Dissertationen zum Thema „Cross-learning“

Key to achieving more efficient machine intelligence is the capability to analysing and understanding data across different views - which can be camera views or modality views (such as visual and textual). One generic learning paradigm for automated understanding data from different views called cross-view learning which includes cross-view matching, cross-view fusion and cross-view generation. Specifically, this thesis investigates two of them, cross-view matching and cross-view generation, by developing new methods for addressing the following specific computer vision problems. The first problem is cross-view matching for person re-identification which a person is captured by multiple non-overlapping camera views, the objective is to match him/her across views among a large number of imposters. Typically a person's appearance is represented using features of thousands of dimensions, whilst only hundreds of training samples are available due to the difficulties in collecting matched training samples. With the number of training samples much smaller than the feature dimension, the existing methods thus face the classic small sample size (SSS) problem and have to resort to dimensionality reduction techniques and/or matrix regularisation, which lead to loss of discriminative power for cross-view matching. To that end, this thesis proposes to overcome the SSS problem in subspace learning by matching cross-view data in a discriminative null space of the training data. The second problem is cross-view matching for zero-shot learning where data are drawn from different modalities each for a different view (e.g. visual or textual), versus single-modal data considered in the first problem. This is inherently more challenging as the gap between different views becomes larger. Specifically, the zero-shot learning problem can be solved if the visual representation/view of the data (object) and its textual view are matched. Moreover, it requires learning a joint embedding space where different view data can be projected to for nearest neighbour search. This thesis argues that the key to make zero-shot learning models succeed is to choose the right embedding space. Different from most existing zero-shot learning models utilising a textual or an intermediate space as the embedding space for achieving crossview matching, the proposed method uniquely explores the visual space as the embedding space. This thesis finds that in the visual space, the subsequent nearest neighbour search would suffer much less from the hubness problem and thus become more effective. Moreover, a natural mechanism for multiple textual modalities optimised jointly in an end-to-end manner in this model demonstrates significant advantages over existing methods. The last problem is cross-view generation for image captioning which aims to automatically generate textual sentences from visual images. Most existing image captioning studies are limited to investigate variants of deep learning-based image encoders, improving the inputs for the subsequent deep sentence decoders. Existing methods have two limitations: (i) They are trained to maximise the likelihood of each ground-truth word given the previous ground-truth words and the image, termed Teacher-Forcing. This strategy may cause a mismatch between training and testing since at test-time the model uses the previously generated words from the model distribution to predict the next word. This exposure bias can result in error accumulation in sentence generation during test time, since the model has never been exposed to its own predictions. (ii) The training supervision metric, such as the widely used cross entropy loss, is different from the evaluation metrics at test time. In other words, the model is not directly optimised towards the task expectation. This learned model is therefore suboptimal. One main underlying reason responsible is that the evaluation metrics are non-differentiable and therefore much harder to be optimised against. This thesis overcomes the problems as above by exploring the reinforcement learning idea. Specifically, a novel actor-critic based learning approach is formulated to directly maximise the reward - the actual Natural Language Processing quality metrics of interest. As compared to existing reinforcement learning based captioning models, the new method has the unique advantage of a per-token advantage and value computation is enabled leading to better model training.

An ontology is a knowledge-representation structure, where words, terms or concepts are defined by their mutual hierarchical relations. Ontologies are becoming ever more prevalent in the world of natural language processing, where we currently see a tendency towards using semantics for solving a variety of tasks, particularly tasks related to information access. Ontologies, taxonomies and thesauri (all related notions) are also used in various variants by humans, to standardize business transactions or for finding conceptual relations between terms in, e.g., the medical domain. The acquisition of machine-readable, domain-specific semantic knowledge is time consuming and prone to inconsistencies. The field of ontology learning therefore provides tools for automating the construction of domain ontologies (ontologies describing the entities and relations within a particular field of interest), by analyzing large quantities of domain-specific texts. This thesis studies three main topics within the field of ontology learning. First, we examine which sources of information are useful within an ontology learning system and how the information sources can be combined effectively. Secondly, we do this with a special focus on cross-language text collections, to see if we can learn more from studying several languages at once, than we can from a single-language text collection. Finally, we investigate new approaches to formal and automatic evaluation of the quality of a learned ontology. We demonstrate how to combine information sources from different languages and use them to train automatic classifiers to recognize lexico-semantic relations. The cross-language data is shown to have a positive effect on the quality of the learned ontologies. We also give theoretical and experimental results, showing that our ontology evaluation method is a good complement to and in some aspects improves on the evaluation measures in use today.<br>För att köpa boken skicka en beställning till exp@ling.su.se/ To order the book send an e-mail to exp@ling.su.se

Sparse high dimensional time series are common in industry, such as in supply chain demand and retail sales. Accurate and reliable forecasting of high dimensional time series is essential for supply chain planning and business management. In practical applications, sparse high dimensional time series prediction faces three challenges: (1) simple models cannot capture complex patterns, (2) insufficient data prevents us from pursuing more advanced models, and (3) time series in the same dataset may have widely different properties. These challenges prevent the currently prevalent models and theoretically successful advanced models (e.g., neural networks) from working in actual use. We focus our research on a pharmaceutical (pharma) demand forecasting problem. To overcome the challenges faced by sparse high dimensional time series, we develop a cross-series learning framework that trains a machine learning model on multiple related time series and uses cross-series information to improve forecasting accuracy. Cross-series learning is further optimized by dividing the global time series into subgroups based on three grouping schemes to balance the tradeoff between sample size and sample quality. Moreover, downstream inventory is introduced as an additional feature to support demand forecasting. Combining the cross-series learning framework with advanced machine learning models, we significantly improve the accuracy of pharma demand predictions. To verify the generalizability of cross-series learning, a generic forecasting framework containing the operations required for cross-series learning is developed and applied to retail sales forecasting. We further confirm the benefits of cross-series learning for advanced models, especially RNN. In addition to the grouping schemes based on product characteristics, we also explore two grouping schemes based on time series clustering, which do not require domain knowledge and can be applied to other fields. Using a retail sales dataset, our cross-series machine learning models are still superior to the baseline models. This dissertation develops a collection of cross-series learning techniques optimized for sparse high dimensional time series that can be applied to pharma manufacturers, retailers, and possibly other industries. Extensive experiments are carried out on real datasets to provide empirical value and insights for relevant theoretical studies. In practice, our work guides the actual use of cross-series learning.

Cross-media games are evolving as a new exciting platform for gaming where different devices are used to create a type of game play were a variant of devices, such as mobile phones and laptops are used. This thesis investigates the possibility of merging cross-media games into the domain of Mobile Learning to create a type of mobile learning game where collaboration becomes a vital part of the game play and style enhances collaboration between the users. By studying cross-media games, key features are captured and converted into requirements that are realised in a prototype that enables cross-media gaming with the intention of creating an environment in which learning could be supported. The development process of the prototype is described and evaluated in the thesis. The result presents a categorization of the key features for cross-media gaming and a prototype of a cross-media game. The thesis investigates which are the key technical features for creating cross-medial games for learning that can be identified for supporting the development process? The results presents a categorization of identified features along with potential future work based on the thesis. It is shown that features related to data sharing are highly prioritized and that certain features are absolutely required to enable cross-media gaming whilst others have less priority.

Automatic analysis of visual data is a key objective of computer vision research; and performing visual recognition of objects from images is one of the most important steps towards understanding and gaining insights into the visual data. Most existing approaches in the literature for the visual recognition are based on a supervised learning paradigm. Unfortunately, they require a large amount of labelled training data which severely limits their scalability. On the other hand, recognition is instantaneous and effortless for humans. They can recognise a new object without seeing any visual samples by just knowing the description of it, leveraging similarities between the description of the new object and previously learned concepts. Motivated by humans recognition ability, this thesis proposes novel approaches to tackle cross-class transfer learning (crossclass recognition) problem whose goal is to learn a model from seen classes (those with labelled training samples) that can generalise to unseen classes (those with labelled testing samples) without any training data i.e., seen and unseen classes are disjoint. Specifically, the thesis studies and develops new methods for addressing three variants of the cross-class transfer learning: Chapter 3 The first variant is transductive cross-class transfer learning, meaning labelled training set and unlabelled test set are available for model learning. Considering training set as the source domain and test set as the target domain, a typical cross-class transfer learning assumes that the source and target domains share a common semantic space, where visual feature vector extracted from an image can be embedded using an embedding function. Existing approaches learn this function from the source domain and apply it without adaptation to the target one. They are therefore prone to the domain shift problem i.e., the embedding function is only concerned with predicting the training seen class semantic representation in the learning stage during learning, when applied to the test data it may underperform. In this thesis, a novel cross-class transfer learning (CCTL) method is proposed based on unsupervised domain adaptation. Specifically, a novel regularised dictionary learning framework is formulated by which the target class labels are used to regularise the learned target domain embeddings thus effectively overcoming the projection domain shift problem. Chapter 4 The second variant is inductive cross-class transfer learning, that is, only training set is assumed to be available during model learning, resulting in a harder challenge compared to the previous one. Nevertheless, this setting reflects a real-world setting in which test data is available after the model learning. The main problem remains the same as the previous variant, that is, the domain shift problem occurs when the model learned only from the training set is applied to the test set without adaptation. In this thesis, a semantic autoencoder (SAE) is proposed building on an encoder-decoder paradigm. Specifically, first a semantic space is defined so that knowledge transfer is possible from the seen classes to the unseen classes. Then, an encoder aims to embed/project a visual feature vector into the semantic space. However, the decoder exerts a generative task, that is, the projection must be able to reconstruct the original visual features. The generative task forces the encoder to preserve richer information, thus the learned encoder from seen classes is able generalise better to the new unseen classes. Chapter 5 The third one is unsupervised cross-class transfer learning. In this variant, no supervision is available for model learning i.e., only unlabelled training data is available, leading to the hardest setting compared to the previous cases. The goal, however, is the same, learning some knowledge from the training data that can be transferred to the test data composed of completely different labels from that of training data. The thesis proposes a novel approach which requires no labelled training data yet is able to capture discriminative information. The proposed model is based on a new graph regularised dictionary learning algorithm. By introducing a l1- norm graph regularisation term, instead of the conventional squared l2-norm, the model is robust against outliers and noises typical in visual data. Importantly, the graph and representation are learned jointly, resulting in further alleviation of the effects of data outliers. As an application, person re-identification is considered for this variant in this thesis.

