When using an intermittent source of energy such as wind power together with a synchronous machine a frequency converter system is needed to decouple the generator from the grid, due to the fluctuations in wind speed resulting in fluctuating electrical frequency. The aim of this master's thesis is to investigate how different types of rectification methods affect permanent magnet synchronous machines of different saliency ratios. A literature study was carried out to review the research within the area and to acquire the necessary knowledge to carry out the work. Two simulation models were created that include a permanent magnet synchronous generator driven by a wind turbine and connected to the grid via a frequency converter, where one model utilizes active rectification and one utilizes passive rectification. The simulation models were verified by carrying out an experiment on a similar setup, which showed that the simulation results coincide well with the results of the experiment. The results of the simulation study were then used to compare the rectification systems as well as investigate the affect that rotor saliency has on the system. It was shown that the active rectification provided a higher efficiency than the passive rectification system, however the saliency of the rotor had little effect on the system

abstract: Video deinterlacing is a key technique in digital video processing, particularly with the widespread usage of LCD and plasma TVs. This thesis proposes a novel spatio-temporal, non-linear video deinterlacing technique that adaptively chooses between the results from one dimensional control grid interpolation (1DCGI), vertical temporal filter (VTF) and temporal line averaging (LA). The proposed method performs better than several popular benchmarking methods in terms of both visual quality and peak signal to noise ratio (PSNR). The algorithm performs better than existing approaches like edge-based line averaging (ELA) and spatio-temporal edge-based median filtering (STELA) on fine moving edges and semi-static regions of videos, which are recognized as particularly challenging deinterlacing cases. The proposed approach also performs better than the state-of-the-art content adaptive vertical temporal filtering (CAVTF) approach. Along with the main approach several spin-off approaches are also proposed each with its own characteristics. / Dissertation/Thesis / M.S. Electrical Engineering 2012

abstract: Mixture of experts is a machine learning ensemble approach that consists of individual models that are trained to be ``experts'' on subsets of the data, and a gating network that provides weights to output a combination of the expert predictions. Mixture of experts models do not currently see wide use due to difficulty in training diverse experts and high computational requirements. This work presents modifications of the mixture of experts formulation that use domain knowledge to improve training, and incorporate parameter sharing among experts to reduce computational requirements. First, this work presents an application of mixture of experts models for quality robust visual recognition. First it is shown that human subjects outperform deep neural networks on classification of distorted images, and then propose a model, MixQualNet, that is more robust to distortions. The proposed model consists of ``experts'' that are trained on a particular type of image distortion. The final output of the model is a weighted sum of the expert models, where the weights are determined by a separate gating network. The proposed model also incorporates weight sharing to reduce the number of parameters, as well as increase performance. Second, an application of mixture of experts to predict visual saliency is presented. A computational saliency model attempts to predict where humans will look in an image. In the proposed model, each expert network is trained to predict saliency for a set of closely related images. The final saliency map is computed as a weighted mixture of the expert networks' outputs, with weights determined by a separate gating network. The proposed model achieves better performance than several other visual saliency models and a baseline non-mixture model. Finally, this work introduces a saliency model that is a weighted mixture of models trained for different levels of saliency. Levels of saliency include high saliency, which corresponds to regions where almost all subjects look, and low saliency, which corresponds to regions where some, but not all subjects look. The weighted mixture shows improved performance compared with baseline models because of the diversity of the individual model predictions. / Dissertation/Thesis / Doctoral Dissertation Electrical Engineering 2018

abstract: Video object segmentation (VOS) is an important task in computer vision with a lot of applications, e.g., video editing, object tracking, and object based encoding. Different from image object segmentation, video object segmentation must consider both spatial and temporal coherence for the object. Despite extensive previous work, the problem is still challenging. Usually, foreground object in the video draws more attention from humans, i.e. it is salient. In this thesis we tackle the problem from the aspect of saliency, where saliency means a certain subset of visual information selected by a visual system (human or machine). We present a novel unsupervised method for video object segmentation that considers both low level vision cues and high level motion cues. In our model, video object segmentation can be formulated as a unified energy minimization problem and solved in polynomial time by employing the min-cut algorithm. Specifically, our energy function comprises the unary term and pair-wise interaction energy term respectively, where unary term measures region saliency and interaction term smooths the mutual effects between object saliency and motion saliency. Object saliency is computed in spatial domain from each discrete frame using multi-scale context features, e.g., color histogram, gradient, and graph based manifold ranking. Meanwhile, motion saliency is calculated in temporal domain by extracting phase information of the video. In the experimental section of this thesis, our proposed method has been evaluated on several benchmark datasets. In MSRA 1000 dataset the result demonstrates that our spatial object saliency detection is superior to the state-of-art methods. Moreover, our temporal motion saliency detector can achieve better performance than existing motion detection approaches in UCF sports action analysis dataset and Weizmann dataset respectively. Finally, we show the attractive empirical result and quantitative evaluation of our approach on two benchmark video object segmentation datasets. / Dissertation/Thesis / M.S. Computer Science 2013

