure to traditional financial measures for construction contractors for assessment of their learning capabilities. The objective of the developed learning framework is to encourage a proactive stance when addressing improvement of construction contractors for the purpose of meeting the challenges of the evolving business environment. The link between construction contractors' learning and the factors that set the condition for double loop or generative learning were found to exhibit satisfactory levels of reliability and validity. Equally, construction contractors' learning increased with performance in terms of average profit and turnover per employee from the empirical analysis. The learning mechanisms by which construction contractors address their improvement by imbibing knowledge from their internal and external business environments were identified and ranked according to the various learning dimensions. The relationships between application of learning mechanisms were examined in order to enriched the understanding of learning practices of large and medium construction contractors.

This thesis explores participation in adult learning and focuses upon three key areas of interest: reasons for participation, the challenges of participation, and the enabling factors relating to participation. The purpose of the research is to expand understanding in order to enhance and improve learning support practice, through a study of a university based, professionally accredited, part-time, Master's Degree programme in Human Resource Management, which serves as the research setting. The study of participation in adult learning is a well-trodden path, beginning with the seminal work of Cyril Houle in the early 1960s. Since then, researchers have continuously sought to prove, disprove or adapt existing typologies. Research has focused on generating groups of single identified factors, motivational indicators and specific challenges influencing participation in adult learning. Specific models and frameworks related to the enablement of participation are identified as being missing from the participation literature, with reference to enablers existing only within the disparate literature relating to adult learning and its broader contexts and influences. A review of the key literature reveals a lack of a single open framework that considers the reasons for, the challenges to, and the enablers of participation across defined contextual dimensions, for the purposes of understanding the nature of participation. This research presents an original conceptual framework matrix, developed from this existing literature, intended to fill this gap. The matrix affords two key opportunities. Firstly, as a theoretical device by which to organise and review current literature in the field and secondly, as a means to identify, explore and present the dominant factors relating to participation in adult learning. To achieve this the matrix identifies the three key areas of interest: i) the reasons participants have for joining the learning activity; ii) the challenges they have faced in doing so, and finally; iii) the elements and influences that enable them to successfully participate in the learning activity. These areas are reviewed further across four dimensions of the participants' life world, that of the psychological, the professional, the practical and the personal. Utilising a critical realist ontology and a post-positivist epistemology the conceptual framework matrix is used to structure the research design. The study adopts a linear, mixed methods approach to collecting data using types of thematic analysis (quantitative and qualitative), achieved through the use of an online questionnaire and one-to-one interviews with the target population. Viewed through the lens of the conceptual framework matrix, findings from within the research setting demonstrate that participants chose to engage with the learning activity as a result of a wide range of influencing factors. Reasons for participation were dominated by two of the dimensions, professional and psychological. Challenges to participation were found to be dominated by psychological factors, alongside issues of a restrictive learning environment and difficulties in achieving work life balance. The dominant enablers were people, deriving from all aspects of the participants' life-world. To aid successful participation in the learning activity under investigation two key recommendations are made to the programme managers and facilitators: i) the facilitation and encouragement of communities of practice and, ii) the development of links between the programme provider and employers. Further to this, this study suggests that, following further research to establish transferability and usability, the matrix has the potential to contribute to wider practice as an open, exploratory framework to be applied to a variety of different learning activities as a means of identifying areas of improvement or change.

The focus of this thesis is the study of mobile learning, specifically learning in different locations and under various contextual situations, from the perspective of university students. I initially derived and designed a theoretical mobile context-aware learning schedule (mCALS) framework from an extensive literature review. Its objective is to recommend appropriate learning materials to students based on their current locations and circumstances. The framework uses a learning schedule (i.e. electronic-based diary) to inform the location and available time a student has for learning/studying at a particular location. Thereafter, a number of factors are taken into consideration for the recommendation of appropriate learning materials. These are the student’s learning styles, knowledge level, concentration level, frequency of interruption at that location and their available time for learning/studying. In order to determine the potential deployment of the framework as a mobile learning application by intended users, I carried out three types of feasibility studies. First, a pedagogical study was conducted using interviews to explore together with students (a) what their learning requirements were when studying in a mobile environment, (b) whether the framework could potentially be used effectively to support their studies and, (c) using this user-centred understanding, refined user requirements of the framework. Second, a diary study was conducted where I collected data and analysed the usability feasibility of the framework by (a) determining whether students could plan their daily schedule ahead and keep to it, (b) ascertaining which learning contexts were important and, (c) establishing which learning materials were appropriate under which situations. Two validation studies were conducted. The first one was an online experiment utilising Java learning objects. Participants of this study were suggested appropriate learning objects to study with, based on their amount of available time, current motivation level for learning and their proficiency level of Java. The second validation study was an investigation into high-quality Java learning objects available in the public domain. Finally, a technical design of the framework was carried out to determine whether the framework at present could realistically be implemented using current mobile technologies. The data analyses of the feasibility studies show that (a) a learning schedule approach is successful to an extent in obtaining location and available time information to indicate accurate values of these contexts, (b) different learners may require different personalisation strategies when selecting appropriate learning materials for them in mobile environments, and (c) the mCALS framework is particularly well-suited for self-regulated students. I also proposed a set of suggestion rules which can be used to recommend appropriate Java learning materials to students in different contexts. The validation studies show that 1) the proposed suggestion rules are effective in recommending appropriate materials to learners in their situation, in order to enhance their learning experiences, and 2) there are a sufficiently large number of high-quality LOs available in the public domain that can be incorporated for use within my framework. Finally, the development of mCALS has been considered from three perspectives – pedagogical, usability and technical. These perspectives consist of critical components that should be considered when developing and evaluating mobile learning software applications. The results demonstrated that the mCALS framework can potentially be used by students in different locations and situations, and appropriate learning materials can be selected to them, in order to enhance their learning experiences.

