The collection of large speech databases is not a trivial task (if done properly). It is not always possible to collect, segment and annotate large databases for every task or language. It is also often the case that there are imbalances in the databases, as a result of little data being available for a specific subset of individuals. An example of one such imbalance is the fact that there are often more male speakers than female speakers (or vice-versa). If there are, for example, far fewer female speakers than male speakers, then the recognizers will tend to work poorly for female speakers (as compared to performance for male speakers). This thesis focuses on using Bayesian and discriminative training algorithms to improve continuous speech recognition systems in scenarios where there is a limited amount of training data available. The research reported in this thesis can be divided into three categories: • Overspecialization is characterized by good recognition performance for the data used during training, but poor recognition performance for independent testing data. This is a problem when too little data is available for training purposes. Methods of reducing overspecialization in the minimum classification error algo¬rithm are therefore investigated. • Development of new Bayesian and discriminative adaptation/training techniques that can be used in situations where there is a small amount of data available. One example here is the situation where an imbalance in terms of numbers of male and female speakers exists and these techniques can be used to improve recognition performance for female speakers, while not decreasing recognition performance for the male speakers. • Bayesian learning, where Bayesian training is used to improve recognition perfor¬mance in situations where one can only use the limited training data available. These methods are extremely computationally expensive, but are justified by the improved recognition rates for certain tasks. This is, to the author's knowledge, the first time that Bayesian learning using Markov chain Monte Carlo methods have been used in hidden Markov model speech recognition. The algorithms proposed and reviewed are tested using three different datasets (TIMIT, TIDIGITS and SUNSpeech), with the tasks being connected digit recognition and con¬tinuous speech recognition. Results indicate that the proposed algorithms improve recognition performance significantly for situations where little training data is avail¬able. / Thesis (PhD (Electronic Engineering))--University of Pretoria, 2006. / Electrical, Electronic and Computer Engineering / unrestricted

O objetivo do presente estudo foi verificar a influência do modelo tradicional de periodização (com picos múltiplos) sobre a aptidão física em testes de campo, desempenho competitivo e variação técnica em competições de judô. Neste sentido, a amostra foi composta por atletas das classes Juvenil e Júnior que foram monitorados ao longo de duas temporadas anuais de treinamento (12 atletas por temporada), nas quais foram quantificados parâmetros da carga interna (Carga de Treinamento e o Training Strain) de cada período de treinamento, por meio da percepção subjetiva de esforço da sessão. O desempenho físico foi avaliado em momentos de alteração dos períodos de treinamento, mensurando a potência de membros inferiores (salto horizontal), a resistência de força dinâmica de membros superiores (teste de barra com o judogi) e a aptidão aeróbia e anaeróbia em situação específica (SJFT - Special Judo Fitness Test). Adicionalmente, nas competições principais, os combates foram filmados para posterior quantificação do desempenho competitivo (percentual de vitórias, pontos conquistados, pontos sofridos, índice de eficiência de ataque e índices de efetividade) e da variação técnica (técnicas de projeção, técnicas de domínio e direções dos ataques efetivos). As variáveis foram comparadas via modelo misto para análise de medidas repetidas, seguido por teste de Bonferroni. As associações entre as variáveis foram testadas via coeficiente de correlação de Pearson ou de Spearman. Na temporada 2011, houve diferenças (P < 0,05) na Carga de Treinamento, Training Strain, salto horizontal, teste de barra com o judogi, percentual de vitórias, pontos conquistados, pontos sofridos, índice de eficiência de ataque, direções de ataque, técnicas de projeção e de domínio, bem como correlação entre a Carga de Treinamento três dias antes do Campeonato Paulista Fase Regional e o índice de eficiência de ataque nesta competição (r = - 0,832; P = 0,001; muito grande). Na temporada 2012, verificou-se diferenças (P < 0,05) na Carga de Treinamento, Training Strain, salto horizontal, teste de barra com o judogi, nas variáveis número de projeções, frequência cardíaca após um minuto de repouso e índice do SJFT, pontos conquistados e técnicas de projeção efetivas, além de correlações entre o número de projeções no SJFT com o índice de eficiência de ataque (r = 0,629; P = 0,028; grande) e o percentual de vitórias (r = 0,634; P = 0,027; grande) no Campeonato Paulista Fase Inter-regional. A relevância do monitoramento da carga interna de treinamento e do desempenho físico foi reforçada pelas associações com parâmetros de desempenho competitivo. Embora os indicadores de desempenho físico, desempenho competitivo e variação técnica tenham sofrido alterações ao longo das temporadas, estas variações ocorreram de maneira dessincronizadas, aparentemente sem influência das cargas de treinamento