Defect prediction models assist tester practitioners on prioritizing the most defect-prone parts of the software. The approach called Cross-Project Defect Prediction (CPDP) refers to the use of known external projects to compose the training set. This approach is useful when the amount of historical defect data of a company to compose the training set is inappropriate or insufficient. Although the principle is attractive, the predictive performance is a limiting factor. In recent years, several methods were proposed aiming at improving the predictive performance of CPDP models. However, to the best of our knowledge, there is no evidence of which CPDP methods typically perform best. Moreover, there is no evidence on which CPDP methods perform better for a specific application domain. In fact, there is no machine learning algorithm suitable for all domains. The decision task of selecting an appropriate algorithm for a given application domain is investigated in the meta-learning literature. A meta-learning model is characterized by its capacity of learning from previous experiences and adapting its inductive bias dynamically according to the target domain. In this work, we investigate the feasibility of using meta-learning for the recommendation of CPDP methods. In this thesis, three main goals were pursued. First, we provide an experimental analysis to investigate the feasibility of using Feature Selection (FS) methods as an internal procedure to improve the performance of two specific CPDP methods. Second, we investigate which CPDP methods present typically best performances. We also investigate whether the typically best methods perform best for the same project datasets. The results reveal that the most suitable CPDP method for a project can vary according to the project characteristics, which leads to the third investigation of this work. We investigate the several particularities inherent to the CPDP context and propose a meta-learning solution able to learn from previous experiences and recommend a suitable CDPD method according to the characteristics of the project being predicted. We evaluate the learning capacity of the proposed solution and its performance in relation to the typically best CPDP methods.<br>Modelos de predição de defeitos auxiliam profissionais de teste na priorização de partes do software mais propensas a conter defeitos. A abordagem de predição de defeitos cruzada entre projetos (CPDP) refere-se à utilização de projetos externos já conhecidos para compor o conjunto de treinamento. Essa abordagem é útil quando a quantidade de dados históricos de defeitos é inapropriada ou insuficiente para compor o conjunto de treinamento. Embora o princípio seja atrativo, o desempenho de predição é um fator limitante nessa abordagem. Nos últimos anos, vários métodos foram propostos com o intuito de melhorar o desempenho de predição de modelos CPDP. Contudo, na literatura, existe uma carência de estudos comparativos que apontam quais métodos CPDP apresentam melhores desempenhos. Além disso, não há evidências sobre quais métodos CPDP apresentam melhor desempenho para um domínio de aplicação específico. De fato, não existe um algoritmo de aprendizado de máquina que seja apropriado para todos os domínios de aplicação. A tarefa de decisão sobre qual algoritmo é mais adequado a um determinado domínio de aplicação é investigado na literatura de meta-aprendizado. Um modelo de meta-aprendizado é caracterizado pela sua capacidade de aprender a partir de experiências anteriores e adaptar seu viés de indução dinamicamente de acordo com o domínio alvo. Neste trabalho, nós investigamos a viabilidade de usar meta-aprendizado para a recomendação de métodos CPDP. Nesta tese são almejados três principais objetivos. Primeiro, é conduzida uma análise experimental para investigar a viabilidade de usar métodos de seleção de atributos como procedimento interno de dois métodos CPDP, com o intuito de melhorar o desempenho de predição. Segundo, são investigados quais métodos CPDP apresentam um melhor desempenho em um contexto geral. Nesse contexto, também é investigado se os métodos com melhor desempenho geral apresentam melhor desempenho para os mesmos conjuntos de dados (ou projetos de software). Os resultados revelam que os métodos CPDP mais adequados para um projeto podem variar de acordo com as características do projeto sendo predito. Essa constatação conduz à terceira investigação realizada neste trabalho. Foram investigadas as várias particularidades inerentes ao contexto CPDP a fim de propor uma solução de meta-aprendizado capaz de aprender com experiências anteriores e recomendar métodos CPDP adequados, de acordo com as características do software. Foram avaliados a capacidade de meta-aprendizado da solução proposta e a sua performance em relação aos métodos base que apresentaram melhor desempenho geral.

La spécification de requêtes est généralement une tâche difficile pour les utilisateurs non-experts. Le problème devient encore plus difficile quand les utilisateurs ont besoin d'interroger des bases de données de grande taille et donc difficiles à visualiser. Le schéma pourrait aider à cette spécification, mais celui-ci manque souvent ou est incomplet quand les données viennent de sources hétérogènes. Dans cette thèse, nous abordons le problème de la spécification de requêtes pour les utilisateurs non-experts. Nous identifions deux approches pour attaquer ce problème : apprendre les requêtes à partir d'exemples ou transformer les données dans un format plus facilement interrogeable par l'utilisateur. Nos contributions suivent ces deux directions et concernent trois modèles de données parmi les plus populaires : XML, relationnel et orienté graphe. Cette thèse comprend deux parties, consacrées à (i) la définition et la transformation de schémas, et (ii) l'apprentissage de schémas et de requêtes. Dans la première partie, nous définissons des formalismes de schémas pour les documents XML non-ordonnés et nous analysons leurs propriétés computationnelles; nous étudions également la complexité du problème d'échange de données entre une source relationnelle et une cible orientée graphe. Dans la deuxième partie, nous étudions le problème de l'apprentissage à partir d'exemples pour les schémas XML proposés dans la première partie, ainsi que pour les requêtes de jointures relationnelles et les requêtes de chemins sur les graphes. Nous proposons notamment un scénario interactif qui permet d'aider des utilisateurs non-experts à définir des requêtes dans ces deux classes<br>Specifying a database query using a formal query language is typically a challenging task for non-expert users. In the context of big data, this problem becomes even harder because it requires the users to deal with database instances of large size and hence difficult to visualize. Such instances usually lack a schema to help the users specify their queries, or have an incomplete schema as they come from disparate data sources. In this thesis, we address the problem of query specification for non-expert users. We identify two possible approaches for tackling this problem: learning queries from examples and translating the data in a format that the user finds easier to query. Our contributions are aligned with these two complementary directions and span over three of the most popular data models: XML, relational, and graph. This thesis consists of two parts, dedicated to (i) schema definition and translation, and to (ii) learning schemas and queries. In the first part, we define schema formalisms for unordered XML and we analyze their computational properties; we also study the complexity of the data exchange problem in the setting of a relational source and a graph target database. In the second part, we investigate the problem of learning from examples the schemas for unordered XML proposed in the first part, as well as relational join queries and path queries on graph databases. The interactive scenario that we propose for these two classes of queries is immediately applicable to assisting non-expert users in the process of query specification

This phenomenographic study, explores the collaborative open learning experience of academic staff and open learners in cross-institutional academic development settings, and adds to what is known in these settings. It provides new insights for academic developers and course designers about the benefits of crossing boundaries (i.e. open learning) in an academic development context and proposes an alternative model to traditional academic Continuing Professional Development (CPD). It engages academic staff in experiencing novel approaches to learning and teaching and developing as practitioners through engagement in academic CPD that stretches beyond institutional boundaries, characterised by diversity and based on collaboration and openness. Data collection was conducted using a collective case study approach to gain insights into the collective lived collaborative open learning experience in two authentic cross-institutional academic development settings with collaborative learning features designed in. At least one of the institutions involved in each course was based in the United Kingdom. Twenty two individual phenomenographic interviews were conducted and coded. The findings illustrate that collaborative open learning was experienced as two dynamic immersive and selective patterns. Boundary crossing as captured in the categories of description and their qualitatively different variations, shaped that experience and related to modes of participation; time, place and space; culture and language as well as diverse professional contexts. Facilitator support and the elasticity of the design also positively shaped this experience. The community aspect influenced study participants' experience at individual and course level and illuminated new opportunities for academic development practice based on cross-boundary community-led approaches. The findings synthesised in the phenomenographic outcome space, depicting the logical relationships of the eleven categories of description in this study, organised in structural factors, illustrate how these contributed and shaped the lived experience, together with a critical discussion of these with the literature, aided the creation of the openly licensed cross-boundary collaborative open learning framework for cross-institutional academic development, the final output of this study. A design tool developed from the results is included that aims to inform academic developers and other course designers who may be considering and planning to model and implement such approaches in their own practice.

This work outlines a possible technique for identifying webpages that contain product  specifications. Using support vector machines a product web page classifier was constructed and tested with various settings. The final result for this classifier ended up being 0.958 in precision and 0.796 in recall for product pages. The scores imply that the method could be considered a valid technique in real world web classification tasks if additional features and more data were made available.

Thesis: M. Eng., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2014.<br>Cataloged from PDF version of thesis.<br>Includes bibliographical references (pages 77-79).<br>Cross Document Coreference Resolution (CDCR) is the problem of learning which mentions, coming from several different documents, correspond to the same entity. This thesis approaches the CDCR problem by first turning it into a structure learning problem. A latent tree structure, in which leaves correspond to observed mentions and internal nodes correspond to latent sub-entities, is learned. A greedy clustering heuristic can then be used to select subtrees from the learned tree structure as entities. As with other structure learning problems, it is prudent to envoke Occam's razor and perform regularization to obtain the simplest hypothesis. When the state space consists of tree structures, we can impose a bias on the possible structure. Different aspects of tree structure (i.e. number of edges, depth of the leaves, etc.) can be penalized in these models to improve the generalization of thes models. This thesis draws upon these ideas to provide a new model for CDCR. To learn parameters, we implement a parameter estimation algorithm based on existing stochastic gradient-descent based algorithms and show how to further tune regularization parameters. The latent tree structure is then learned using MCMC inference. We show how structural regularization plays a critical role in the inference procedure. Finally, we empirically show that our model out-performs previous work, without using a sophisticated set of features.<br>by Eric Shyu.<br>M. Eng.

Our face is a key modality to convey emotions and infer intention. This makes face analysis an important factor in understanding the underlying mechanisms of interaction. Automatic solutions for facial expression recognition promise to deliver a significant fraction of the currently missing component of non-verbal communication to the human-machine interaction enabling more fulfilling experience closely modelling interpersonal communication. This thesis presents three major contributions aimed to overcome a number of issues currently preventing modern face analysis solutions from being applied in practice. The problem of reliable automatic discovery of facial actions is first considered from the point of view of manual feature craft, exploring ways to highlight features related to interpersonal commonalities in facial expression appearance, disregarding those corresponding to environmental conditions and subjective differences. It is then approached from the Multi-Task and Transfer learning perspective, presenting solutions for cost and performance efficient training of facial expression detection algorithms. Finally, a novel solution is proposed for multi-database heterogeneous data representation aimed to provide an environment for better generalisable face analysis solutions training and evaluation.

Cette thèse de doctorat est le résultat de mes travaux de recherche dans le domaine de l'apprentissage automatique, du traitement d'image et du transport intelligent pour résoudre le problème du système de protection des piétons (PPS) multi-tâches comprenant non seulement la classification, la détection et le suivi des piétons, mais aussi l'action des piétons- classification et prédiction des unités, et enfin estimation du risque piéton. De plus, notre système PPS utilise des approches originales d'apprentissage en profondeur inter-modalités. Le but de notre travail de recherche est de développer un composant de protection des piétons intelligent basé uniquement sur un système de vision stéréo unique utilisant une architecture d'apprentissage en profondeur cross-modalité optimale afin de classer l'action piétonne actuelle, de prédire leurs prochaines actions et enfin d'estimer le piéton risque au moment de traverser pour chaque piéton. Premièrement, nous étudions la composante de classification où nous avons analysé comment les représentations d'apprentissage d'une modalité permettraient de reconnaître d'autres modalités au sein de divers apprentissages profonds, un terme comme apprentissage multimodal. Deuxièmement, nous étudions comment l'apprentissage inter-modalité améliore la détection de l'action piétonne de bout en bout.Troisièmement, nous analysons la prédiction de l'action des piétons et l'estimation du temps à traverser<br>This PhD thesis is the result of my research work in the machine learning, image processing and intelligent transportation field for solving the problem of multi-task pedestrian protection system (PPS) including not only pedestrian classification, detection and tracking, but also pedestrian action-unit classification and prediction, and finally pedestrian risk estimation. Moreover, our PPS system uses original cross-modality deep learning approaches. The goal of our research work is to develop an intelligent pedestrian protection component-based only on single stereo vision system using an optimal cross-modality deep learning architecture in order to classify the current pedestrian action, predict their next actions and finally to estimate the pedestrian risk by the time to cross for each pedestrian. First, we investigate the classification component where we analyzed how learning representations from one modality would enable recognition for other modalities within various deep learning, which one term as cross-modality learning. Second, we study how the cross-modality learning improves an end-to-end the pedestrian action