Deep learning techniques have achieved profound success in many challenging real-world applications, including image recognition, speech recognition, and machine translation. This success has increased the demand for developing deep neural networks and more effective learning approaches. The aim of this thesis is to consider the problem of learning a neural network classifier and to propose a novel approach to solve this problem under the Information Bottleneck (IB) principle. Based on the IB principle, we associate with the classification problem a representation learning problem, which we call ``IB learning". A careful investigation shows there is an unconventional quantization problem that is closely related to IB learning. We formulate this problem and call it ``IB quantization". We show that IB learning is, in fact, equivalent to the IB quantization problem. The classical results in rate-distortion theory then suggest that IB learning can benefit from a vector quantization approach, namely, simultaneously learning the representations of multiple input objects. Such an approach assisted with some variational techniques, result in a novel learning framework that we call ``Aggregated Learning (AgrLearn)", for classification with neural network models. In this framework, several objects are jointly classified by a single neural network. In other words, AgrLearn can simultaneously optimize against multiple data samples which is different from standard neural networks. In this learning framework, two classes are introduced, ``deterministic AgrLearn (dAgrLearn)" and ``probabilistic AgrLearn (pAgrLearn)". We verify the effectiveness of this framework through extensive experiments on standard image recognition tasks. We show the performance of this framework over a real world natural language processing (NLP) task, sentiment analysis. We also compare the effectiveness of this framework with other available frameworks for the IB learning problem.

This study is predicated on the belief that there does not now exist, nor will there ever exist, any single theory of learning that is broad enough to account for all types of learning yet specific enough to be maximally useful in practical application. Perhaps this dichotomy is the reason for the apparent gap between existing theories of learning and the practice of instructional design. As an alternative to any supposed grand theory of learning—and following the lead of prominent thinkers in the fields of clinical psychology and language teaching—this study proposes a shift toward principles. It presents a principle-based conceptual framework of learning, and recommends use of the framework as a guide for creating domain-specific theories of learning. The purpose of this study was to review theories of learning in the behavioral, cognitive, constructive, human, and social traditions to identify principles of learning local to those theories that might represent specific instances of more universal principles, fundamentally requisite to the facilitation of learning in general. Many of the ideas reviewed have resulted from, or been supported by, direct empirical evidence. Others have been suggested based on observational or practical experience of the theorist. The ideas come from different points in time, are described from a variety of perspectives, and emphasize different aspects and types of learning; yet there are a number of common themes shared among them regarding the means by which learning occurs. It is hypothesized that such themes represent universal and fundamental principles of learning. These principles were the objective of the present study. They have been sought through careful review and analysis of both theoretical and empirical literature by methods of textual research (Clingan, 2008) and constant comparative analysis (Glaser & Strauss, 1967). By way of textual research a methodological lens was defined to identify general themes, and by way of constant comparative analysis these themes were developed further through the analysis and classification of specific instances of those themes in the texts reviewed. Ten such principles were identified: repetition, time, step size, sequence, contrast, significance, feedback, context, engagement, and agency. These ten facilitative principles were then organized in the context of a comprehensive principles-of-learning framework, which includes the four additional principles of potential, target, change, and practice.

This thesis focuses on three of the main components of an e-learning system: Infrastructure model, data integration and personalization. For the infrastructure model, our aim is to get best use of heterogeneously structured, geographically distributed data resources. Therefore, a detailed analysis of the available infrastructure models is carried out and an open source reference implementation based on grid technology is implemented. Furthermore, a simple data integration mechanism is proposed for the suggested reference implementation. For personalization, a statistical algorithm is proposed based on extracting and utilizing the learner model. The learner model based on Felder-Silverman learning style is extracted automatically using NBTree classification algorithm in conjunction with Binary relevance classifier for basic science learners. Experimental results show that the performance of the proposed automated learner modelling approach is consistent with the results, obtained by the questionnaires traditionally used for learning style assessment. In the thesis, the classification results are further utilized for providing the user with personalized queries. Keywords: Interactive learning environments
personalization in e-learning

The objective of this thesis is to propose a Web services oriented framework for distributed e-learning systems aimed at providing a flexible integration model in which all the learning components and applications are well defined, effectively discovered and loosely connected. Web services provide an essential deploy environment to realize dynamic e-learning/e-business systems by facilitating application-to-application interaction. Using the proposed framework, learning service providers will be able to publish their learning objects or services universally and learning service requesters can retrieve those services anywhere, any time with any device (wired or wireless) through common communication protocols. The key values of interoperability and accessibility in the proposed architecture enhance the future distributed e-learning systems to communicate more efficiently and share data more easily. A proof of concept of the proposed system is designed and implemented in a J2EE (Java 2 Enterprise Edition) combined with J2ME (Java 2 Micro Edition) environment, using JAX Pack (Java for XML Pack) for building essential Web services and kSOAP package for parsing SOAP (Simple Object Access Protocol) messages on lightweight platforms. The implementation is a successful demonstration that learning services can be easily accessed through standard Web services interface. A cross-platform service invocation (C# to Java, Windows to Linux) is successfully accomplished in this implementation.

This thesis explores learning experiences in ultrasound scanning by examining the ways that sonographers, at various stages in their professional development, scan patients. The qualitative study developed a framework, the themed content of which emerged from the sonographers' own narratives of their learning experiences. The focus on, and the consequent analysis of the sonographers' narratives at different points in their learning led to the construction of a staged framework. The study's sample was designed to cover a broad spectrum of experience and was divided into two groups, differentiated by their qualification status. Purposive sampling was used, recruiting ten participants who recounted their learning experiences to the researcher. The researcher took the role of participant observer. Data was generated via direct observation of sonographers in their working context and the use of semi-structured interviews facilitated the telling of narratives of the individual's learning experiences. These narratives were then formally analysed to seek a better understanding of why the subject performed in the way that they did. The resulting framework was constructed from the analyses of the narratives; it comprises seven stages ranging from 'starting to scan' to 'excellence in scanning'. This Framework of Learning Experiences in Ultrasound Scanning is the first experiential framework of its kind which shows how progress in scanning develops over time. Its focus on process also adds empirical evidence to the sparse literature concerning scanning performance. The framework is foundational in the sense that it has potential implications for curricula, training, and service delivery related to the Sonographers' profession and role. However, the intent of the framework is to understand better the experience of learning to scan and therefore it is a necessary precursor to any future work that seeks to apply that understanding to practice.

