Submitted by Fabio Sobreira Campos da Costa (fabio.sobreira@ufpe.br) on 2016-03-15T12:52:14Z No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) Tese_Jarley_Nobrega_CORRIGIDA.pdf: 12392055 bytes, checksum: 30d9ff36e7236d22ddc3a16dd942341f (MD5) / Made available in DSpace on 2016-03-15T12:52:14Z (GMT). No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) Tese_Jarley_Nobrega_CORRIGIDA.pdf: 12392055 bytes, checksum: 30d9ff36e7236d22ddc3a16dd942341f (MD5) Previous issue date: 2015-08-24 / Em aplicações de aprendizagem de máquina, é comum encontrar situações onde o conjunto de entrada não está totalmente disponível no início da fase de treinamento. Uma solução conhecida para essa classe de problema é a realização do processo de aprendizagem através do fornecimento sequencial das instâncias de treinamento. Entre as abordagens mais recentes para esses métodos, encontram-se as baseadas em redes neurais do tipo Single Layer Feedforward Network (SLFN), com destaque para as extensões da Extreme Learning Machine (ELM) para aprendizagem sequencial. A versão sequencial da ELM, chamada de Online Sequential Extreme Learning Machine (OS-ELM), utiliza uma solução recursiva de mínimos quadrados para atualizar os pesos de saída da rede através de uma matriz de covariância. Entretanto, a implementação da OS-ELM e suas extensões sofrem com o problema de multicolinearidade entre os elementos da matriz de covariância. Essa tese introduz um novo método para aprendizagem sequencial com capacidade para tratar os efeitos da multicolinearidade. Chamado de Kalman Learning Machine (KLM), o método proposto utiliza o filtro de Kalman para a atualização sequencial dos pesos de saída de uma SLFN baseada na OS-ELM. Esse trabalho também propõe uma abordagem para a estimativa dos parâmetros do filtro, com o objetivo de diminuir a complexidade computacional do treinamento. Além disso, uma extensão do método chamada de Extended Kalman Learning Machine (EKLM) é apresentada, voltada para problemas onde a natureza do sistema em estudo seja não linear. O método proposto nessa tese foi comparado com alguns dos mais recentes e efetivos métodos para o tratamento de multicolinearidade em problemas de aprendizagem sequencial. Os experimentos executados mostraram que o método proposto apresenta um desempenho melhor que a maioria dos métodos do estado da arte, quando medidos o de erro de previsão e o tempo de treinamento. Um estudo de caso foi realizado, aplicando o método proposto a um problema de previsão de séries temporais para o mercado financeiro. Os resultados confirmaram que o KLM consegue simultaneamente reduzir o erro de previsão e o tempo de treinamento, quando comparado com os demais métodos investigados nessa tese. / In machine learning applications, there are situations where the input dataset is not fully available at the beginning of the training phase. A well known solution for this class of problem is to perform the learning process through the sequential feed of training instances. Among most recent approaches for sequential learning, we can highlight the methods based on Single Layer Feedforward Network (SLFN) and the extensions of the Extreme Learning Machine (ELM) approach for sequential learning. The sequential version of the ELM algorithm, named Online Sequential Extreme Learning Machine (OS-ELM), uses a recursive least squares solution for updating the output weights through a covariance matrix. However, the implementation of OS-ELM and its extensions suffer from the problem of multicollinearity for the hidden layer output matrix. This thesis introduces a new method for sequential learning in which the effects of multicollinearity is handled. The proposed Kalman Learning Machine (KLM) updates sequentially the output weights of an OS-ELM based network by using the Kalman filter iterative procedure. In this work, in order to reduce the computational complexity of the training process, a new approach for estimating the filter parameters is presented. Moreover, an extension of the method, named Extended Kalman Learning Machine (EKLM), is presented for problems where the dynamics of the model are non linear. The proposed method was evaluated by comparing the related state-of-the-art methods for sequential learning based on the original OS-ELM. The results of the experiments show that the proposed method can achieve the lowest forecast error when compared with most of their counterparts. Moreover, the KLM algorithm achieved the lowest average training time when all experiments were considered, as an evidence that the proposed method can reduce the computational complexity for the sequential learning process. A case study was performed by applying the proposed method for a problem of financial time series forecasting. The results reported confirm that the KLM algorithm can decrease the forecast error and the average training time simultaneously, when compared with other sequential learning algorithms.