Multi-modality medical imaging is increasingly used for comprehensive assessment of complex diseases in either diagnostic examinations or as part of medical research trials. Different imaging modalities provide complementary information about living tissues. However, multi-modal examinations are not always possible due to adversary factors such as patient discomfort, increased cost, prolonged scanning time and scanner unavailability. In addition, in large imaging studies, incomplete records are not uncommon owing to image artifacts, data corruption or data loss, which compromise the potential of multi-modal acquisitions. Moreover, independently of how well an imaging system is, the performance of the imaging equipment usually comes to a certain limit through different physical devices. Additional interferences arise (particularly for medical imaging systems), for example, limited acquisition times, sophisticated and costly equipment and patients with severe medical conditions, which also cause image degradation. The acquisitions can be considered as the degraded version of the original high-quality images. In this dissertation, we explore the problems of image super-resolution and cross-modality synthesis for one Magnetic Resonance Imaging (MRI) modality from an image of another MRI modality of the same subject using an image synthesis framework for reconstructing the missing/complex modality data. We develop models and techniques that allow us to connect the domain of source modality data and the domain of target modality data, enabling transformation between elements of the two domains. In particular, we first introduce the models that project both source modality data and target modality data into a common multi-modality feature space in a supervised setting. This common space then allows us to connect cross-modality features that depict a relationship between each other, and we can impose the learned association function that synthesizes any target modality image. Moreover, we develop a weakly-supervised method that takes a few registered multi-modality image pairs as training data and generates the desired modality data without being constrained a large number of multi-modality images collection of well-processed (\textit{e.g.}, skull-stripped and strictly registered) brain data. Finally, we propose an approach that provides a generic way of learning a dual mapping between source and target domains while considering both visually high-fidelity synthesis and task-practicability. We demonstrate that this model can be used to take any arbitrary modality and efficiently synthesize the desirable modality data in an unsupervised manner. We show that these proposed models advance the state-of-the-art on image super-resolution and cross-modality synthesis tasks that need jointly processing of multi-modality images and that we can design the algorithms in ways to generate the practically beneficial data to medical image analysis.

Data similarity search is widely regarded as a classic topic in the realms of computer vision, machine learning and data mining. Providing a certain query, the retrieval model sorts out the related candidates in the database according to their similarities, where representation learning methods and nearest-neighbour search apply. As matching data features in Hamming space is computationally cheaper than in Euclidean space, learning to hash and binary representations are generally appreciated in modern retrieval models. Recent research seeks solutions in deep learning to formulate the hash functions, showing great potential in retrieval performance. In this thesis, we gradually extend our research topics and contributions from unsupervised single-modal deep hashing to supervised cross-modal hashing _nally zero-shot hashing problems, addressing the following challenges in deep hashing. First of all, existing unsupervised deep hashing works are still not attaining leading retrieval performance compared with the shallow ones. To improve this, a novel unsupervised single-modal hashing model is proposed in this thesis, named Deep Variational Binaries (DVB). We introduce the popular conditional variational auto-encoders to formulate the encoding function. By minimizing the reconstruction error of the latent variables, the proposed model produces compact binary codes without training supervision. Experiments on benchmarked datasets show that our model outperform existing unsupervised hashing methods. The second problem is that current cross-modal hashing methods only consider holistic image representations and fail to model descriptive sentences, which is inappropriate to handle the rich semantics of informative cross-modal data for quality textual-visual search tasks. To handle this problem, we propose a supervised deep cross-modal hashing model called Textual-Visual Deep Binaries (TVDB). Region-based neural networks and recurrent neural networks are involved in the image encoding network in order to make e_ective use of visual information, while the text encoder is built using a convolutional neural network. We additionally introduce an e_cient in-batch optimization routine to train the network parameters. The proposed mode successfully outperforms state-of-the-art methods on large-scale datasets. Finally, existing hashing models fail when the categories of query data have never been seen during training. This scenario is further extended into a novel zero-shot cross-modal hashing task in this thesis, and a Zero-shot Sketch-Image Hashing (ZSIH) scheme is then proposed with graph convolution and stochastic neurons. Experiments show that the proposed ZSIH model signi_cantly outperforms existing hashing algorithms in the zero-shot retrieval task. Experiments suggest our proposed and novel hashing methods outperform state-of-the-art researches in single-modal and cross-modal data retrieval.

Thesis (O.T.D.)--Boston University PLEASE NOTE: Boston University Libraries did not receive an Authorization To Manage form for this thesis or dissertation. It is therefore not openly accessible, though it may be available by request. If you are the author or principal advisor of this work and would like to request open access for it, please contact us at open-help@bu.edu. Thank you.<br>Developing cultural awareness and effectiveness is critical to meaningful and successful occupational therapy practice and of concern to occupational therapists worldwide (World Federation of Occupational Therapists, 2010). Occupational therapy graduates, not fully representative of the demographics of the populations they will meet clinically, must be able to work effectively with individuals and systems and acquire the ability to understand the interconnectedness of culture and its influence on socioeconomics, health, wellness, specific diagnostic conditions and health disparities (Black & Wells, 2007). This doctoral project argues that the term cultural effectiveness communicates a more collaborative process and realistic outcome than the term cultural competence. Based on a review of theories and effective approaches for teaching and learning about cultural effectiveness this doctoral project includes a review of the St. Catherine University Occupational Science and Occupational Therapy Masters of Arts in Occupational Therapy curriculum. Suggestions are recommended for content and outcome measures to enhance occupational therapy students' reflective, client-centered, culturally effective practice.

This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections.<br>Thesis: S.M., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2019<br>Cataloged from student-submitted PDF version of thesis.<br>Includes bibliographical references (pages 71-81).<br>Deep learning is one of the most prominent machine learning techniques nowadays, being the state-of-the-art on a broad range of applications in computer vision, natural language processing, and speech and audio processing. Current deep learning models, however, rely on signicant amounts of supervision for training to achieve exceptional performance. For example, commercial speech recognition systems are usually trained on tens of thousands of hours of annotated data, which take the form of audio paired with transcriptions for training acoustic models, collections of text for training language models, and (possibly) linguist-crafted lexicons mapping words to their pronunciations. The immense cost of collecting these resources makes applying state-of-the-art speech recognition algorithm to under-resourced languages infeasible. In this thesis, we propose a general framework for mapping sequences between speech and text. Each component in this framework can be trained without any labeled data so the entire framework is unsupervised. We first propose a novel neural architecture that learns to represent a spoken word in an unlabeled speech corpus as an embedding vector in a latent space, in which word semantics and relationships between words are captured. In parallel, we train another latent space that captures similar information about written words using a corpus of unannotated text. By exploiting the geometrical properties exhibited in the speech and text embedding spaces, we develop an unsupervised learning algorithm that learns a cross-modal alignment between speech and text. As an example application of the learned alignment, we develop a unsupervised speech-to-text translation system using only unlabeled speech and text corpora.<br>This work was supported in part by iFlytek<br>by Yu-An Chung.<br>S.M.<br>S.M. Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science

This study examined how computer technology had an effect on a Year 9 visual arts education class with regard to the Western Australian four Arts Learning Outcomes (WA 4ALO). The research was administered concurrently with a learning approach called Cross-Linked Leaming (CLL) (Zervos, 1997), which consisted of three components: (1) the subject (e.g., visual arts education in relation to the WA 4ALO); (2) the learner (i.e., a target group and how they learn); and (3) the tool (e.g., computer technology). This study addressed the problem of how to promote learning in visual arts education, especially with visual arts theory. Historically, students have preferred to make art than to study art theory subjects such as art history and art criticism/response. Furthermore, many students may have found, traditional ways of learning theory to be less engaging and stimulating than making art. For this study, a sample consisted of 19 female students from an independent secondary school in Perth for one school term. The students were divided into three groups for the three data collections methods: (l) the whole class completed pre and post-questionnaires; (2) five pairs of students participated in pre- and post-interviews; and, (3) nine students' art portfolios representative of different levels, of achievement, that were analysed at the end of the school term. The methodology was action research. Data was collected and interpreted to answer the primary research question through four sub-questions as follows: (1.0) What was the effect of CLL on students; (1.1) What were students' attitudes towards CLL; (1.2) What skills did students require for CLL; (1.3) What knowledge did students exercise with CLL; and (1.4) What were students' preferences for !earning with CLL?" The results showed that the three components of CLL had a predominately positive effect upon most students in terms of their attitudes, skills, knowledge, and preferences. Furthermore, the students showed a first preference for learning visual arts theory in a CLL framework reflecting a social constructivist and student-centered way of learning that included using, computers 75% of the-•time for visual arts theory instruction. This thesis demonstrates that CLL is an effective framework for the Year 9 visual arts students who participated in this study.

Les plongements lexicaux sont un composant standard des architectures modernes de traitement automatique des langues (TAL). Chaque fois qu'une avancée est obtenue dans l'apprentissage de plongements lexicaux, la grande majorité des tâches de traitement automatique des langues, telles que l'étiquetage morphosyntaxique, la reconnaissance d'entités nommées, la recherche de réponses à des questions, ou l'inférence textuelle, peuvent en bénéficier. Ce travail explore la question de l'amélioration de la qualité de plongements lexicaux monolingues appris par des modèles prédictifs et celle de la mise en correspondance entre langues de plongements lexicaux contextuels créés par des modèles préentraînés de représentation de la langue comme ELMo ou BERT.Pour l'apprentissage de plongements lexicaux monolingues, je prends en compte des informations globales au corpus et génère une distribution de bruit différente pour l'échantillonnage d'exemples négatifs dans word2vec. Dans ce but, je précalcule des statistiques de cooccurrence entre mots avec corpus2graph, un paquet Python en source ouverte orienté vers les applications en TAL : il génère efficacement un graphe de cooccurrence à partir d'un grand corpus, et lui applique des algorithmes de graphes tels que les marches aléatoires. Pour la mise en correspondance translingue de plongements lexicaux, je relie les plongements lexicaux contextuels à des plongements de sens de mots. L'algorithme amélioré de création d'ancres que je propose étend également la portée des algorithmes de mise en correspondance de plongements lexicaux du cas non-contextuel au cas des plongements contextuels<br>Word embeddings are a standard component of modern natural language processing architectures. Every time there is a breakthrough in word embedding learning, the vast majority of natural language processing tasks, such as POS-tagging, named entity recognition (NER), question answering, natural language inference, can benefit from it. This work addresses the question of how to improve the quality of monolingual word embeddings learned by prediction-based models and how to map contextual word embeddings generated by pretrained language representation models like ELMo or BERT across different languages.For monolingual word embedding learning, I take into account global, corpus-level information and generate a different noise distribution for negative sampling in word2vec. In this purpose I pre-compute word co-occurrence statistics with corpus2graph, an open-source NLP-application-oriented Python package that I developed: it efficiently generates a word co-occurrence network from a large corpus, and applies to it network algorithms such as random walks. For cross-lingual contextual word embedding mapping, I link contextual word embeddings to word sense embeddings. The improved anchor generation algorithm that I propose also expands the scope of word embedding mapping algorithms from context independent to contextual word embeddings

This is a comparative study of parental involvement in children’s learning at Key Stage 2. It focuses on mothers and children of white English and Chinese ethnic origin, using a two-group sample which also mixes gender and socio-economic status. The illustrative medium for the study is an official DfES book designed for parents to use with their children to support their school learning. The book, <i>Learning Journey (ages 7-11): a parent’s guide to the primary school curriculum</i> is published in both English and Chinese versions. The research examines how this material is viewed and used by mothers and children in the two cultural groups, and places its analysis in the wider context of UK government policy on education and home-school relations. Three methods were used to collect data for the study: interview, observation and analysis of the chosen illustrative medium. The interviews were with the mothers in the two groups and with DfES and Primary National Strategy team officials, a home-school academic expert, and a representative of a national parental organisation. The mothers and children were observed undertaking activities from the English and Chinese variations of <i>Learning Journey (ages 7-11).</i> All interviews and observation sessions were tape-recorded.  In the latter case this facilitated a subsidiary comparison of the task-related mother-child discourse in the two groups. The two published versions of <i>Learning Journey (ages 7-11) </i>were compared for cultural bias and for disparities arising from translation. The findings reveal both similarities and differences within and across the cultural and socio-economic groups involved in the study. There is also variation in the views of those interviewed. Some of the most striking differences are in the areas of homework, children’s home life, and the way parents approach the task of helping their children with their school work. In the latter case, the discourse analysis shows the impact of culture and parental educational experience.