In this work, a framework for a system which will intelligently assign an edge detection filter to an image based on features taken from the image is introduced. The framework has four parts: the learning stage, image feature extraction, training filter creation, and filter selection training. Two prototypes systems of this framework are given. The learning stage for these systems is the Berkeley Segmentation Database coupled with the Baddelay Delta Metric. Feature extraction is performed using a GIST methodology which extracts color, intensity, and orientation information. The set of image features are used as the input to a single hidden layer feed forward neural network trained using back propagation. The system trains against a set of linear cellular automata filters which are determined to best solve the edge image according to the Baddelay Delta Metric. One system uses cellular automata augmented with a fuzzy rule. The systems are trained and tested against the images from the Berkeley Segmentation Database. The results from the testing indicate that systems built on this framework can perform better than standard methods of edge detection on average across many types of images.

This thesis represents the culmination of work carried out as part of an ongoing research into hypermedia authoring for Computer Assisted Language Learning (CALL). It originates from, and is the natural continuation of previous research activities in user interface design, which addressed the problem of transferring existing human factors expertise derived from the field of human-computer interaction (HCI) to the hypermedia CALL authoring process. Problems identified with the development of specific design guidelines for authoring hypermedia CALL led to a need for a thorough examination of the usability field with a view to creating a coherent and exhaustive theoretical framework providing a comprehensive contextual and conceptual design support. At the conceptual level, emphasis is placed on defining the design process from an HCI perspective, on delineating the authoring input and explicating the potential of the hypermedia CALL platform, in terms of specificity, scope and limitations. At the contextual level, this research presents an in-depth study of mental models and user requirements elicited and formulated by students as targeted users on the basis of a selection of relevant applications. The resulting usability field is central to the design of the theoretical framework, inasmuch as it feeds into conceptual design considerations and is instrumental in facilitating and validating a realistic transition from theory into practice. Ultimately, the theoretical framework provides a comprehensive design support encapsulating design guidelines and generating design solutions. The main contribution made to hypermedia CALL rests on providing an extensive contextualized design support in the form of a practical and applicable framework with a sound theoretical underpinning designed to stimulate a conceptual approach to authoring hypermedia CALL environments. Therefore, it is designed to develop a much greater awareness of the design process and the role authors must play within it, as well as to provide a methodology and an approach to further identify and understand student requirements. Last but not least, it is conceived to promote and facilitate the use of design guidelines to turn a complex process into a successful, student-centred design outcome.

I många industriella tillämpningar används robotar för tunga och repetetiva uppgifter. För dessa tillämpningar är iterative learning control (ILC) ett sätt att fånga upp och utnyttja repeterbarheten för att förbättra någon form av referenseföljning. I det här examensarbetet har det tagits fram ett ramverk som ska hjälpa en användare att kunna untyttja ILC. Det visas handgripliga exempel på hur man enkelt kan avända ramverket. Övergången från den betydligt mer vanliga diskreta ILC algoritmen till det kontinuerliga tillvägagångssättet som anänds av ramverket underlättas av teroretisk  underbygga inställningsregler. Den uppnåeliga prestandan demonstreras med hjälp av ramverkets inbyggda plotfunktioner.
In many industrial applications robots are used for heavy and repetitive tasks. For these applications iterative learning control (ILC) is a way to capture the repetitive nature and use it to improve some kind of reference tracking. In this master thesis a framework has been developed to help a user getting started with ILC. Some hands-on examples are given on how to easily use the framework. The transition from the far more common discrete time domain to the continuous time domain used by the framework is eased by tuning theory. The achievable performance is demonstrated with the help of the built-in plot functions of the framework.

Thesis: M. Eng. in Computer Science, Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2018.
This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections.
Cataloged from student-submitted PDF version of thesis.
Includes bibliographical references (page 81).
In this thesis, we present DeepMining, a framework to search for machine learning pipelines. The high-level goal of DeepMining is to solve the pipeline search problem: given a problem and a dataset, find the pipeline best-suited to solve that problem. The DeepMining platform serves as a testbed for developers to experiment with different methods of computing and evaluating machine learning pipelines. Specifically, developers have autonomy over how to evaluate different configurations in parallel, score a pipeline given a dataset and hyperparameter configuration, and efficiently search over the pipeline space. DeepMining was designed with modularity and extensibility in mind: developers can easily implement new search algorithms, scoring functions, and computation frameworks. At the same time, users can switch between these modules with minimal effort.
by Akshay Ravikumar.
M. Eng. in Computer Science

Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Electrical Engineering and Computer Science, 2006.
Includes bibliographical references (p. 147-154).
In recent years, the study of classification shifted to algorithms for training the classifier from data that may be missing the class label. While traditional supervised classifiers already have the ability to cope with some incomplete data, the new type of classifiers do not view unlabeled data as an anomaly, and can learn from data sets in which the large majority of training points are unlabeled. Classification with labeled and unlabeled data, or semi-supervised classification, has important practical significance, as training sets with a mix of labeled an unlabeled data are commonplace. In many domains, such as categorization of web pages, it is easier to collect unlabeled data, than to annotate the training points with labels. This thesis is a study of the information regularization method for semi-supervised classification, a unified framework that encompasses many of the common approaches to semi-supervised learning, including parametric models of incomplete data, harmonic graph regularization, redundancy of sufficient features (co-training), and combinations of these principles in a single algorithm.
(cont.) We discuss the framework in both parametric and non-parametric settings, as a transductive or inductive classifier, considered as a stand-alone classifier, or applied as post-processing to standard supervised classifiers. We study theoretical properties of the framework, and illustrate it on categorization of web pages, and named-entity recognition.
by Adrian Corduneanu.
Ph.D.