Ce travail est consacré à l’étude contrastive du fonctionnement de l’anaphore résomptive à tête nominale (ARN) en allemand et en français. Il s’appuie principalement sur un corpus de textes journalistiques. Le journaliste est un scripteur qui, par le biais de son article, désire informer son lecteur, voire le faire adhérer à sa vision du monde. Mais il est soumis à des contraintes de place. L’ARN est un moyen linguistique de choix, car elle permet un compactage par abstraction et par généralisation des informations sous la forme d’un concept introduit par le nom-tête de l’ARN. Il en ressort que les constituants de l’ARN que sont le déterminatif, le nom-tête et son expansion, et l’ARN en soi jouent un rôle non négligeable au sein de l’énoncé et du paragraphe. Afin de rendre compte de la dimension cognitive du phénomène anaphorique, le recours à la notion de saillance permet de montrer le rôle central des ARN dans la cohérence textuelle. Ce type d’expressions anaphoriques joue en outre un rôle au niveau textuel et au niveau argumentatif. L’ARN est en effet une balise saillante au service de l’argumentation. Elle permet de structurer et d’organiser le discours, ainsi que de participer à la stratégie argumentative du journaliste. / The purpose of this work is to develop a contrastive study of nominal anaphoric encapsulation in German and in French. It is mainly based on a corpus of newspaper articles. Thanks to their articles, journalists want to inform their readers, and sometimes make them share their own world view. But journalists are forced to do with limited space. Nominal anaphoric encapsulation is a perfect linguistic tool because it allows concision through the abstraction and generalization of information – a concept being introduced by the head noun of the nominal anaphoric encapsulation. Therefore, constituent parts of nominal anaphoric encapsulation (determinative, head noun and its expansion) and nominal anaphoric encapsulation itself play an important role in the clause and in the paragraph. In order to analyse the cognitive dimension of the anaphoric phenomenon, we use the notion of saliency to show the central role of nominal anaphoric encapsulation in textual coherence. Furthermore, such anaphoric expressions play a role at the textual and argumentative levels. Nominal anaphoric encapsulation is actually a salient buoy supporting the argumentation. It serves to structure and organize the speech, and to participate in the argumentative strategy of journalists.

In today’s society of increased globalization and digitalization openness has become a buzzword. This raises questions about what we mean by openness and how it is interpreted in various contexts. This thesis has two aims; to explore how openness is interpreted in planning, politics and political decision-making, and to develop an analytical tool to assess openness in different contexts. A new institutional theory framework that centers on the interplay between institutions and actors has been used, and three empirical case studies in a Swedish context were conducted to analyze how openness is interpreted in planning in metropolitan regions, in politics through the political parties and in political decision-making in the Stockholm region. The research concludes that openness in planning, politics and political decision-making is interpreted along two inter-linked narrative lines: ’openness to people’ and ’openness to knowledge, information and ideas’. It was more common to talk about peoples’ accessibility to public services and participation in different parts of society (’openness to people’) than to talk about issues of transparency and ’openness to knowledge, information and ideas’. The institutional framework shows how openness is interpreted at different institutional levels. To what degree openness is expressed at different institutional levels vary by context. In planning for instance, openness is mainly interpreted in terms of governance, whereas in politics and political decision-making, openness is interpreted in an inter-play between culture and norms, institutions, governance and practice. The institutional framework complementary context-specific theories and elaborated into an analytical model, was found useful to explain what mechanisms are at play when dealing with openness in planning, politics and political decision-making, and can be applicable in future research of openness in other geographical or organizational contexts. / <p>QC 20170914</p>

京都大学 / 0048 / 新制・課程博士 / 博士(情報学) / 甲第17967号 / 情博第511号 / 新制||情||91(附属図書館) / 30797 / 京都大学大学院情報学研究科知能情報学専攻 / (主査)教授 松山 隆司, 教授 乾 敏郎, 教授 石井 信 / 学位規則第4条第1項該当 / Doctor of Informatics / Kyoto University / DFAM

No description available.