adotadas. Assim, os resultados do presente estudo indicam que, para jovens atletas de judô, o modelo tradicional de periodização (com picos múltiplos) não desenvolve adaptações contínuas sobre os parâmetros de desempenho analisados, mas parece ser eficiente na elevação do desempenho físico no início da temporada e posterior estabilização destas adaptações / The aim of this study was investigate the influence of the traditional periodization model (with multiple peaks) on physical fitness in field tests, competitive performance and technical variation in judo competitions. In this sense, the sample consisted of Junior and Cadet athletes who were monitored over two annual training seasons (12 athletes per season) in which internal training load parameters (Training Load and Training Strain) of each training period were quantified through the session rating of perceived exertion. Physical performance was evaluated when the periods of training were changed, measuring lower limbs muscle power (standing long jump), upper limbs dynamic strength endurance (dynamic judogi chin up) and aerobic and anaerobic fitness in a specific situation (SJFT - Special Judo Fitness Test). Additionally, in main competitions, fights were filmed for later quantification of competitive performance (winning percentage, points scored, points conceded, attack efficiency index and effectiveness index) and technical variation (throwing techniques, groundwork techniques and directions of throws). The variables were compared by mixed model for repeated measures analysis, followed by Bonferroni test. The associations among variables were tested by Pearson\'s or Spearman\'s correlation coefficient. In the 2011 season, there were differences (P <0.05) in the Training Load, Training Strain, standing long jump, dynamic judogi chin up, winning percentage, points scored, points conceded, attack efficiency index, directions of throws and throwing techniques, in addition to correlation between the Training Load three days before a regional championship and the attack efficiency index in this competition (r = - 0.832; P = 0.001; very large). In 2012 season, there were differences (P <0.05) in Training Load, Training Strain, standing long jump, dynamic judogi chin up, SJFT variables (number of throws, heart rate one minute after the test and index), points scored and throwing techniques, in addition to correlations between the number of throws in SJFT with the attack efficiency index (r = 0.629; P = 0.028; large) and winning percentage (r = 0.634; P = 0.027; large) in a inter-regional championship. The relevance of monitoring the internal training load and the physical performance was reinforced by the associations with competitive performance parameters. Although physical performance, competitive performance and technical variation have changed over the seasons, these variations occurred in an unsynchronized way, apparently without influence of the training loads adopted. Thus, the results of this study indicate that for young judo athletes, the traditional periodization model (with multiple peaks) does not develop continuous adaptations of the performance parameters analyzed, but it seems to be effective in increasing the physical performance early in the season and subsequent stabilization of the adaptations

O objetivo do presente estudo foi verificar a influência do modelo tradicional de periodização (com picos múltiplos) sobre a aptidão física em testes de campo, desempenho competitivo e variação técnica em competições de judô. Neste sentido, a amostra foi composta por atletas das classes Juvenil e Júnior que foram monitorados ao longo de duas temporadas anuais de treinamento (12 atletas por temporada), nas quais foram quantificados parâmetros da carga interna (Carga de Treinamento e o Training Strain) de cada período de treinamento, por meio da percepção subjetiva de esforço da sessão. O desempenho físico foi avaliado em momentos de alteração dos períodos de treinamento, mensurando a potência de membros inferiores (salto horizontal), a resistência de força dinâmica de membros superiores (teste de barra com o judogi) e a aptidão aeróbia e anaeróbia em situação específica (SJFT - Special Judo Fitness Test). Adicionalmente, nas competições principais, os combates foram filmados para posterior quantificação do desempenho competitivo (percentual de vitórias, pontos conquistados, pontos sofridos, índice de eficiência de ataque e índices de efetividade) e da variação técnica (técnicas de projeção, técnicas de domínio e direções dos ataques efetivos). As variáveis foram comparadas via modelo misto para análise de medidas repetidas, seguido por teste de Bonferroni. As associações entre as variáveis foram testadas via coeficiente de correlação de Pearson ou de Spearman. Na temporada 2011, houve diferenças (P < 0,05) na Carga de Treinamento, Training Strain, salto horizontal, teste de barra com o judogi, percentual de vitórias, pontos conquistados, pontos sofridos, índice de eficiência de ataque, direções de ataque, técnicas de projeção e de domínio, bem como correlação entre a Carga de Treinamento três dias antes do Campeonato Paulista Fase Regional e o índice de eficiência de ataque nesta competição (r = - 0,832; P = 0,001; muito grande). Na temporada 2012, verificou-se diferenças (P < 0,05) na Carga de Treinamento, Training Strain, salto horizontal, teste de barra com o judogi, nas variáveis número de projeções, frequência cardíaca após um minuto de repouso e índice do SJFT, pontos