The electric power grid is evolving in terms of generation, transmission and distribution network architecture. On the generation side, distributed energy resources (DER) are participating at a much larger scale. Transmission and distribution networks are transforming to a decentralized architecture from a centralized one. Residential and commercial buildings are now considered as active elements of the electric grid which can participate in grid operation through applications such as the Demand Response (DR). DR is an application through which electric power consumption during the peak demand periods can be curtailed. DR applications ensure an economic and stable operation of the electric grid by eliminating grid stress conditions. In addition to that, DR can be utilized as a mechanism to increase the participation of green electricity in an electric grid. The DR applications, in general, are passive in nature. During the peak demand periods, common practice is to shut down the operation of pre-selected electrical equipment i.e., heating, ventilation and air conditioning (HVAC) and lights to reduce power consumption. This approach, however, is not optimal and does not take into consideration any user preference. Furthermore, this does not provide any information related to demand flexibility beforehand. Under the broad concept of grid modernization, the focus is now on the applications of data analytics in grid operation to ensure an economic, stable and resilient operation of the electric grid. The work presented here utilizes data analytics in DR application that will transform the DR application from a static, look-up-based reactive function to a dynamic, context-aware proactive solution. The dynamic demand response framework presented in this dissertation performs three major functionalities: electrical load forecast, electrical load disaggregation and peak load reduction during DR periods. The building-level electrical load forecasting quantifies required peak load reduction during DR periods. The electrical load disaggregation provides equipment-level power consumption. This will quantify the available building-level demand flexibility. The peak load reduction methodology provides optimal HVAC setpoint and brightness during DR periods to reduce the peak demand of a building. The control scheme takes user preference and context into consideration. A detailed methodology with relevant case studies regarding the design process of the network architecture of a deep learning algorithm for electrical load forecasting and load disaggregation is presented. A case study regarding peak load reduction through HVAC setpoint and brightness adjustment is also presented. To ensure the scalability and interoperability of the proposed framework, a layer-based software architecture to replicate the framework within a cloud environment is demonstrated.
Doctor of Philosophy
The modern power grid, known as the smart grid, is transforming how electricity is generated, transmitted and distributed across the US. In a legacy power grid, the utilities are the suppliers and the residential or commercial buildings are the consumers of electricity. However, the smart grid considers these buildings as active grid elements which can contribute to the economic, stable and resilient operation of an electric grid. Demand Response (DR) is a grid application that reduces electrical power consumption during peak demand periods. The objective of DR application is to reduce stress conditions of the electric grid. The current DR practice is to shut down pre-selected electrical equipment i.e., HVAC, lights during peak demand periods. However, this approach is static, pre-fixed and does not consider any consumer preference. The proposed framework in this dissertation transforms the DR application from a look-up-based function to a dynamic context-aware solution. The proposed dynamic demand response framework performs three major functionalities: electrical load forecasting, electrical load disaggregation and peak load reduction. The electrical load forecasting quantifies building-level power consumption that needs to be curtailed during the DR periods. The electrical load disaggregation quantifies demand flexibility through equipment-level power consumption disaggregation. The peak load reduction methodology provides actionable intelligence that can be utilized to reduce the peak demand during DR periods. The work leverages functionalities of a deep learning algorithm to increase forecasting accuracy. An interoperable and scalable software implementation is presented to allow integration of the framework with existing energy management systems.

This research aims at identifying mobile learning characteristics. In this exploratory research, focus groups involving students, teachers and administrative stakeholders from Pakistani universities informed the research findings. Socio-cultural aspect is a major contributing factor, in addition to pedagogical and technological factors. A Mobile Learning Framework (MLF) incorporating these factors was developed to cater for developing countries' learning environments. MLF may be generalized to other developing and developed countries with similar higher education and socio-cultural environments.

Mobile Ad-hoc Networks (MANET) are design and implemented without the need for any infrastructure support. The properties of MANET inherently provide greater challenges in areas like security and reliability. This thesis presents three security protocols which were developed for addressing the MANET security needs. A novel trust calculation methodology and intelligent secure route prediction was designed to an existing MANET routing protocol. These protocols will help to implement a trustworthy MANET, providing a dynamic and secure environment.

Master of Landscape Architecture
Department of Landscape Architecture/Regional and Community Planning
Mary C. Kingery-Page
School environments of today’s urban children are generally inflexible, restricting and uninspiring places for learning and exploration that are disconnected from their surrounding community and nature. Facilities and teaching methods do not keep pace with the evolving needs of the workforce and varying child learning styles (Stanbury 2009). Organized sports, limited free time and standardized testing steal the zest out of childhood discovery once felt by children who grew up with a connection to their surroundings, especially nature. Many adverse effects are seen as a result. “Nature-deficit disorder describes the human costs of alienation from nature, among them: diminished use of the senses, attention difficulties and higher rates of physical and emotional illnesses,” (Louv 2008, 36). Children are left to face the world’s escalating environmental dilemmas with hindered social and cognitive skills, diseases related to association and disassociation from nature and an impaired relationship with their extended community. Programs like University Colorado Denver’s Learning Landscapes and California’s Collaborative for High Performance Schools (CHPS) and have individually worked to improve learning facilities, reconnect students with outdoor curriculum-based learning and establish a bond with their communities. But implemented designs reveal unmet potential, calling for advancement and further evolution of the school learning environment. MontClair Elementary in Oakland, California is a typical urban school with paved schoolyard, restricted boundary, weak link between curriculum and schoolyard, disconnect from the community and disassociation from nature. New CHPS verified facilities are being implemented on their existing campus to accommodate an increase in student population but the link between schoolyard and curriculum has only been minimally addressed in the proposed design. Integrating Learning Landscapes with the Collaborative for High Performance Schools to create a hybrid learning landscape framework will reconnect MontClair Elementary with the surrounding community and nature. Advancement of the CHPS program, through adaptation of their existing scorecard, will allow Hybrid Learning Landscape Framework to be quantitatively applied to MontClair Elementary.