published_or_final_version / Industrial and Manufacturing Systems Engineering / Doctoral / Doctor of Philosophy

Le pronostic industriel vise à étendre le cycle de vie d’un dispositif physique, tout en réduisant les couts d’exploitation et de maintenance. Pour cette raison, le pronostic est considéré comme un processus clé avec des capacités de prédiction. En effet, des estimations précises de la durée de vie avant défaillance d’un équipement, Remaining Useful Life (RUL), permettent de mieux définir un plan d’action visant à accroitre la sécurité, réduire les temps d’arrêt, assurer l’achèvement de la mission et l’efficacité de la production.Des études récentes montrent que les approches guidées par les données sont de plus en plus appliquées pour le pronostic de défaillance. Elles peuvent être considérées comme des modèles de type boite noire pour l’ étude du comportement du système directement `a partir des données de surveillance d’ état, pour définir l’ état actuel du système et prédire la progression future de défauts. Cependant, l’approximation du comportement des machines critiques est une tâche difficile qui peut entraîner des mauvais pronostic. Pour la compréhension de la modélisation du pronostic guidé par les données, on considère les points suivants. 1) Comment traiter les données brutes de surveillance pour obtenir des caractéristiques appropriées reflétant l’ évolution de la dégradation? 2) Comment distinguer les états de dégradation et définir des critères de défaillance (qui peuvent varier d’un cas `a un autre)? 3) Comment être sûr que les modèles définis seront assez robustes pour montrer une performance stable avec des entrées incertaines s’ écartant des expériences acquises, et seront suffisamment fiables pour intégrer des données inconnues (c’est `a dire les conditions de fonctionnement, les variations de l’ingénierie, etc.)? 4) Comment réaliser facilement une intégration sous des contraintes et des exigence industrielles? Ces questions sont des problèmes abordés dans cette thèse. Elles ont conduit à développer une nouvelle approche allant au-delà des limites des méthodes classiques de pronostic guidé par les données. / Prognostics and Health Management (PHM) aims at extending the life cycle of engineerin gassets, while reducing exploitation and maintenance costs. For this reason,prognostics is considered as a key process with future capabilities. Indeed, accurateestimates of the Remaining Useful Life (RUL) of an equipment enable defining furtherplan of actions to increase safety, minimize downtime, ensure mission completion andefficient production.Recent advances show that data-driven approaches (mainly based on machine learningmethods) are increasingly applied for fault prognostics. They can be seen as black-boxmodels that learn system behavior directly from Condition Monitoring (CM) data, usethat knowledge to infer its current state and predict future progression of failure. However,approximating the behavior of critical machinery is a challenging task that canresult in poor prognostics. As for understanding, some issues of data-driven prognosticsmodeling are highlighted as follows. 1) How to effectively process raw monitoringdata to obtain suitable features that clearly reflect evolution of degradation? 2) Howto discriminate degradation states and define failure criteria (that can vary from caseto case)? 3) How to be sure that learned-models will be robust enough to show steadyperformance over uncertain inputs that deviate from learned experiences, and to bereliable enough to encounter unknown data (i.e., operating conditions, engineering variations,etc.)? 4) How to achieve ease of application under industrial constraints andrequirements? Such issues constitute the problems addressed in this thesis and have ledto develop a novel approach beyond conventional methods of data-driven prognostics.