conquistados e técnicas de projeção efetivas, além de correlações entre o número de projeções no SJFT com o índice de eficiência de ataque (r = 0,629; P = 0,028; grande) e o percentual de vitórias (r = 0,634; P = 0,027; grande) no Campeonato Paulista Fase Inter-regional. A relevância do monitoramento da carga interna de treinamento e do desempenho físico foi reforçada pelas associações com parâmetros de desempenho competitivo. Embora os indicadores de desempenho físico, desempenho competitivo e variação técnica tenham sofrido alterações ao longo das temporadas, estas variações ocorreram de maneira dessincronizadas, aparentemente sem influência das cargas de treinamento adotadas. Assim, os resultados do presente estudo indicam que, para jovens atletas de judô, o modelo tradicional de periodização (com picos múltiplos) não desenvolve adaptações contínuas sobre os parâmetros de desempenho analisados, mas parece ser eficiente na elevação do desempenho físico no início da temporada e posterior estabilização destas adaptações / The aim of this study was investigate the influence of the traditional periodization model (with multiple peaks) on physical fitness in field tests, competitive performance and technical variation in judo competitions. In this sense, the sample consisted of Junior and Cadet athletes who were monitored over two annual training seasons (12 athletes per season) in which internal training load parameters (Training Load and Training Strain) of each training period were quantified through the session rating of perceived exertion. Physical performance was evaluated when the periods of training were changed, measuring lower limbs muscle power (standing long jump), upper limbs dynamic strength endurance (dynamic judogi chin up) and aerobic and anaerobic fitness in a specific situation (SJFT - Special Judo Fitness Test). Additionally, in main competitions, fights were filmed for later quantification of competitive performance (winning percentage, points scored, points conceded, attack efficiency index and effectiveness index) and technical variation (throwing techniques, groundwork techniques and directions of throws). The variables were compared by mixed model for repeated measures analysis, followed by Bonferroni test. The associations among variables were tested by Pearson\'s or Spearman\'s correlation coefficient. In the 2011 season, there were differences (P <0.05) in the Training Load, Training Strain, standing long jump, dynamic judogi chin up, winning percentage, points scored, points conceded, attack efficiency index, directions of throws and throwing techniques, in addition to correlation between the Training Load three days before a regional championship and the attack efficiency index in this competition (r = - 0.832; P = 0.001; very large). In 2012 season, there were differences (P <0.05) in Training Load, Training Strain, standing long jump, dynamic judogi chin up, SJFT variables (number of throws, heart rate one minute after the test and index), points scored and throwing techniques, in addition to correlations between the number of throws in SJFT with the attack efficiency index (r = 0.629; P = 0.028; large) and winning percentage (r = 0.634; P = 0.027; large) in a inter-regional championship. The relevance of monitoring the internal training load and the physical performance was reinforced by the associations with competitive performance parameters. Although physical performance, competitive performance and technical variation have changed over the seasons, these variations occurred in an unsynchronized way, apparently without influence of the training loads adopted. Thus, the results of this study indicate that for young judo athletes, the traditional periodization model (with multiple peaks) does not develop continuous adaptations of the performance parameters analyzed, but it seems to be effective in increasing the physical performance early in the season and subsequent stabilization of the adaptations

京都大学 / 0048 / 新制・課程博士 / 博士(情報学) / 甲第19849号 / 情博第600号 / 新制||情||104(附属図書館) / 32885 / 京都大学大学院情報学研究科知能情報学専攻 / (主査)教授 河原 達也, 教授 黒橋 禎夫, 教授 鹿島 久嗣 / 学位規則第4条第1項該当 / Doctor of Informatics / Kyoto University / DFAM

The aim of this thesis is to create a competency model for the jobs of new members of a student organization. The competency model is used to assemble a list of the types of educational activities that should be undertaken by the member in a half year frame after they start working on the new position in the organization. This goal is reflected by the structure of this thesis. The theoretical part is divided into two chapters. In the first one, the reader is acquainted with the concepts of competence, employee education, competency model, and management through competencies. The knowledge of these terms and their meaning is crucial for the understanding of the domain of education management. The contents of the second chapter deal with the various methods of development and educational activities, the systematical education and its phases, and the creation of competency model in an organization. The knowledge that is presented in the theoretical part comes originally from the corporate environment. But there it was created to address the same problems that are also faced by educational organizations, and also by organizations in which the education of its members is one of their secondary goals. The theoretical part is followed by the research part. The research methodology is described first. This...