In past research we have demonstrated how methodologies used in the training of performers can both encourage whole-brain learning and answer to the demands of South Africa’s current educational paradigm, outcomes-based education (OBE). OBE is a needs-driven, outcomes-driven and competency-orientated pedagogy, which aims at incorporating learners as active agents within the learning process as opposed to the previous content-driven, teacher-orientated approach to education (Coetzee 2004). Our research was prompted by the constant need for our Drama departments to validate their existence in the light of changing funding structures for the arts, governmental and institutional demands for measured outcomes and our institutions’ emphasis on whole-brain learning as the preferred pedagogical approach to education and training. We explored the ways in which the changes in the South African educational dispensation impact on the work of educators within a Drama department in the Higher Education and Training band (HET) in South Africa. These changes include a focus on competencies and critical outcomes across learning areas and across the qualification bands identified by the new National Qualifications Framework. In our search for ways in which to implement the critical outcomes2 demanded by the OBE framework, we turned to Herrmann’s argument (1995) that optimal, deep structure learning can only take place when whole-brain modes are operative.

Supported by technological development and its impact on everyday activities, e-Learning and b-Learning (Blended Learning) have experienced rapid growth mainly in higher education and training. Its inherent ability to break both physical and cultural distances, to disseminate knowledge and decrease the costs of the teaching-learning process allows it to reach anywhere and anyone. The educational community is divided as to its role in the future. It is believed that by 2019 half of the world's higher education courses will be delivered through e-Learning. While supporters say that this will be the educational mode of the future, its detractors point out that it is a fashion, that there are huge rates of abandonment and that their massification and potential low quality, will cause its fall, assigning it a major role of accompanying traditional education. There are, however, two interrelated features where there seems to be consensus. On the one hand, the enormous amount of information and evidence that Learning Management Systems (LMS) generate during the e-Learning process and which is the basis of the part of the process that can be automated. In contrast, there is the fundamental role of e-tutors and etrainers who are guarantors of educational quality. These are continually overwhelmed by the need to provide timely and effective feedback to students, manage endless particular situations and casuistics that require decision making and process stored information. In this sense, the tools that e-Learning platforms currently provide to obtain reports and a certain level of follow-up are not sufficient or too adequate. It is in this point of convergence Information-Trainer, where the current developments of the LMS are centered and it is here where the proposed thesis tries to innovate. This research proposes and develops a platform focused on decision support in e-Learning environments. Using soft computing and data mining techniques, it extracts knowledge from the data produced and stored by e-Learning systems, allowing the classification, analysis and generalization of the extracted knowledge. It includes tools to identify models of students' learning behavior and, from them, predict their future performance and enable trainers to provide adequate feedback. Likewise, students can self-assess, avoid those ineffective behavior patterns, and obtain real clues about how to improve their performance in the course, through appropriate routes and strategies based on the behavioral model of successful students. The methodological basis of the mentioned functionalities is the Fuzzy Inductive Reasoning (FIR), which is particularly useful in the modeling of dynamic systems. During the development of the research, the FIR methodology has been improved and empowered by the inclusion of several algorithms. First, an algorithm called CR-FIR, which allows determining the Causal Relevance that have the variables involved in the modeling of learning and assessment of students. In the present thesis, CR-FIR has been tested on a comprehensive set of classical test data, as well as real data sets, belonging to different areas of knowledge. Secondly, the detection of atypical behaviors in virtual campuses was approached using the Generative Topographic Mapping (GTM) methodology, which is a probabilistic alternative to the well-known Self-Organizing Maps. GTM was used simultaneously for clustering, visualization and detection of atypical data. The core of the platform has been the development of an algorithm for extracting linguistic rules in a language understandable to educational experts, which helps them to obtain patterns of student learning behavior. In order to achieve this functionality, the LR-FIR algorithm (Extraction of Linguistic Rules in FIR) was designed and developed as an extension of FIR that allows both to characterize general behavior and to identify interesting patterns. In the case of the application of the platform to several real e-Learning courses, the results obtained demonstrate its feasibility and originality. The teachers' perception about the usability of the tool is very good, and they consider that it could be a valuable resource to mitigate the time requirements of the trainer that the e-Learning courses demand. The identification of student behavior models and prediction processes have been validated as to their usefulness by expert trainers. LR-FIR has been applied and evaluated in a wide set of real problems, not all of them in the educational field, obtaining good results. The structure of the platform makes it possible to assume that its use is potentially valuable in those domains where knowledge management plays a preponderant role, or where decision-making processes are a key element, e.g. ebusiness, e-marketing, customer management, to mention just a few. The Soft Computing tools used and developed in this research: FIR, CR-FIR, LR-FIR and GTM, have been applied successfully in other real domains, such as music, medicine, weather behaviors, etc.