Pour estimer la répartition de la puissance au sein d’un réacteur nucléaire, il est nécessaire de coupler des modélisations neutroniques et thermohydrauliques. De telles simulations doivent disposer des valeurs sections efficaces homogénéisées à peu de groupes d’énergies qui décrivent les interactions entre les neutrons et la matière. Cette thèse est consacrée à la modélisation des sections efficaces par des techniques académiques innovantes basées sur l’apprentissage machine. Les premières méthodes utilisent les modèles à noyaux du type RKHS (Reproducing Kernel Hilbert Space) et les secondes par réseaux de neurones. La performance d’un modèle est principalement définie par le nombre de coefficients qui le caractérisent (c’est-à-dire l’espace mémoire nécessaire pour le stocker), la vitesse d’évaluation, la précision, la robustesse au bruit numérique, la complexité, etc. Dans cette thèse, un assemblage standard de combustible UOX REP est analysé avec trois variables d’état : le burnup, la température du combustible et la concentration en bore. La taille de stockage des bibliothèques est optimisée en cherchant à maximiser la vitesse et la précision de l’évaluation, tout en cherchant à réduire l’erreur de reconstruction des sections efficaces microscopiques, macroscopiques et du facteur de multiplication infini. Trois techniques d’approximation sont étudiées. Les méthodes de noyaux, qui utilisent le cadre général d’apprentissage machine, sont capables de proposer, dans un espace vectoriel normalisé, une grande variété de modèles de régression ou de classification. Les méthodes à noyaux peuvent reproduire différents espaces de fonctions en utilisant un support non structuré, qui est optimisé avec des techniques d’apprentissage actif. Les approximations sont trouvées grâce à un processus d’optimisation convexe facilité par "l’astuce du noyau”. Le caractère modulaire intrinsèque de la méthode facilite la séparation des phases de modélisation : sélection de l’espace de fonctions, application de routines numériques, et optimisation du support par apprentissage actif. Les réseaux de neurones sont des méthodes d’approximation universelles capables d’approcher de façon arbitraire des fonctions continues sans formuler de relations explicites entre les variables. Une fois formés avec des paramètres d’apprentissage adéquats, les réseaux à sorties multiples (intrinsèquement parallélisables) réduisent au minimum les besoins de stockage tout en offrant une vitesse d’évaluation élevée. Les stratégies que nous proposons sont comparées entre elles et à l’interpolation multilinéaire sur une grille cartésienne qui est la méthode utilisée usuellement dans l’industrie. L’ensemble des données, des outils, et des scripts développés sont disponibles librement sous licence MIT<br>Modern nuclear reactors utilize core calculations that implement a thermo-hydraulic feedback requiring accurate homogenized few-group cross sections.They describe the interactions of neutrons with matter, and are endowed with the properties of smoothness and regularity, steaming from their underling physical phenomena. This thesis is devoted to the modeling of these functions by industry state-of-theart and innovative machine learning techniques. Mathematically, the subject can be defined as the analysis of convenient mapping techniques from one multi-dimensional space to another, conceptualize as the aggregated sum of these functions, whose quantity and domain depends on the simulations objectives. Convenient is intended in terms of computational performance, such as the model’s size, evaluation speed, accuracy, robustness to numerical noise, complexity,etc; always with respect to the engineering modeling objectives that specify the multidimensional spaces of interest. In this thesis, a standard UO₂ PWR fuel assembly is analyzed for three state-variables, burnup,fuel temperature, and boron concentration.Library storage requirements are optimized meeting the evaluation speed and accuracy targets in view of microscopic, macroscopic cross sections and the infinite multiplication factor. Three approximation techniques are studied: The state-of-the-art spline interpolation using computationally convenient B-spline basis, that generate high order local approximations. A full grid is used as usually donein the industry. Kernel methods, that are a very general machine learning framework able to pose in a normed vector space, a large variety of regression or classification problems. Kernel functions can reproduce different function spaces using an unstructured support,which is optimized with pool active learning techniques. The approximations are found through a convex optimization process simplified by the kernel trick. The intrinsic modular character of the method facilitates segregating the modeling phases: function space selection, application of numerical routines and support optimization through active learning. Artificial neural networks which are“model free” universal approximators able Artificial neural networks which are“model free” universal approximators able to approach continuous functions to an arbitrary degree without formulating explicit relations among the variables. With adequate training settings, intrinsically parallelizable multi-output networks minimize storage requirements offering the highest evaluation speed. These strategies are compared to each other and to multi-linear interpolation in a Cartesian grid, the industry standard in core calculations. The data set, the developed tools, and scripts are freely available under aMIT license

Traditional engineering and science teaching methodology has been to train like-minded students within the discipline of their respective majors. Curriculum time constraints, however, limit the number and nature of out of discipline elective courses. As a result, students are well trained within their respective fields of study but lack the breadth of experience in interacting with other diverse disciplines. Industry, particularly technology-based companies, has observed that solutions to problems have a greater probability of success when all interested parties (purchasing, innovation, marketing, sales, manufacturing, etc.) have input in developing a plan to achieve a desired corporate outcome. It is through this collective action of diverse disciplines that unique solutions are conceived. Many times breakthroughs in innovation and product development occur not through the actions of companies in direct competition but through new entrant companies by modifying technology currently residing in different markets and applications. The breakthrough occurs because the new entrants are not bound by the technology paradigms constraining innovation in their particular market arena. Our goal is to take the diversity lessons gleaned from industry and incorporate them into coursework that creates diverse cross-functional teams such that students learn the benefits of cross-discipline diversity. The College of Business and Technology at ETSU is itself a diverse blend of disciplines (Engineering Technology, Entrepreneurship, Human Nutrition, Marketing, Digital Media, etc) and several graduate and undergraduate courses residing in different departments within the college have intentional programs that encourage cross-discipline enrollment. This action is further facilitated through dual course listings between departments for the same course. Examples of diverse discipline teams will be discussed with attention to outcomes and challenges. Through this diverse cooperative program, students from the technology, business, applied human sciences and digital media disciplines gain a perspective for each other’s expertise and learn to develop teams with diverse skills to meet the increasing challenges for managing business and technology.

Les modèles développés dans cette thèse sont le résultat d'un processus itératif d'analyse et de synthèse entre les théories existantes et nos études réalisées. Plus spécifiquement, nous souhaitons étudier l'apprentissage inter-modal pour la compréhension contextualisée sur les composants des documents à travers le langage et la vision. Cette thèse porte sur l'avancement de la recherche sur l'apprentissage inter-modal et apporte des contributions sur quatre fronts : (i) proposer une approche inter-modale avec des réseaux profonds pour exploiter conjointement les informations visuelles et textuelles dans un espace de représentation sémantique commun afin d'effectuer et de créer automatiquement des prédictions sur les documents multimodaux; (ii) à étudier des stratégies concurrentielles pour s'attaquer aux tâches de classification de documents intermodaux, de récupération basée sur le contenu et de classification few-shot de documents ; (iii) pour résoudre les problèmes liés aux données comme l'apprentissage lorsque les données ne sont pas annotées, en proposant un réseau qui apprend des représentations génériques à partir d'une collection de documents non étiquetés ; et (iv) à exploiter les paramètres d'apprentissage few-shot lorsque les données ne contiennent que peu d’exemples<br>The frameworks developed in this thesis were the outcome of an iterative process of analysis and synthesis between existing theories and our performed studies. More specifically, we wish to study cross-modality learning for contextualized comprehension on document components across language and vision. The main idea is to leverage multimodal information from document images into a common semantic space. This thesis focuses on advancing the research on cross-modality learning and makes contributions on four fronts: (i) to proposing a cross-modal approach with deep networks to jointly leverage visual and textual information into a common semantic representation space to automatically perform and make predictions about multimodal documents (i.e., the subject matter they are about); (ii) to investigating competitive strategies to address the tasks of cross-modal document classification, content-based retrieval and few-shot document classification; (iii) to addressing data-related issues like learning when data is not annotated, by proposing a network that learns generic representations from a collection of unlabeled documents; and (iv) to exploiting few-shot learning settings when data contains only few examples

The rapid development of Internet technologies has resulted in a sharp increase in the number of Internet users who create content online. User-generated content often represents people's opinions, thoughts, speculations and sentiments and is a valuable source of information for companies, organisations and individual users. This has led to the emergence of the field of sentiment analysis, which deals with the automatic extraction and classification of sentiments expressed in texts. Sentiment analysis has been intensively researched over the last ten years, but there are still many issues to be addressed. One of the main problems is the lack of labelled data necessary to carry out precise supervised sentiment classification. In response, research has moved towards developing semi-supervised and cross-domain techniques. Semi-supervised approaches still need some labelled data and their effectiveness is largely determined by the amount of these data, whereas cross-domain approaches usually perform poorly if training data are very different from test data. The majority of research on sentiment classification deals with the binary classification problem, although for many practical applications this rather coarse sentiment scale is not sufficient. Therefore, it is crucial to design methods which are able to perform accurate multiclass sentiment classification. The aims of this thesis are to address the problem of limited availability of data in sentiment analysis and to advance research in semi-supervised and cross-domain approaches for sentiment classification, considering both binary and multiclass sentiment scales. We adopt graph-based learning as our main method and explore the most popular and widely used graph-based algorithm, label propagation. We investigate various ways of designing sentiment graphs and propose a new similarity measure which is unsupervised, easy to compute, does not require deep linguistic analysis and, most importantly, provides a good estimate for sentiment similarity as proved by intrinsic and extrinsic evaluations. The main contribution of this thesis is the development and evaluation of a graph-based sentiment analysis system that a) can cope with the challenges of limited data availability by using semi-supervised and cross-domain approaches b) is able to perform multiclass classification and c) achieves highly accurate results which are superior to those of most state-of-the-art semi-supervised and cross-domain systems. We systematically analyse and compare semi-supervised and cross-domain approaches in the graph-based framework and propose recommendations for selecting the most pertinent learning approach given the data available. Our recommendations are based on two domain characteristics, domain similarity and domain complexity, which were shown to have a significant impact on semi-supervised and cross-domain performance.