Training program for new financial advisors analysis of the company XY examines the whole training process in this company. The amalysis concentrates on distribution employees training. It tryes to find out the problems existing here. The analysis especially solves the evaluation of the training process. The work involves the recommendations leading to improve the training program in the future.

Image augmentation is a field that covers the subject area of altering existing data to create more for the use of model training processes. It may be seen as the practice of expanding upon existing data using a range of techniques that employ transformations to improve the diversity of training sets when applied to machine learning. In our case of image recognition, triplet loss is utilised to pair a reference image to a matching and non-matching input. However, since there are many single images, augmentation techniques are relied upon to expand upon our data set to improve the recognition of images and create true positives. True positives are created using standard augmentation techniques like perspective transformation, contrast, cropping, and more. Despite this, the same images may undergo other types of alterations, natural disturbances such as transformations and warping, that are not captured by standard augmentation techniques. Such instances constitute to the variance in identification. Therefore, the analysis of augmentations by artificial intelligence (AI) based recognition is proposed; AI is used in order to identify what contributes to realistic disturbances of single images that better imitate real-life transformations. Analysing existing standard image augmentation techniques should provide further insight within this scope, as to better determine ways of emulating natural disturbances, and the formulation of non-standard practices in tandem. How this is done is by the use of an image's identity, the pixels it's comprised of and their distributions. Through a methodology of inspecting image identities, the breaking down of augmentations, and the inquiry into practices of non-standard image augmentation techniques, we detect the variance in accuracy of generated models, analysing the comprised data sets. Our results show that augmentations improve accuracy on a basis of variance and divergence from the original image. Subsequent discussion expands upon the identities of images and how augmentations must still resemble true positives, with the potential of an augmentation gauged by its influence on the rate of growth and highest accuracy of a model. / <p></p><p></p><p></p>

在音樂中，組曲是一種特別的創作形式。組曲將多首音樂段落組合排列，並且在音樂段落之間加入間奏，形成一首音樂組曲。組曲的編曲重點在於音樂段落的編排順序及段落之間的連結。平時在宴會、舞會、餐廳、賣場等場合中，往往都會連續播放多首流行音樂。利用電腦編曲自動產生流行音樂組曲，將可提升播放音樂的銜接與流暢感。 因此，本研究利用資料探勘技術及音樂編曲理論，將多首音樂重新改編成一首組曲。系統首先將每首音樂分段並找出每首音樂的代表段落。接著，系統根據代表段落間的相似度編排順序。最後，為了達到組曲中音樂段落連接的流暢性，我們以模型訓練的方式在段落連結間加入間奏。系統從訓練資料學習產生旋律發展、和弦進程與節奏的模型，接著分析代表段落的動機、旋律、和弦及節奏，使得組曲編曲後的段落連結更為流暢且完整。本研究以流行音樂鋼琴伴奏曲為測試資料，我們分別邀請三十四位受過音樂訓練與未受音樂訓練的測試者，針對本研究所提出的鋼琴伴奏節奏辨識、代表段落萃取、段落順序編排及間奏產生，評估其效果。實驗結果顯示，本研究所提出的順序編排與間奏產生技術，對於組曲的流暢感，有著相當大的幫助。 / In music, a medley is an organized piece composed from segments of existing pieces. Ordering and bridge for connection between segments are the important elements for medley arrangement. Automatic medley arrangement is helpful for playing a set of music continuously which is common in banquet, party, restaurant, shopping mall, etc.. This thesis aims to develop the automatic medley arrangement method by using data mining techniques and music arrangement theory. The proposed method first segments each music and discovers the significant segment from each music. Then, the linear arrangement based on the similarities between significant segments is generated. Finally, in order to connect segments smoothly in the medley, the bridge between two segments is generated and inserted by using model training. Three models, melody progression, chord progression and rhythm models are learned from training data. For the experiments, testing data is collected from popular piano music and thirty-four people are invited to evaluate the effectiveness of the rhythm recognition of accompaniment, the extraction of significant segment, the linear arrangement of segments, and the creation of bridge. Experimental results show that the proposed medley arrangement method performs well.