Soportado por el desarrollo tecnológico y su impacto en las diferentes actividades cotidianas, el e-Learning (o aprendizaje electrónico) y el b-Learning (Blended Learning o aprendizaje mixto), han experimentado un crecimiento vertiginoso principalmente en la educación superior y la capacitación. Su habilidad inherente para romper distancias tanto físicas como culturales, para diseminar conocimiento y disminuir los costes del proceso enseñanza aprendizaje le permite llegar a cualquier sitio y a cualquier persona. La comunidad educativa se encuentra dividida en cuanto a su papel en el futuro. Se cree que para el año 2019 la mitad de los cursos de educación superior del mundo se impartirá a través del e-Learning. Mientras que los partidarios aseguran que ésta será la modalidad educativa del futuro, sus detractores señalan que es una moda, que hay enormes índices de abandono y que su masificación y potencial baja calidad, provocará su caída, reservándole un importante papel de acompañamiento a la educación tradicional. Hay, sin embargo, dos características interrelacionadas donde parece haber consenso. Por un lado, la enorme generación de información y evidencias que los sistemas de gestión del aprendizaje o LMS (Learning Management System) generan durante el proceso educativo electrónico y que son la base de la parte del proceso que se puede automatizar. En contraste, está el papel fundamental de los e-tutores y e-formadores que son los garantes de la calidad educativa. Éstos se ven continuamente desbordados por la necesidad de proporcionar retroalimentación oportuna y eficaz a los alumnos, gestionar un sin fin de situaciones particulares y casuísticas que requieren toma de decisiones y procesar la información almacenada. En este sentido, las herramientas que las plataformas de e-Learning proporcionan actualmente para obtener reportes y cierto nivel de seguimiento no son suficientes ni demasiado adecuadas. Es en este punto de convergencia Información-Formador, donde están centrados los actuales desarrollos de los LMS y es aquí donde la tesis que se propone pretende innovar. La presente investigación propone y desarrolla una plataforma enfocada al apoyo en la toma de decisiones en ambientes e-Learning. Utilizando técnicas de Soft Computing y de minería de datos, extrae conocimiento de los datos producidos y almacenados por los sistemas e-Learning permitiendo clasificar, analizar y generalizar el conocimiento extraído. Incluye herramientas para identificar modelos del comportamiento de aprendizaje de los estudiantes y, a partir de ellos, predecir su desempeño futuro y permitir a los formadores proporcionar una retroalimentación adecuada. Así mismo, los estudiantes pueden autoevaluarse, evitar aquellos patrones de comportamiento poco efectivos y obtener pistas reales acerca de cómo mejorar su desempeño en el curso, mediante rutas y estrategias adecuadas a partir del modelo de comportamiento de los estudiantes exitosos. La base metodológica de las funcionalidades mencionadas es el Razonamiento Inductivo Difuso (FIR, por sus siglas en inglés), que es particularmente útil en el modelado de sistemas dinámicos. Durante el desarrollo de la investigación, la metodología FIR ha sido mejorada y potenciada mediante la inclusión de varios algoritmos. En primer lugar un algoritmo denominado CR-FIR, que permite determinar la Relevancia Causal que tienen las variables involucradas en el modelado del aprendizaje y la evaluación de los estudiantes. En la presente tesis, CR-FIR se ha probado en un conjunto amplio de datos de prueba clásicos, así como conjuntos de datos reales, pertenecientes a diferentes áreas de conocimiento. En segundo lugar, la detección de comportamientos atípicos en campus virtuales se abordó mediante el enfoque de Mapeo Topográfico Generativo (GTM), que es una alternativa probabilística a los bien conocidos Mapas Auto-organizativos. GTM se utilizó simultáneamente para agrupamiento, visualización y detección de datos atípicos. La parte medular de la plataforma ha sido el desarrollo de un algoritmo de extracción de reglas lingüísticas en un lenguaje entendible para los expertos educativos, que les ayude a obtener los patrones del comportamiento de aprendizaje de los estudiantes. Para lograr dicha funcionalidad, se diseñó y desarrolló el algoritmo LR-FIR, (extracción de Reglas Lingüísticas en FIR, por sus siglas en inglés) como una extensión de FIR que permite tanto caracterizar el comportamiento general, como identificar patrones interesantes. En el caso de la aplicación de la plataforma a varios cursos e-Learning reales, los resultados obtenidos demuestran su factibilidad y originalidad. La percepción de los profesores acerca de la usabilidad de la herramienta es muy buena, y consideran que podría ser un valioso recurso para mitigar los requerimientos de tiempo del formador que los cursos e-Learning exigen. La identificación de los modelos de comportamiento de los estudiantes y los procesos de predicción han sido validados en cuanto a su utilidad por los formadores expertos. LR-FIR se ha aplicado y evaluado en un amplio conjunto de problemas reales, no todos ellos del ámbito educativo, obteniendo buenos resultados. La estructura de la plataforma permite suponer que su utilización es potencialmente valiosa en aquellos dominios donde la administración del conocimiento juegue un papel preponderante, o donde los procesos de toma de decisiones sean una pieza clave, por ejemplo, e-business, e-marketing, administración de clientes, por mencionar sólo algunos. Las herramientas de Soft Computing utilizadas y desarrolladas en esta investigación: FIR, CR-FIR, LR-FIR y GTM, ha sido aplicadas con éxito en otros dominios reales, como música, medicina, comportamientos climáticos, etc.