In this thesis I set out to explore the area of specific learning difficulties, an area of conflicting theories, understandings, policies and provision. The purpose of this comparative research in such a heavily debated area was to illuminate the commonalities and differences which can be observed across countries. Comparative research in a policy related area has a long tradition. However, Greece and Scotland provided two different cultural and educational backgrounds which made the comparisons particularly interesting. The nature of, as well as the provision for, specific learning difficulties is investigated in this research through the eyes of those involved. The perceptions of policy agents, head teachers, learning support teachers, mainstream teachers, parents and pupils, as well as the underlying constructs evident in policy documentation and literature in both countries, provided the data on which this thesis was based. This thesis seeks to compare current policies and provision in Scotland and Greece, to investigate the discrepancies between policy and provision, to highlight the differences in perceptions about the nature of specific learning difficulties among the different groups within and between the countries, and to identify factors which might have influenced these perceptions and the current provision. In addition, as both countries are members of the European Union, the impact that the EU had in forming the current policies or provision is also examined. The case-study schools were selected by policy agents in Scotland and from a list provided by the Ministry of Education in Greece. Case-study pupils were selected by the learning support teachers of the schools selected, or the head teachers using the learning support teachers files. The aim was that no preconceptions held by the researcher about the nature of specific learning difficulties influenced the selection of the case-study schools and/or pupils, consistent with the ethnographic principles of investigation. The data was gathered through semi-structured interview schedules which, although they maintaineda structure, allowed the respondents to play the leading role. The interviews were supported by observation of the case-study pupils, from which examples were drawn to use as exemplification during the interviews. Relevant policy documents and literature, not only those explicitly about specific learning difficulties but also those rather more generally about special educational needs were also studied and compared with the constructs held by professionals and consumers. The findings of this study indicated that culture, societal and educational context had influenced the perceptions of, and the provision for, specific learning difficulties. This was highlighted by the fact that the differences among the various groups within the same country were substantially less distinctive than those between Scotland and Greece. These differences highlighted the `inclusive' Scottish society, supporting the notion of `rights' of individuals, whilst in Greece the attitudes were focused on `exclusion' and the `protective' role of the family. The educational systems also played a significant role; the Greek system is heavily hierarchical, with a prescriptive curriculum based on knowledge and delivered by common-to-all books which focus on the `average' child. In contrast, the Scottish system has been characterised as task-oriented and able to differentiate according to children's needs. In addition, the Scottish curriculum is designed for all pupils, and includes guidelines for 'support for learning' targeted at those with special educational needs. The distinctiveness of the Greek and the Scottish societies and educational systems was reflected in the different understandings of special educational needs. In Scotland, they were seen as a continuum of needs including specific learning difficulties. In relation to specific learning difficulties the location of problems was perceived to be to a large extent within the learning environment and, in conjunction with the dominance of the `rights' discourse, responsibilities were placed explicitly on mainstream and head teachers as well as learning support. The latter's role was perceived as co-operative teaching and consultancy. In Greece, concerns were raised about the system itself and its limitations. Characteristics of this system were the lack of clear responsibility on the part of head teachers, and the lack of co-operation between learning support teachers (regarded as responsible for specific learning difficulties) and mainstream teachers. The construct of special educational needs as set of categories of impairment, the distinctive special and general education systems, the provision for specific learning difficulties in 'special classrooms' and the locus of the problem perceived to be within the child, all reflected the dominant position of the 'medical and charity' discourses in the society. In conclusion, although the aim of the education systems has been stated as being `inclusive' education in both Greece and Scotland, I argue that the two countries are at different points, closer or further apart, from their goal. However, the complexity of the various factors involved in the educational development of the two countries presented in this thesis makes a linear comparison a simplistic one, and hence unsuitable. Nevertheless, as both Greece and Scotland reiterate their objective towards &quot;one school for all&quot;, a goal set also by the EU, the latter's impact in Greece is stronger. EU acts through its role as `expert' and co-ordinator of exchanges and by funding projects to support inclusive education. This comparative research has indicated how studies of this kind can raise the awareness of the impact of characteristics of national societies on an area of education which has common rhetoric ('inclusion') across countries but where practice and provision can look very different `on the ground'.