La conception optimale de turbomachines repose usuellement sur des méthodes itératives avec des évaluations soit expérimentales, soit numériques qui peuvent conduire à des coûts élevés en raison des nombreuses manipulations ou de l’utilisation intensive de CPU. Afin de limiter ces coûts et de raccourcir les temps de développement, le présent travail propose d’intégrer une méthode de paramétrisation et de métamodélisation dans un cycle de conception d’une turbomachine axiale basse vitesse. La paramétrisation, réalisée par l’étude de sensibilité d’ordre élevé des équations de Navier-Stokes, permet de construire une base de données paramétrée qui contient non seulement les résultats d’évaluations, mais aussi les dérivées simples et les dérivées croisées des objectifs en fonction des paramètres. La plus grande quantité d’informations apportée par les dérivées est avantageusement utilisée lors de la construction de métamodèles, en particulier avec une méthode de Co-Krigeage employée pour coupler plusieurs bases de données. L’intérêt économique de la méthode par rapport à une méthode classique sans dérivée réside dans l’utilisation d’un nombre réduit de points d’évaluation. Lorsque ce nombre de points est véritablement faible, il peut arriver qu’une seule valeur de référence soit disponible pour une ou plusieurs dimensions, et nécessite une hypothèse de répartition d’erreur. Pour ces dimensions, le Co-Krigeage fonctionne comme une extrapolation de Taylor à partir d’un point et de ses dérivées. Cette approche a été expérimentée avec la construction d’un méta-modèle pour une hélice présentant un moyeu conique. La méthodologie fait appel à un couplage de bases de données issues de deux géométries et deux points de fonctionnement. La précision de la surface de réponse a permis de conduire une optimisation avec un algorithme génétique NSGA-2, et les deux optima sélectionnés répondent pour l’un à une maximisation du rendement, et pour l’autre à un élargissement de la plage de fonctionnement. Les résultats d’optimisation sont finalement validés par des simulations numériques supplémentaires. / The turbomachinery optimal design usually relies on some iterative methods with either experimental or numerical evaluations that can lead to high cost due to numerous manipulations and intensive usage of CPU. In order to limit the cost and shorten the development time, the present thesis work proposes to integrate a parameterization method and the meta-modelization method in an optimal design cycle of an axial low speed turbomachine. The parameterization, realized by the high order sensitivity study of Navier-Stokes equations, allows to construct a parameterized database that contains not only the evaluations results, but also the simple and cross derivatives of objectives as a function of parameters. Enriched information brought by the derivatives are utilized during the meta-model construction, particularly by the Co-Kriging method employed to couple several databases. Compared to classical methods that are without derivatives, the economic benefit of the proposed method lies in the use of less reference points. Provided the number of reference points is small, chances are a unique point presenting at one or several dimensions, which requires a hypothesis on the error distribution. For those dimensions, the Co-Kriging works like a Taylor extrapolation from the reference point making the most of its derivatives. This approach has been experimented on the construction of a meta-model for a conic hub fan. The methodology recalls the coupling of databases based on two fan geometries and two operating points. The precision of the meta-model allows to perform an optimization with help of NSGA-2, one of the optima selected reaches the maximum efficiency, and another covers a large operating range. The optimization results are eventually validated by further numerical simulations.