This thesis aims to apply the reinforcement learning into soccer robot and show the great power of reinforcement learning for the RoboCup. In the first part, the background of reinforcement learning is briefly introduced before showing the previous work on it. Therefore the difficulty in implementing reinforcement learning is proposed. The second section demonstrates basic concepts in reinforcement learning, including three fundamental elements, state, action and reward respectively, and three classical approaches, dynamic programming, monte carlo methods and temporal-difference learning respectively. When it comes to keepaway framework, more explanations are given to further combine keepaway with reinforcement learning. After the suggestion about sarsa algorithm with two function approximation, artificial neural network and tile coding, it is implemented successfully during the simulations. The results show it significantly improves the performance of soccer robot.

In this paper we elaborate on a framework, a set of guidelines, for teachers when designing project based courses. The emergent framework presented in this paper will focus on six themes: (1) overall course design, (2) project task, (3) project group, (4) examination, (5) feedback and (6) course evaluation and improvement and is initially grounded in theory and practice. The framework elaborated in this paper should support teachers' development of a professional autonomy within the norms of a professional group and an active curriculum.

Decision-making under uncertainty has become critical as autonomous and semi-autonomous agents become more ubiquitious in our society. These agents must deal with uncertainty and ambiguity from the environment and still perform desired tasks robustly. Partially observable Markov decision processes (POMDPs) provide a principled mathematical framework for modelling agents operating in such an environment. These models are able to capture the uncertainty from noisy sensors, inaccurate actuators, and perform decision-making in light of the agent's incomplete knowledge of the world. POMDPs have been applied successfully in domains ranging from robotics to dialogue management to medical systems. Extensive research has been conducted on methods for optimizing policies for POMDPs. However, these methods typically assume a model of the environment is known. This thesis presents a Bayesian reinforcement learning framework for learning POMDP parameters during execution. This framework takes advantage of agents which work alongside an operator who can provide optimal policy information to help direct the learning. By using Bayesian reinforcement learning, the agent can perform learning concurrently with execution, incorporate incoming data immediately, and take advantage of prior knowledge of the world. By using such a framework, an agent is able to adapt its policy to that of the operator. This framework is validated on data collected from the interaction manager of an autonomous wheelchair. The interaction manager acts as an intelligent interface between the user and the robot, allowing the user to issue high-level commands through natural interface such as speech. This interaction manager is controlled using a POMDP and acts as a rich scenario for learning in which the agent must adjust to the needs of the user over time.
Comme le nombre d'agents autonomes et semi-autonomes dansnotre société ne cesse de croître, les prises de décisions sous incertitude constituent désormais un problème critique. Malgré l'incertitude et l'ambiguité inhérentes à leurs environnements, ces agents doivent demeurer robustes dans l'exécution de leurs tâches. Les processus de décision markoviens partiellement observables (POMDP) offrent un cadre mathématique permettant la modélisation des agents et de leurs environnements. Ces modèles sont capables de capturer l'incertitude due aux perturbations dans les capteurs ainsi qu'aux actionneurs imprécis. Ils permettent conséquemment une prise de décision tenant compte des connaissances imparfaites des agents. À ce jour, les POMDP ont été utilisés avec succès dans un éventail de domaines, allant de la robotique à la gestion de dialogue, en passant par la médecine. Plusieurs travaux de recherche se sont penchés sur des méthodes visant à optimiser les POMDP. Cependant, ces méthodes requièrent habituellement un modèle environnemental préalablement connu. Dans ce mémoire, une méthode bayésienne d'apprentissage par renforcement est présentée, avec laquelle il est possible d'apprendre les paramètres du modèle POMDP pendant l'éxécution. Cette méthode tire avantage d'une coopération avec un opérateur capable de guider l'apprentissage en divulguant certaines données optimales. Avec l'aide du renforcement bayésien, l'agent peut apprendre pendant l'éxécution, incorporer immédiatement les données nouvelles et profiter des connaissances précédentes, pour finalement pouvoir adapter sa politique de décision à celle de l'opérateur. La méthodologie décrite est validée à l'aide de données produites par le gestionnaire d'interactions d'une chaise roulante autonome. Ce gestionnaire prend la forme d'une interface intelligente entre le robot et l'usager, permettant à celui-ci de stipuler des commandes de haut niveau de façon naturelle, par exemple en parlant à voix haute. Les fonctions du gestionnaire sont accomplies à l'aide d'un POMDP et constituent un scénario d'apprentissage idéal, dans lequel l'agent doit s'ajuster progressivement aux besoins de l'usager.

PONTIFÍCIA UNIVERSIDADE CATÓLICA DO RIO DE JANEIRO
COORDENAÇÃO DE APERFEIÇOAMENTO DO PESSOAL DE ENSINO SUPERIOR
PROGRAMA DE EXCELENCIA ACADEMICA
Hoje em dia, a enorme quantidade de dados disponíveis online apresenta um novo desafio para os processos de descoberta de conhecimento. As abordagens mais utilizadas para enfrentar esse desafio são baseadas em técnicas de aprendizado de máquina. Apesar de serem muito poderosas, essas técnicas exigem que seus parâmetros sejam calibrados para gerar modelos com melhor qualidade. Esses processos de calibração são demorados e dependem das habilidades dos especialistas da área de aprendizado de máquinas. Neste contexto, esta pesquisa apresenta uma estrutura baseada em agentes de software para automatizar a calibração de modelos de aprendizagem de máquinas. Esta abordagem integra conceitos de Engenharia de Software Orientada a Agentes (AOSE) e Aprendizado de Máquinas (ML). Como prova de conceito, foi utilizado o conjunto de dados Iris para mostrar como nossa abordagem melhora a qualidade dos novos modelos gerados por nosso framework. Além disso, o framework foi instanciado para um dataset de imagens médicas e finalmente foi feito um experimento usando o dataset Grid Sector.
Nowadays, the challenge of knowledge discovery is to mine massive amounts of data available online. The most widely used approaches to tackle that challenge are based on machine learning techniques. In spite of being very powerful, those techniques require their parameters to be calibrated in order to generate models with better quality. Such calibration processes are time-consuming and rely on the skills of machine learning experts. Within this context, this research presents a framework based on software agents for automating the calibration of machine learning models. This approach integrates concepts from Agent Oriented Software Engineering (AOSE) and Machine Learning (ML). As a proof of concept, we first train a model for the Iris dataset and then we show how our approach improves the quality of new models generated by our framework. Then, we create instances of the framework to generate models for a medical images dataset and finally we use the Grid Sector dataset for a final experiment.

In this thesis, we propose a new framework for the generation of training data for machine learning techniques used for classification in communications applications. Machine learning-based signal classifiers do not generalize well when training data does not describe the underlying probability distribution of real signals. The simplest way to accomplish statistical similarity between training and testing data is to synthesize training data passed through a permutation of plausible forms of noise. To accomplish this, a framework is proposed that implements arbitrary channel conditions and baseband signals. A dataset generated using the framework is considered, and is shown to be appropriately sized by having $11\%$ lower entropy than state-of-the-art datasets. Furthermore, unsupervised domain adaptation can allow for powerful generalized training via deep feature transforms on unlabeled evaluation-time signals. A novel Deep Reconstruction-Classification Network (DRCN) application is introduced, which attempts to maintain near-peak signal classification accuracy despite dataset bias, or perturbations on testing data unforeseen in training. Together, feature transforms and diverse training data generated from the proposed framework, teaching a range of plausible noise, can train a deep neural net to classify signals well in many real-world scenarios despite unforeseen perturbations.