Visuospatial abilities are considered essential to our interaction with the environment and are involved in many every-day activities (Hegarty & Waller, 2005; Jansen, Wiedenbauer, & Hahn, 2010). A useful way to approach this neuropsychological domain is the global-local paradigm, according to which, people may attend an event using a global processing style, in which they consider the gestalt of a set of stimuli, or a local processing style, in which they focus on details (Förster & Dannenberg, 2010; Navon, 1977; Schooler, 2002). An abundance of research on global versus local processing has revealed preferential processing styles (with a global or local bias) in specific neurodevelopmental disorders, particularly as concerns Autism Spectrum Disorders (ASD) (Caron, Mottron, Dawson, Bertiaume, & Dawson, 2006; Kuschner, Bodner, & Minshew, 2009). Conflicting findings have often emerged in the literature (see for example Van der Hallen, Evers, Brewaeys, Van den Noortgate, & Wagemans, 2015), however, showing that participants with different developmental disorders can process both global and local information, depending on the task requirements and the cognitive domain involved, but in different and atypical ways (Dukette & Stiles, 2001). These results prevent possible generalizations and need to be further explored. Differently, global and local processing styles have never been studied in children with other neurodevelopmental disorders, such as Nonverbal Learning Disabilities (NLD), even though there is evidence to suggest that the issue could be relevant in individuals with NLD as well (Chow & Skuy, 1999). For this reason, cross-task and cross-syndrome comparisons are suggested as the best way to analyze these processing abilities and reveal similarities and differences in global and local processing styles in neurodevelopmental disorders (D’Souza, Booth, Connolly, Happé, & Karmiloff-Smith, 2016). The main aim of this PhD dissertation is to improve our understanding of the role of global and local visuospatial processing in the neuropsychological profile of specific neurodevelopmental disorders, using cross-task and cross-disorder comparisons. Children with ASD without intellectual disability (ID) or NLD were tested in terms of their performance in different domains of visuospatial skills, comparing them with each other and with children who had other neurodevelopmental disorders, such as dyslexia or Attention Deficit Hyperactivity Disorder (ADHD). The assessment focused on visuospatial processing speed, visuo-perceptual and visuo-constructive abilities, visuospatial working memory (VSWM), and their interplay with local and global processing. Based on the modified Block Design Task (BDT) paradigm (Caron et al., 2006), new tasks and stimuli have been devised in order to assess the previously mentioned visuospatial abilities, and four studies have been carried out. Study I aimed to make a cross-task comparison on global-local visuospatial processing in two groups of participants with ASD without ID – with and without a visuospatial peak (–P and –NP) – comparing them with matched typically developing (TD) individuals. The results helped us to clarify the visuospatial profile of the two groups of individuals with ASD, demonstrating the importance of taking specific factors into account (i.e. the visuospatial domains examined and the perceptual reasoning abilities). Participants with ASD-NP performed poorly in all domains, revealing weaker spatial integration abilities in the visuo-perceptual domain and a diminished sensitivity to perceptual coherence in the VSWM, while the ASD-P group used both global and local processing effectively according to the task, and a local bias only emerged in the visuo-constructive task. In agreement with D’Souza and coauthors (2016), our results support the conviction that labelling individuals with ASD as ‘local processors’ is restrictive. They may use both local and global processing styles depending on the demands of the task in hand, the visuospatial domain involved and their cognitive visuospatial functioning. Study II (Chapter 3) aimed to investigate global and local visuospatial processing in children with symptoms of NLD comparing them with children with symptoms of dyslexia and with TD controls. The results showed that children with symptoms of NLD were less accurate in visuo-constructive tasks, while children with symptoms of dyslexia were only slightly impaired in a visuo-constructive task, but clearly slower in the perceptual task. Children with symptoms of NLD were less able to benefit from different levels of coherence of the stimuli, probably as a consequence of their less flexible and efficient visuospatial processes (Mammarella, & Cornoldi, 2005). In particular, the global dominance mechanism (Navon, 1977) made it more complicated for the group with symptoms of NLD to switch from a global to a local processing, which was needed to complete the visuo-constructive task correctly. After investigating the issue of global and local visuospatial processing separately for ASD without ID and NLD, the aim of Study III (Chapter 4) was to draw a cross-disorders comparison, highlighting similarities and differences across three clinical profiles - ASD without ID, NLD and ADHD - as compared with TD controls. Our results revealed different visuospatial profiles for the groups considered, and suggested the utility of manipulating the coherence of stimuli to investigate visuospatial skills. Marked deficit in all the visuospatial domains emerged for the group with NLD, confirming that impairments in the visuospatial domain are core and distinctive symptoms of this disorder (Cornoldi, Mammarella, & Fine, 2016; Semrud-Clikeman, Walkowiak, Wilkinson, & Christopher, 2010). In addition, difficulty in integrating local configurations in a coherent whole emerged for the visuo-perceptual domain. A heterogeneous profile emerged for children with ADHD, which showed, consistently with previous studies, impairment in the visuospatial processing speed domain and in VSWM (Martinussen, Hayden, Hogg-Johnson, & Tannock, 2005; Weigard & Huang-Pollock, 2017). Moreover, these participants presented some difficulties in visuo-constructive abilities when they had to deal with global configurations, while they performed normally in visuo-perceptual task. Differently, participants with ASD performed normally in all the examined domains, using effectively both global and local visuospatial processes, with the sole exception of the visuo-constructive task in which this group showed slower response times and a diminished sensitivity to perceptual coherence (Caron et al., 2006; Shah & Frith, 1993). Finally, since individuals with NLD and those with High Functioning Autism or Asperger Syndrome (DSM-IV TR, American Psychiatric Association [APA], 2000) are often confused, Study IV (Chapter 5) included a further comparison between ASD and NLD. Visuo-constructive abilities and VSWM were investigated in a subgroup of participants with ASD without ID and without a visuospatial peak (ASD-NP) and in a group with NLD. Thus, Study IV aimed to analyze whether ASD-NP – though not representative of the ASD without ID population as a whole– shared any characteristics with the NLD group. Once again, our results differentiate the visuospatial profile of children with NLD from that of children with ASD. The former group showed an impaired performance in all the domains examined affecting both global and local levels of processing. The ASD group had a more heterogeneous profile, with normal performance in VSWM and in the drawing of a complex figure, slower response times in the segmented condition of visuoconstructive BDT and a more local and fragmented drawing style in the recall of a complex figure. Here again, local bias affected the performance of participants with ASD in tasks demanding visuoconstructive skills that specifically involved combining parts to form a single whole (Simic, Khan, & Rovet, 2013). General conclusions derived from the main findings of the four studies, and both clinical and educational implications will be thus highlighted in the final chapter of this dissertation. To conclude, investigating visuospatial abilities and global-local processing in individuals with neurodevelopmental disorders offer crucial insight for the analysis of the strengths and weaknesses of the clinical profiles examined and for their differential diagnosis. There is still space for further research on the domains of visuospatial abilities, and on the general neuropsychological functioning of children with different neurodevelopmental disorders. This dissertation was an effort to raise and clarify some points, however other questions remain open and will require further studies.<br>Le abilità visuospaziali sono un insieme di abilità considerate essenziali nell’interazione con l’ambiente e sono coinvolte in numerose attività quotidiane (Hegarty & Waller, 2005; Jansen, Wiedenbauer, & Hahn, 2010). Il paradigma di elaborazione globale-locale (Navon, 1977) costituisce un utile approccio per studiare questo dominio neuropsicologico. Secondo tale paradigma le persone possono percepire un evento usando uno stile di elaborazione globale, per cui considerano la gestalt di un insieme di stimoli, o uno stile di elaborazione locale, per cui si focalizzano sui dettagli (Förster & Dannenberg, 2010; Navon, 1977; Schooler, 2002). Numerose ricerche sull’elaborazione globale-locale hanno rivelato l’uso preferenziale di uno stile di elaborazione (con un bias globale o locale) in specifici disturbi del neurosviluppo, in particolare riguardo al disturbo dello spettro dell’autismo (ASD) (Caron, Mottron, Dawson, Bertiaume, & Dawson, 2006; Kuschner, Bodner, & Minshew, 2009). Tuttavia, risultati conflittuali sono spesso emersi in letteratura (vedi Van der Hallen, Evers, Brewaeys, Van den Noortgate, & Wagemans, 2015) e mostrano come i partecipanti con differenti disturbi dello sviluppo possono elaborare sia informazioni locali che globali, a seconda delle richieste del compito e del dominio cognitivo coinvolto, ma in modi differenti e atipici (Dukette & Stiles, 2001). Questi risultati prevengono possibili generalizzazioni e necessitano di essere ulteriormente esplorati. Al contrario, gli stili di elaborazione globale-locale non sono mai stati studiati in bambini con altri disturbi del neurosviluppo, come il disturbo dell’apprendimento nonverbale (NLD), nonostante evidenze abbiano suggerito che questi aspetti possano essere rilevanti anche nell’NLD (Chow & Skuy, 1999). Per tale ragione, confronti tra differenti disturbi del neurosviluppo e attraverso l’uso di diversi compiti vengono suggeriti come il metodo migliore per analizzare queste abilità ed evidenziare similitudini o differenze nell’uso degli stili di elaborazione (D’Souza, Booth, Connolly, Happé, & Karmiloff-Smith, 2016). L'obiettivo principale della presente tesi di Dottorato è quello di migliorare la nostra comprensione del ruolo dell’elaborazione visuospaziale globale-locale nel profilo neuropsicologico di specifici disturbi del neurosviluppo, attraverso la comparazione di diversi disturbi e l’uso di prove differenti. Sono state indagate le prestazioni di partecipanti con ASD senza disabilità intellettiva (ID) o NLD in diversi domini di abilità visuospaziali, confrontandoli tra loro e con bambini aventi altri disturbi del neurosviluppo, come la dislessia o il deficit di attenzione/iperattività (ADHD). L’assessment si è concentrato sull’indagine della velocità di elaborazione visuospaziale, delle abilità visuo-percettive, visuo-costruttive e di memoria di lavoro visuospaziale (VSWM). È stata inoltre indagata l’interazione tra le performance in questi domini e l'elaborazione globale-locale. Sulla base del paradigma modificato di disegno con cubi (BDT) (Caron et al., 2006), sono stati elaborati nuovi compiti e stimoli per valutare le abilità visuospaziali menzionate. In particolare, sono stati condotti quattro studi. Lo Studio I ha indagato gli stili di elaborazione visuospaziale globale-locale in due gruppi di partecipanti con ASD senza ID - con e senza un picco visuospaziale (-P e -NP) - confrontandoli con individui a sviluppo tipico (TD). I risultati hanno permesso di chiarire il profilo visuospaziale dei due gruppi di partecipanti con ASD, dimostrando l’importanza di tenere in considerazione fattori specifici (come i domini di abilità visuospaziali esaminati e le abilità di ragionamento percettivo dei partecipanti). I partecipanti con ASD-NP hanno ottenuto scarsi risultati in tutti i domini, mostrando inferiori capacità di integrazione spaziale nel dominio visuo-percettivo e una ridotta sensibilità alla coerenza percettiva nella VSWM, mentre il gruppo ASD-P ha utilizzato entrambe le strategie di elaborazione globale e locale in modo efficace in base al compito e un bias locale è emerso solo nel compito visuo-costruttivo. In accordo con D'Souza et al. (2016), i nostri risultati sostengono la convinzione che etichettare gli individui con ASD come "local processors" sia restrittivo. Infatti, essi possono utilizzare entrambi gli stili di elaborazione locale e globale a seconda delle richieste del compito, del dominio visuospaziale coinvolto e del loro funzionamento cognitivo di tipo visuospaziale. Lo studio II (Capitolo 3) ha indagato l'elaborazione visuospaziale globale-locale nei bambini con sintomi di NLD confrontandoli con bambini con sintomi di dislessia e con TD. I risultati hanno mostrato un’accuratezza inferiore per i bambini con sintomi di NLD nel compito visuo-costruttivo, mentre i bambini con sintomi di dislessia hanno mostrato lievi difficoltà nel compito visuo-costruttivo e una chiara lentezza in quello viuso-percettivo. Inoltre, i bambini con sintomi di NLD si sono mostrati meno in grado di beneficiare dei diversi livelli di coerenza degli stimoli, probabilmente come conseguenza dei loro processi visuospaziali meno flessibili ed efficienti (Mammarella & Cornoldi, 2005). In particolare, il meccanismo di dominanza globale (Navon, 1977) ha reso più complicato per il gruppo con sintomi di NLD il passaggio dall’elaborazione globale a quella locale, necessario per completare correttamente il compito visuo-costruttivo. Dopo aver esaminato l’elaborazione visuospaziale globale-locale separatamente per ASD senza ID e NLD, lo scopo dello Studio III (Capitolo 4) era quello di effettuare un confronto tra disturbi, evidenziando somiglianze e differenze tra tre profili clinici - ASD senza ID, NLD e ADHD - rispetto ai TD. I nostri risultati hanno rivelato diversi profili visuospaziali per i gruppi considerati e suggerito l'utilità di manipolare la coerenza degli stimoli per l’indagine di tali abilità. Per il gruppo con NLD è emerso un deficit marcato in tutti i domini visuospaziali, a conferma che le difficoltà in tale dominio costituiscono sintomi fondamentali e distintivi di questo disturbo (Cornoldi, Mammarella & Fine, 2016, Semrud-Clikeman, Walkowiak, Wilkinson e Christopher, 2010). Inoltre, per il dominio visuo-percettivo è emersa la difficoltà di integrare le configurazioni locali in un insieme coerente. Per il gruppo con ADHD si è evidenziato un profilo eterogeneo, i partecipanti con tale diagnosi hanno mostrato, in linea con gli studi precedenti, un deficit nel dominio di velocità di elaborazione visuospaziale e nella VSWM (Martinussen, Hayden, Hogg-Johnson & Tannock, 2005, Weigard & Huang-Pollock, 2017). Inoltre, questi partecipanti hanno presentato alcune difficoltà nelle abilità viso-costruttive quando dovevano ricostruire configurazioni globali, mentre sono emerse abilità visuo-percettive in norma. Diversamente, i partecipanti con ASD hanno mostrato prestazioni in norma in tutti i domini esaminati, utilizzando efficacemente processi visuospaziali globali e locali, con l'unica eccezione del compito visuo-costruttivo in cui questo gruppo ha mostrato tempi di risposta più lenti e una sensibilità ridotta alla coerenza percettiva (Caron et al., 2006; Shah & Frith, 1993). Infine, considerato che i profili di individui con NLD e con autismo ad alto funzionamento o sindrome di Asperger (DSM-IV TR, American Psychiatric Association [APA], 2000) sono spesso confusi, nello Studio IV (Capitolo 5) è stato proposto un ulteriore confronto tra ASD e NLD. Le abilità visuo-costruttive e la VSWM sono state studiate in un sottogruppo di partecipanti con ASD senza ID e senza picco visuospaziale (ASD-NP) e in partecipanti con NLD. Lo scopo era quello di analizzare se il gruppo con ASD-NP - sebbene non rappresentativo dell'intera popolazione con ASD senza ID – condividesse o meno caratteristiche con il gruppo NLD. Ancora una volta, i nostri risultati hanno permesso di differenziare il profilo visuospaziale dei bambini con NLD da quello dei bambini con ASD. Il primo gruppo ha mostrato prestazioni deficitarie in tutti i domini esaminati sia per il livello di elaborazione locale sia per quello globale. Il gruppo con ASD ha mostrato invece un profilo più eterogeneo, con prestazioni in norma nella VSWM e nel disegno di una figura complessa, tempi di risposta più lenti nella condizione segmentata della prova visuo-costruttiva e uno stile di disegno locale e frammentato nel disegno a memoria di una figura complessa. Anche qui, il bias locale ha influenzato le prestazioni dei partecipanti con ASD in compiti che richiedevano competenze visuo-costruttive e nello specifico di combinare le parti per formare un unico insieme (Simic, Khan, & Rovet, 2013). Infine, le conclusioni generali derivate dai principali risultati dei quattro studi e le loro implicazioni cliniche ed educative sono state evidenziate nel capitolo conclusivo della presente tesi. Per concludere, l'analisi delle capacità visuospaziali e l'elaborazione globale-locale in individui con disturbi del neurosviluppo offrono una visione cruciale per l'analisi dei punti di forza e di debolezza dei profili clinici esaminati e per la loro diagnosi differenziale. C'è ancora molto spazio per ulteriori ricerche sulle capacità visuospaziali e sul funzionamento neuropsicologico generale dei bambini con diversi disturbi del neurosviluppo. La presente tesi ha avuto l’obiettivo di sollevare e chiarire alcuni punti, ma altre domande restano aperte e richiederanno ulteriori studi.

This study explored the effects on self-concept of a Cross Age Learning Program (CALP) where students in grades four to six taught mathematics to students in grades one to three. All 27 participating students were judged, by their teachers, to be having difficulty in mathematics but were capable of achieving in a regular environment. These children were divided into four groups, Older Learning Partners (OLPs), Younger Learning Partners (YLPs), Older Non Learning Partners (ONLPs), and Younger Non Learning Partners (YNLPs). The OLPs and the nPs participated in learning sessions where each OLP was trained and then taught basic math to a YLP for approximately four months. The ONLPs and the YNLPs did not participate in the program. A self-concept measure, the Self Description Questionnaire-I (SDQ-I, Marsh, 1990) was administered three times to all students: before the program began, four and a half weeks into the learning sessions and then four weeks after that. The scale scores were compared for the two older groups (OLPs and ONLPs), the two younger groups (YLPs and YNLPs) and for the Learning Partners versus the Non Learning Partners (LPs and NLPs). (Abstract shortened by UMI.)