In the recent years, a huge success has been accomplished in prediction of human eye fixations. Several studies employed deep learning to achieve high accuracy of prediction of human eye fixations. These studies rely on pre-trained deep learning for object classification. They exploit deep learning either as a transfer-learning problem, or the weights of the pre-trained network as the initialization to learn a saliency model. The utilization of such pre-trained neural networks is due to the relatively small datasets of human fixations available to train a deep learning model. Another relatively less prioritized problem is amount of computation of such deep learning models requires expensive hardware. In this dissertation, two approaches are proposed to tackle abovementioned problems. The first approach, codenamed DeepFeat, incorporates the deep features of convolutional neural networks pre-trained for object and scene classifications. This approach is the first approach that uses deep features without further learning. Performance of the DeepFeat model is extensively evaluated over a variety of datasets using a variety of implementations. The second approach is a deep learning saliency model, codenamed ClassNet. Two main differences separate the ClassNet from other deep learning saliency models. The ClassNet model is the only deep learning saliency model that learns its weights from scratch. In addition, the ClassNet saliency model treats prediction of human fixation as a classification problem, while other deep learning saliency models treat the human fixation prediction as a regression problem or as a classification of a regression problem.

This study investigates Japanese high school teachers’ learning of grammar proceduralization strategies designed to promote oral fluency. It is a multiple case study of six Japanese EFL teachers who learn to use their declarative knowledge of L2 grammar while engaging in tasks that enable them to compare their oral output with a native English speaker’s reformulations of it. Past studies of language learning strategies have been primarily focused either on the learners’ general study habits toward the target language or on their skill-specific language learning strategies in the areas of listening, reading, speaking, writing, and vocabulary. Although the effectiveness of these strategies on learning outcomes is known to be highly constrained by learners’ prior linguistic knowledge, strategies to proceduralize grammar, a core component of one’s linguistic knowledge, have not been well researched. Therefore, little is known about how learners’ volitional efforts contribute to the proceduralization of L2 grammar. Research into oral fluency development has provided evidence that the use of formulas promotes fluency, but it has not revealed how formulas and other varieties of multiword units contribute to different aspects of oral fluency; namely, temporal, repair, and perceived fluency. This study fills these gaps in research by defining, investigating, and creating a set of grammar proceduralization strategies as a promising construct that sheds light on what learners can proactively do to proceduralize their knowledge of L2 grammar. The three main purposes of this study are to (a) investigate Japanese EFL teachers’ grammar proceduralization strategies for appropriating, refining, and using their grammar knowledge, (b) identify L2 morphosyntactic forms and multiword units that facilitate Japanese EFL teachers’ oral production during oral summary and personal anecdote tasks, and (c) investigate the possible relationships between the participants’ L2 grammar proceduralization strategies, their use of specific grammar forms, and their oral fluency development. The participants are six Japanese teachers of English who teach at public senior high schools in Japan. To gain a detailed understanding of the participants’ complex learning processes, their learning trajectories were investigated for a period of six months, using a longitudinal mixed-methods design, with detailed analyses of their English learning history, post-task protocols, linguistic measures, and rubric-based assessment of their oral fluency development. The results provide (a) a typology of L2 grammar proceduralization strategies created based on models of communicative competence and speech production, (b) 16 categories of grammar items that have potential impact on oral fluency development, with insights into factors that facilitate and debilitate the participants’ use of these grammar items, and (c) insights into how the participants’ goal orientation leads to their orchestration of L2 grammar proceduralization strategies, their use of 16 categories of grammar items, and to the different trajectories of their temporal, repair, and perceived fluency development. This study presents data to support the conclusion that a reverse-saliency strategy to learn L2 grammar in concepts, propositions, and discourse is a key to effective EFL pedagogy. / Teaching & Learning