Thesis: S.M. in Engineering and Management, Massachusetts Institute of Technology, School of Engineering, System Design and Management Program, Engineering and Management Program, 2016.
Cataloged from PDF version of thesis.
Includes bibliographical references (pages 93-97).
Machine Learning (ML) is an emerging business capability that have transformed many organizations by enabling them to learn from past data and helping them predict or make decisions on unknown future events. While ML is no longer the preserve of large IT companies, there are abundant opportunities for mid-sized organizations who do not have the resources of the larger IT companies to exploit their data through ML so as to gain deeper insights. This thesis outlines these opportunities and provide guidance for the adoption of ML by these organizations. This thesis examines available literature on current state of adoption of ML by organizations which highlight the gaps that motivate the thesis in providing a guiding framework for applying ML. To achieve this, the thesis provides the practitioner with an overview of ML from both technology and business perspectives that are integrated from multiple sources, categorized for ease of reference and communicated at the decision making level without delving into the mathematics behind ML. The thesis thereafter proposes the ML Integration framework for the System Architect to review the enterprise model, identify opportunities, evaluate technology adoption and architect the ML System. In this framework, system architecting methodologies as well as Object-Process Diagrams are used to illustrate the concepts and the architecture. The ML Integration framework is subsequently applied in the context of a hypothetical mid-sized hospital to illustrate how an architect would go about utilizing this framework. Future work is needed to validate the ML Integration framework, as well as improve the overview of ML specific to application domains such as recommender systems and speech/image recognition.
by Alan Tham.
S.M. in Engineering and Management

Thesis (M. Eng.)--Massachusetts Institute of Technology, Dept. of Electrical Engineering and Computer Science, 2010.
Cataloged from PDF version of thesis.
Includes bibliographical references (p. 81-83).
We consider the problem of building a viable multiclass classification system that minimizes training data, is robust to noisy, imbalanced samples, and outputs confidence scores along with its predications. These goals address critical steps along the entire classification pipeline that pertain to collecting data, training, and classifying. To this end, we investigate the merits of a classification framework that uses a robust algorithm known as Regularized Least Squares (RLS) as its basic classifier. We extend RLS to account for data imbalances, perform efficient active learning, and output confidence scores. Each of these extensions is a new result that combines with our other findings to give an altogether novel and effective classification system. Our first set of results investigates various ways to handle multiclass data imbalances and ultimately leads to a derivation of a weighted version of RLS with and without an offset term. Weighting RLS provides an effective countermeasure to imbalanced data and facilitates the automatic selection of a regularization parameter through exact and efficient calculation of the Leave One Out error. Next, we present two methods that estimate multiclass confidence from an asymptotic analysis of RLS and another method that stems from a Bayesian interpretation of the classifier. We show that while the third method incorporates more information in its estimate, the asymptotic methods are more accurate and resilient to imperfect kernel and regularization parameter choices. Finally, we present an active learning extension of RLS (ARLS) that uses our weighting methods to overcome imbalanced data. ARLS is particularly adept to this task because of its intelligent selection scheme.
by Hristo Spassimirov Paskov.
M.Eng.

Computational approaches to linguistic analysis have been used for more than half a century. The main tools come from the field of Natural Language Processing (NLP) and are based on rule-based or corpora-based (supervised) methods. Despite the undeniable success of supervised learning methods in NLP, they have two main drawbacks: on the practical side, it is expensive to produce the manual annotation (or the rules) required and it is not easy to find annotators for less common languages. A theoretical disadvantage is that the computational analysis produced is tied to a specific theory or annotation scheme. Unsupervised methods offer the possibility to expand our analyses into more resourcepoor languages, and to move beyond the conventional linguistic theories. They are a way of observing patterns and regularities emerging directly from the data and can provide new linguistic insights. In this thesis I explore unsupervised methods for inducing parts of speech across languages. I discuss the challenges in evaluation of unsupervised learning and at the same time, by looking at the historical evolution of part-of-speech systems, I make the case that the compartmentalised, traditional pipeline approach of NLP is not ideal for the task. I present a generative Bayesian system that makes it easy to incorporate multiple diverse features, spanning different levels of linguistic structure, like morphology, lexical distribution, syntactic dependencies and word alignment information that allow for the examination of cross-linguistic patterns. I test the system using features provided by unsupervised systems in a pipeline mode (where the output of one system is the input to another) and show that the performance of the baseline (distributional) model increases significantly, reaching and in some cases surpassing the performance of state-of-the-art part-of-speech induction systems. I then turn to the unsupervised systems that provided these sources of information (morphology, dependencies, word alignment) and examine the way that part-of-speech information influences their inference. Having established a bi-directional relationship between each system and my part-of-speech inducer, I describe an iterated learning method, where each component system is trained using the output of the other system in each iteration. The iterated learning method improves the performance of both component systems in each task. Finally, using this iterated learning framework, and by using parts of speech as the central component, I produce chains of linguistic structure induction that combine all the component systems to offer a more holistic view of NLP. To show the potential of this multi-level system, I demonstrate its use ‘in the wild’. I describe the creation of a vastly multilingual parallel corpus based on 100 translations of the Bible in a diverse set of languages. Using the multi-level induction system, I induce cross-lingual clusters, and provide some qualitative results of my approach. I show that it is possible to discover similarities between languages that correspond to ‘hidden’ morphological, syntactic or semantic elements.

This study explored Professional Teacher Technical Identity Development through the use of Mobile Technology. A sample of fifteen teachers was conveniently selected from one school in an urban setting. An action research was designed consisting of three phases. Each phase formed the basis of the next phase to identify the development of professional teacher technical identity. Data was collected using a written questionnaire, two reflective journals, an online questionnaire, focus group discussions, lesson reflections, and interviews. Each instrument was designed using the literature to identify factors that impact on the implementation of mobile technology in classrooms and teachers’ acceptance towards mobile technology. The results were interpreted using three existing models to create a framework: The Technology, Pedagogy and Content Knowledge model, Technology Acceptance Model and Substitution, Augmentation, Modification and Redefinition Model. It was found that there are six factors that affect the perceived usefulness and perceived ease of use of technology. These are attitude, anxiety, ability, subjective norm, facilitating conditions and voluntariness. The perceived ease of use and perceived usefulness determine the level at which technology is implemented in classrooms. The level of integration determines whether or not successful teaching in terms of the three elements of TPCK is being used. During the process whereby teachers attempt to implement technology in their classrooms, it is possible to identify changes in their professional teacher technical identity development. These changes are interpreted and a new framework for Professional Teacher Technical Identity Development is created. It is proposed that this framework can be used to explain the implementation process and behaviour of teachers during the process as their teacher identity is altered.
Thesis (PhD)--University of Pretoria, 2017.
Science, Mathematics and Technology Education
PhD
Unrestricted