This thesis alms to develop a comprehensive understanding of cross-project learning in multiple-project environments. Cross-project learning is the process through which technologies are transferred and reused within organisations. Recent years have seen a growing interest in cross-project learning. However, research in this area has emphasised the rational, classical approach to crossproject learning. Also, the majority of research on cross-project learning has largely been on the automobile industry in Japan and the USA. Thirdly, research in this field has failed to assess the impact that cross-project learning has had on other organisational processes in product development. The conclusions of these studies are context-specific, fragmented and lack any critical assessment of the process of introducing cross-project learning. This study argues that a rather different approach to cross-project learning is needed. A three-level analysis is applied in the present study that highlights operational, dysfunctional and strategic aspects in cross-project learning. The empirical core of the research is the evidence from three in-depth case studies conducted in the Israeli electronics defence industry. Three different approaches to cross-project learning have been identified at the operational level, offering organisational mechanisms and managerial practices that have not previously been reported. At the dysfunctional operations level, the study reveals that the introduction of innovations in cross-project learning has impacted the past harmony between expertise development and knowledge management practices. The findings suggest that this harmony has broken down while the knowledge management and expertise development practices have been further transformed and developed. Lastly, at the strategic level of analysis, two potential cross-project learning strategies have been detected: exploit product success and design to reuse. A contingency model that emphasises the evolutionary development path of 'modes of reusability', subject to the 'strategic development' of the studied companies, concludes this study.

xviii, 216 p. : ill. (some col.)<br>The focus of this dissertation is on how language experience shapes perception of a non-native prosodic contrast. In Tokyo Japanese, fundamental frequency (F0) peak and fall are acoustic cues to lexically contrastive pitch patterns, in which a word may be accented on a particular syllable or unaccented (e.g., tsúru 'a crane', tsurú 'a vine', tsuru 'to fish'). In English, lexical stress is obligatory, and it may be reinforced by F0 in higher-level prosodic groupings. Here I investigate whether English listeners can attend to F0 peaks as well as falls in contrastive pitch patterns and whether training can facilitate the learning of prosodic categories. In a series of categorization and discrimination experiments, where F0 peak and fall were manipulated in one-word utterances, the judgments of prominence by naïve English listeners and native Japanese listeners were compared. The results indicated that while English listeners had phonetic sensitivity to F0 fall in a same-different discrimination task, they could not consistently use the F0 fall to categorize F0 patterns. The effects of F0 peak location and F0 fall on prominence judgments were always larger for Japanese listeners than for English listeners. Furthermore, the interaction between these acoustic cues affected perception of the contrast by Japanese, but not English, listeners. This result suggests that native, but not non-native, listeners have complex and integrated processing of these cues. The training experiment assessed improvement in categorization of Japanese pitch patterns with exposure and feedback. The results suggested that training improved identification of the accented patterns, which also generalized to new words and new contexts. Identification of the unaccented pattern, on the other hand, showed no improvement. Error analysis indicated that native English listeners did not learn to attend specifically to the lack of the F0 fall. To conclude, language experience influences perception of prosodic categories. Although there is some sensitivity to F0 fall in non-native listeners, they rely mostly on F0 peak location in language-like tasks such as categorization of pitch patterns. Learning of new prosodic categories is possible. However, not all categories are learned equally well, which suggests that first language attentional biases affect second language acquisition in the prosodic domain.<br>Committee in charge: Susan Guion Anderson, Chairperson; Melissa A. Redford, Member; Vsevolod Kapatsinki, Member; Kaori Idemaru, Outside Member

In recent years, many machine learning algorithms have been developed and widely applied in various applications. However, most of them have considered the data distributions of the training and test datasets to be similar. This thesis concerns on the decrease of generalization ability in a test dataset when the data distribution is different from that of the training dataset. As labels may be unavailable in the test dataset in practical applications, we follow the effective approach of unsupervised domain adaptation and propose distribution alignment methods to improve the generalization ability of models learned from the training dataset in the test dataset. To solve the problem of joint distribution alignment without target labels, we propose a new criterion of domain-shared group sparsity that is an equivalent condition for equal conditional distribution. A domain-shared group-sparse dictionary learning model is built with the proposed criterion, and a cross-domain label propagation method is developed to learn a target-domain classifier using the domain-shared group-sparse representations and the target-specific information from the target data. Experimental results show that the proposed method achieves good performance on cross-domain face and object recognition. Moreover, most distribution alignment methods have not considered the difference in distribution structures, which results in insufficient alignment across domains. Therefore, a novel graph alignment method is proposed, which aligns both data representations and distribution structural information across the source and target domains. An adversarial network is developed for graph alignment by mapping both source and target data to a feature space where the data are distributed with unified structure criteria. Promising results have been obtained in the experiments on cross-dataset digit and object recognition. Problem of dataset bias also exists in human pose estimation across datasets with different image qualities. Thus, this thesis proposes to synthesize target body parts for cross-domain distribution alignment, to address the problem of cross-quality pose estimation. A translative dictionary is learned to associate the source and target domains, and a cross-quality adaptation model is developed to refine the source pose estimator using the synthesized target body parts. We perform cross-quality experiments on three datasets with different image quality using two state-of-the-art pose estimators, and compare the proposed method with five unsupervised domain adaptation methods. Our experimental results show that the proposed method outperforms not only the source pose estimators, but also other unsupervised domain adaptation methods.

Many healthcare organisations face repetitive incidents because organisations tend to fail to learn from the past. Learning from incidents (LFI) in healthcare is a process through which healthcare professionals and the organisation as a whole seek to understand adverse events that have taken place. The LFI process consists of five main steps: data acquisition, investigation and analysis, planning interventions, implementing interventions, and evaluations. In order to reduce the reoccurrence of incidents, it is important that LFI processes are improved. As a prerequisite, it is necessary to gain insight into the steps of the LFI process to identify hindrances (bottlenecks) and mitigate them. This thesis is a broad comparative study of the LFI processes in Dutch and Swedish healthcare systems. Cross-border comparisons between LFI systems can support mutual learning, and consequently lead to improvements of healthcare organisations’ learning processes. The study consists of an analysis of Swedish and Dutch legislation, national healthcare inspectorates, and hospitals’ learning from incidents processes. Legislation was analysed through a (legal) documentation study. Healthcare inspectorates’ practices in LFI were analysed by a combination of documentation studies, and by conducting interviews with one Dutch inspector, one Swedish inspector, and one Swedish development strategist. For analysis of hospitals’ LFI processes, a questionnaire and interview study with fourteen Dutch and eleven Swedish hospitals were conducted. Analysis of these processes was done at the hand of a number of quality statements developed based on a literature study. The main differences between how the two countries’ learn from incidents are in data acquisition, and investigation and analysis. The Netherlands have various reporting systems, as well as diversity in incident investigation methods. Sweden has more uniformity in these matters. Moreover, Sweden has a national system for sharing lessons learned between hospitals, which can benefit the learning process on a national level. The Netherlands currently does not have such a system. Sweden and the Netherlands have similar strengths and weaknesses in LFI. Both countries have accessible data acquisition systems, and it does not take much time to report incidents. There are however significant disparities between incidents and sentinel events in both countries in the quality of investigations and analyses, planning of interventions and implementation of interventions. The implementation and evaluation phases are also regarded to have the lowest quality, based on analysis of the quality statements. Dutch and Swedish legislation and the supervision of the healthcare inspectorates only cover these last two phases to a limited extent. Requirements with respect to incidents are also only formulated to a limited extent (except data acquisition), which may explain the significant difference of quality when compared to sentinel events. There are resemblances between the scopes of the legal frameworks and inspectorates, and the LFI processes in hospitals. There is therefore reason to believe that hospitals typically do not excel above what is required by legislation or by the healthcare inspectorates.<br>I många vårdorganisationer upprepar sig incidenter eftersom organisationer tenderar att misslyckas med att lära sig från incidenter. Att lära från incidenter (LFI) inom hälso- och sjukvården är en process genom vilket vårdpersonal och organisationen som helhet försöker förstå incidenter som har ägt rum. LFI-processen består av fem huvudsteg: datainsamling, utredning och analys, planering av åtgärder, implementering av åtgärder, och utvärderingar. För att minska upprepande av incidenter är det viktigt att LFIprocesser förbättras. Det här examensarbetet är en jämförande studie av LFI-processerna i holländska och svenska sjukvårdssystem. Gränsöverskridande jämförelser mellan LFI-system kan stödja ömsesidigt lärande och därmed leda till förbättringar av vårdorganisationernas lärande. Studien består av en analys av svensk och holländsk lagstiftning, nationella inspektioner och sjukhusens lärande från incidensprocesser. Lagstiftningen analyserades genom en (juridisk) dokumentationsstudie. Sjukvårdsinspektionernas praxis i LFI analyserades med en kombination av dokumentationsstudier och genom att göra intervjuer med en holländsk inspektör, en svensk inspektör och en svensk utvecklingsstrateg. För analys av sjukhusens LFI-processer genomfördes en enkätstudie och intervjustudie med 14 holländska och 11 svenska sjukhus. Analysen genomfördes med ett kvalitetsindikatorer som är baserade på en litteraturstudie. De viktigaste skillnaderna mellan hur de två länderna lär sig av incidenter är inom datainsamling och incidentutredning. I Nederländerna används många olika rapporteringssystem och utredningsmetoder för händelser. Sverige har mer enhetlighet i dessa frågor. Dessutom har Sverige ett nationellt system för att dela lärdomar mellan sjukhusen, vilket kan gynna lärningsprocessen på nationell nivå. Nederländerna har för närvarande inget liknande system. Sverige och Nederländerna har liknande styrkor och svagheter i LFI. Båda länderna har tillgängliga datainsamlingssystem och det tar inte mycket tid att rapportera incidenter. Det finns betydliga skillnader mellan incidenter och händelser som har medfört allvarliga vårdskador i båda länderna. Detta gäller kvaliteten på utredningar, planering av åtgärder och implementering av årgärder. Implementerings- och utvärderingsfaserna anses ha lägsta kvalitet, baserat på analys av kvalitetsindikatorerna. Holländsk och svensk lagstiftning och tillsynen av inspektionerna täcker dessa två sista faser endast i begränsad utsträckning. Krav på incidenter formuleras också endast i begränsad omfattning (förutom datainsamling), vilket kan förklara skillnaden i kvalitet jämfört med händelser som har medfört en allvarlig vårdskada. Det finns likheter mellan räckvidden av lagstiftningen och inspektionen, och LFIprocesserna på sjukhus i både länder. Det finns därför anledning att tro att sjukhus vanligtvis inte utmärker sig högre än vad som krävs enligt lagstiftningen eller av hälsooch sjukvårdsinspektionerna.

This research was inspired by a pressing question which formed the main aim of the current study--What factors contribute to the differential academic performance of adult learners in the formal setting in Malaysia? It is hoped that by addressing this question, insights obtained may be useful for the Malaysian policy makers in attempting to implement the government's initiative--Malaysia Vision 2020. The current literature informs that in order to achieve the desired goals, Malaysian adult learners, must now more than ever be conscious of the effect of learning values and approaches to learning. Hence, there is a need to develop a more holistic understanding of the interrelated dynamics between learning values and approaches to learning. The current study adopts a transdisciplinary, etic/emic approach, using two culturally sensitive questionnaires, Revised Study Process Questionnaires-2 Factors Malaysia (RSPQ- 2FM) and Learning Values Survey (LVS) on 858 Malay and Chinese adult learners in Malaysia. The study found the significant others can have substantial influence on the 'face value' for both Malay and Chinese adult learners generally, but was more pronounced for the Malay adult learners. This in turn may encourage Malay adult learners to submit to pressure from others in influencing how they perceive the importance of learning and motivation in learning. Because Malay adult learners are constantly driven by external factors to compete with other cultural groups in education or economic achievement, they may tend to avoid challenging tasks such as deeper approaches to learning in order to rapidly achieve their immediate learning goals. Engaging with deep approaches and meaningful learning are effortful and the pressure to save face may result in the likelihood of adopting surface approaches. This coupled with the finding that they do not appreciate the middle way principles as much as the Chinese adult learners suggest that they may be less flexible and/or pragmatic learners. The findings suggest that practice of middle way principles (such as 'Willing to compromise one's own values to suit the situation/issues when I learn') can indeed enhance certain positive learning approaches which implies that Malay adult learners may be disadvantaged in the learning settings due to their lack of appreciation of the middle way principles. It is also interesting to find that Malay adult learners appreciate time factor more than their Chinese counterparts when engaging with Deep Approaches to learning. In contrast, the middle way principle practiced as a way of life by the Chinese culture has made Chinese adult learners more malleable, resulting in a relatively less face conscious cultural group. Being less externally driven and less restrictive, Chinese adult learners are more likely to adopt deep approaches to enhance meaningful learning. In addition, the Chinese culturally ingrained learning approach, Understand and Memorization was found to be more likely to produce positive learning outcome. Unlike their Malay counterparts, Chinese adult learners view work experiences more essential in helping them to engage with Deep Approaches to learning. The above findings are novel and add to previous studies on approaches to learning by introducing the effect of learning values. While previous research has referred to cultural variable in learning, they have not sufficiently explored the effect of culture. Learning values is one significant cultural variable that is considered in the study. The findings underpin the different emphasis placed by the two cultural groups as they engage with professional development activities. It is hoped that by identifying values pertinent to learning in this competitive globalized economy, the study has provided insights for Malaysian policy makers to develop holistic future education plans to assist in achieving Malaysian Vision 2020. Insights gained can also support plans where Malay can be encouraged to become competent global leaders and workers, capable of competing in this knowledge economy.

This research was inspired by a pressing question which formed the main aim of the current study--What factors contribute to the differential academic performance of adult learners in the formal setting in Malaysia? It is hoped that by addressing this question, insights obtained may be useful for the Malaysian policy makers in attempting to implement the government's initiative--Malaysia Vision 2020. The current literature informs that in order to achieve the desired goals, Malaysian adult learners, must now more than ever be conscious of the effect of learning values and approaches to learning. Hence, there is a need to develop a more holistic understanding of the interrelated dynamics between learning values and approaches to learning. The current study adopts a transdisciplinary, etic/emic approach, using two culturally sensitive questionnaires, Revised Study Process Questionnaires-2 Factors Malaysia (RSPQ- 2FM) and Learning Values Survey (LVS) on 858 Malay and Chinese adult learners in Malaysia. The study found the significant others can have substantial influence on the 'face value' for both Malay and Chinese adult learners generally, but was more pronounced for the Malay adult learners. This in turn may encourage Malay adult learners to submit to pressure from others in influencing how they perceive the importance of learning and motivation in learning. Because Malay adult learners are constantly driven by external factors to compete with other cultural groups in education or economic achievement, they may tend to avoid challenging tasks such as deeper approaches to learning in order to rapidly achieve their immediate learning goals. Engaging with deep approaches and meaningful learning are effortful and the pressure to save face may result in the likelihood of adopting surface approaches. This coupled with the finding that they do not appreciate the middle way principles as much as the Chinese adult learners suggest that they may be less flexible and/or pragmatic learners. The findings suggest that practice of middle way principles (such as 'Willing to compromise one's own values to suit the situation/issues when I learn') can indeed enhance certain positive learning approaches which implies that Malay adult learners may be disadvantaged in the learning settings due to their lack of appreciation of the middle way principles. It is also interesting to find that Malay adult learners appreciate time factor more than their Chinese counterparts when engaging with Deep Approaches to learning. In contrast, the middle way principle practiced as a way of life by the Chinese culture has made Chinese adult learners more malleable, resulting in a relatively less face conscious cultural group. Being less externally driven and less restrictive, Chinese adult learners are more likely to adopt deep approaches to enhance meaningful learning. In addition, the Chinese culturally ingrained learning approach, Understand and Memorization was found to be more likely to produce positive learning outcome. Unlike their Malay counterparts, Chinese adult learners view work experiences more essential in helping them to engage with Deep Approaches to learning. The above findings are novel and add to previous studies on approaches to learning by introducing the effect of learning values. While previous research has referred to cultural variable in learning, they have not sufficiently explored the effect of culture. Learning values is one significant cultural variable that is considered in the study. The findings underpin the different emphasis placed by the two cultural groups as they engage with professional development activities. It is hoped that by identifying values pertinent to learning in this competitive globalized economy, the study has provided insights for Malaysian policy makers to develop holistic future education plans to assist in achieving Malaysian Vision 2020. Insights gained can also support plans where Malay can be encouraged to become competent global leaders and workers, capable of competing in this knowledge economy.

The purpose of this thesis was to investigate the potential mechanisms and sites of neural adaptations that mediate cross-limb transfer of strength and motor learning that can occur subsequent to unilateral training. Better understanding of the mechanisms should allow therapeutic benefits of this effect to be assessed. There are two main classes of mechanisms that could contribute to cross-limb transfer. The first is described by the ??bilateral access?? hypothesis, which suggests that neural adaptations induced by training reside in bilaterally projecting motor areas that are accessible to the untrained (ipsilateral) hemisphere during task execution to facilitate performance. According to the alternative ??cross-activation?? hypothesis, activation of the untrained hemisphere during unilateral training leads to adaptations in the untrained hemisphere that cause improved performance with the opposite untrained limb. A series of studies were conducted in this research. We directly tested the cross-activation hypothesis via a reliable twitch interpolation technique involving transcranial magnetic stimulation (TMS). Four-weeks of strength training for the right wrist increased neural drive (from the untrained motor cortex) to the untrained left wrist. The data demonstrate that strength training of one limb can influence the efficacy of corticospinal pathways that project to the opposite untrained limb, consistent with the cross-activation hypothesis. To investigate the contribution of each hemisphere in cross-limb transfer, we applied repetitive TMS (rTMS) to the trained or the untrained motor cortex to disrupt brain processing after unilateral ballistic training. Learning to produce ballistic movements requires optimization of motor drive to the relevant muscles in a way that resembles high-force contractions performed during strength training. Ballistic skill transferred rapidly to the untrained hand and the improved performance was accompanied by bilateral increases in corticospinal excitability. Performance improvement in each hand was specifically suppressed by rTMS of the opposite hemisphere. Thus the motor cortex ipsilateral to the trained hand is critically altered during unilateral training; and neural adaptations within this untrained hemisphere are crucial in cross-limb transfer of ballistic skill. Overall, the data are in agreement with the cross-activation hypothesis for high-force and ballistic tasks, although they do not exclude the potential involvement of bilateral access mechanisms.

See Siu Ching.<br>"Architecture Department, Chinese University of Hong Kong, Master of Architecture Programme 2002-2003, design report."<br>Includes bibliographical references.<br>Chapter 01 --- THE PHENOMENON<br>Chapter 02 --- THE DESIRE TO CROSS OVER<br>Chapter 03 --- THE DESIRE TO LEARN<br>Chapter 04 --- TRANSBORDER

碩士<br>國立中山大學<br>企業管理學系研究所<br>102<br>This research starts with the life-stories of three professional planners, and raised the question that how do planners learn. With narrative for research method and the viewpoint of becoming for theory view, the researcher begins to discuss about the learning process of three planners. Firstly, this research finds that planners are working in a multi-boundary habitat. Planners work together with, for instance, lighting, designer, stage staff, and artist. Based on this, the research finally gets two findings: 1.connection as learning 2.culture translation as learning. Instead of traditional opinion that learning is knowledge acquisition, learning in this research is defined as a process of becoming others. In this way, connection and culture translation are recognized as self-creation and self-recreation.

碩士<br>國立高雄師範大學<br>數學系<br>107<br>Machine learning technologies has attracted a lot of attention, it is becoming pop- ular and be widely applied in most recent years. In the real world applications, we can get a huge amount of data, but these data are unlabeled data. However, many classic classication algorithms we often used cannot be used directly. Since learning a good classier usually need large quantities of labels information, but get labels information is usually dicult or expensive (need time and money). Even if we just labeled some of the training data, the time and money cost of labeling data is unimaginable. Therefore, we used active learning, an algorithm that can reduces the training set and labeling cost as much as possible. And combine transfer learning to solve the weaknesses of active learning algorithms: initial selection. Moreover, we also try to improve the performance of active learning in the beginning. In this paper, we propose a simple active learning framework with cross domain transfer, which using labeled data from a dierent (but related) tasks to improve the perfor- mance of an active learner. We use some commonly used transfer learning data sets to conrm our method analysis. Moreover, the results of experiment verify the eectiveness of the method we proposed.

碩士<br>國立臺灣大學<br>電信工程學研究所<br>102<br>Traditionally, recommender systems make recommendations based on a single domain (e.g., movie or book domain) only. Recently, several cross-domain recommendation models have been proposed. Some of them proposed to leverage the common latent factors in the rating patterns of users-to-items co-clustering between domains and proposed to transfer the knowledge of such common latent factors to enhance the overall recommendation performance. However, these models often restrain themselves to transfer all the common knowledge between domains. Furthermore, these models often include all the domains in theirs participating domain set without selecting and evaluating the effect of including such domain into the transfer learning task. In this thesis, we propose a novel selective transfer learning model for the cross-multiple domains recommendation problem. This model not only can discover and apply the cross-multiple domains rating patterns to enhance the performance of recommendation on each of the participating domain, but also can select the most beneficial and efficient common knowledge then transfer the knowledge to each of the participating domain to improve the recommendation performance. In addition, we define a domain property index to evaluate the benefit of including each domain into the transfer learning task. Hence, this framework is able to discover and leverage the most influential common and cross-multiple domains rating patterns, and select an efficient participating domain set to enhance the recommendation performance. Extensive experiments on several real world datasets indicate that the proposed framework outperforms state-of-the-art methods for cross-domain recommendation task.

碩士<br>國立體育大學<br>休閒產業經營學系<br>103<br>The purpose of this study is to inquire into adventure education students’ cross-cultural learning process. It aims to understand students’ cross-cultural learning motivation, application process, and learning process. (Learning process includes, but is not limited to something such as lifestyles that may influence students’ learning efficiency.) The study uses qualitative research methods and is conducted under semi-structured interviews. It gathers information from three interviewees who have been to U.S National Outdoor Leadership School (NOLS). The result is as follows. Students’ learning motivation are greatly influenced by institutions' words of mouth and their prestige. In addition, whether programs’ web pages provide comprehensive and clear information about the courses they are offering can also be factors that determine students’ motivation and decision to the programs. More importantly, if students’ could gain first hand advice from their mentors, professors, or friends' experience, it could boost their intention to explore the field. According to the three interviewees, there are two possible ways to decrease risks that may be involved in the programs such as NOLS. For instance, a knowledgeable and experienced instructor with extensive technical skills regards to the field and an appropriate instructors/students ratio can decrease the risks that may accompany with the courses. Low instructors/students ratio not only benefits the control of risk management, but also sustains the quality of learning itself and increases each student's practice opportunities. Students would get more attention from their instructors. The two suggestions may also help in planning courses and making proper arrangements. Part of the learning process is about letting students introspect their living styles. According to the interviewees, they all mentioned that it requires time for one to adapt and learn different diets and cooking methods. Besides that, helping students develop good communication skills, properly express their opinions or thoughts, and respect others' opinions and decisions would let students learn early about how to be a part of a group and how to be with others. The most important thing is to always keep positive thinking with oneself, and base one’s intention on goodwill.