Au décours d’un AVC, près de 50% des patients vont développer un trouble de la cognition dans les six premiers mois suivant l’accident. Ces déficits ont la particularité de pouvoir être de natures différentes, en touchant plusieurs domaines cognitifs distincts, parfois simultanément. A l’aide de batteries de tests neuropsychologiques dédiés, ces altérations cognitives ont pu être largement décrites et caractérisées. En revanche, les mécanismes sous-jacents l’apparition de ces troubles sont encore mal compris.Grâce aux possibilités d’analyse structurelle et fonctionnelle du cerveau, l’imagerie par résonance magnétique (IRM) est une technique de plus en plus utilisée pour identifier de nouveaux marqueurs diagnostiques des maladies neurodégénératives. L’objectif principal de de travail de thèse était de mieux comprendre les mécanismes physiopathologiques impliqués dans l’apparition de troubles cognitifs post-AVC à l’aide de méthodologies avancées en IRM.La première étude est une étude transversale comportant un versant pré-clinique chez des rats ischémiés et un versant clinique chez des patients victimes d’un AVC. Chez l’Homme, les résultats ont montré des anomalies morphologiques de l’hippocampe ainsi que des anomalies structurelles du cortex entorhinal chez les patients présentant un déficit cognitif 6 mois après AVC. Chez le rongeur, l’imagerie a montré des déformations des contours de l’hippocampe chez les rats ischémiés présentant des anomalies cognitives 6 mois après occlusion de l’artère cérébrale moyenne.Nous avons ensuite analysé les anomalies de connectivité fonctionnelle spécifiques aux troubles cognitifs survenant dans les 6 mois après un AVC chez l’Homme car certains travaux ont démontré l’importance des anomalies de communication fonctionnelle dans l’apparition des troubles cognitifs post-AVC. Nous avons ainsi identifié un réseau fonctionnel spécifique organisé autour du lobe frontal supérieur et temporal. De plus, chaque fonction cognitive était associée à un motif spécifique de connexions fonctionnelles altérées.Enfin, à l’aide d’algorithmes d’apprentissage machine, nous avons montré que ce réseau fonctionnel impliqué dans la génèse des troubles cognitifs post-AVC était un excellent marqueur prédictif des altérations cognitives chez l’Homme 3 ans après l’AVC.Ainsi, les mesures morphométriques du lobe temporal médian et de connectivité fonctionnelle globale apparaissent comme des marqueurs IRM complémentaires dans la caractérisation des troubles cognitifs post-AVC. L’ensemble de ces résultats suggèrent ainsi que des mécanismes physiopathologiques spécifiques seraient impliqués dans la survenue de des troubles cognitifs, permettant d’envisager dans l’avenir des prises en charge personnalisées pour les fonctions cognitives des patients victimes d’AVC. / After a stroke, nearly 50% of the patients are prompt to develop cognitive disorders in the first 6 months. These deficits can be various, affecting distinct cognitive functions and sometimes simultaneously. Using specific cognitive battery, these disorders can be well described and characterized. However, the mechanisms leading to the development of these cognitive impairments are poorly understood.In that sense, magnetic resonance imaging offers many possibilities for the detection of occurring cognitive disorders after a stroke. The aim of this study of to better define imaging markers that could help to better understand the physiopathology and potentially, using advances methods, predict the long-term outcome of stroke patients.First, a translational study highlighted morphological deformations of hippocampus and structural changes of entorhinal cortex in patients with a cognitive disorder 6 month after stroke compared to patients without. These alterations have also been found in a rat model of ischemia, that presented deformations of hippocampus 6 months after the ischemia compared to control animals.Second, many imaging studies reported that the post-stroke cognitive disorders could be more associated with spread dysfunctions rather than focal alteration at the lesion site. In that sense, we analyzed the functional alterations that could exist in patients with a cognitive disorder 6 months post-stroke compared to patients without. We then identified a functional network that seemed to be organized around regions in superior frontal and temporal lobes. Moreover, each cognitive function presented specific pattern of correlated connections in this network.Lastly, using machine learning algorithms, we showed that this identified functional network 6 months post-stroke can predict the diagnosis of cognitive impairment 30 months later, and also predict alterations of specific cognitive domains at the same time.Then, morphological measures of the medial temporal lobe and the global functional connectivity analyses appeared to be complementary MRI markers for the characterization of patients developing a cognitive disorder after stroke. All these results suggest then that specific physiopathological mechanisms could be involved in the appearance of post-stroke cognitive disorders, permitting to imagine potential new personalized care of post-stroke cognitive disorders.

Extreme Learning Machine (ELM) is a recently discovered way of training Single Layer Feed-forward Neural Networks with an explicitly given solution, which exists because the input weights and biases are generated randomly and never change. The method in general achieves performance comparable to Error Back-Propagation, but the training time is up to 5 orders of magnitude smaller. Despite a random initialization, the regularization procedures explained in the thesis ensure consistently good results. While the general methodology of ELMs is well developed, the sheer speed of the method enables its un-typical usage for state-of-the-art techniques based on repetitive model re-training and re-evaluation. Three of such techniques are explained in the third chapter: a way of visualizing high-dimensional data onto a provided fixed set of visualization points, an approach for detecting samples in a dataset with incorrect labels (mistakenly assigned, mistyped or a low confidence), and a way of computing confidence intervals for ELM predictions. All three methods prove useful, and allow even more applications in the future. ELM method is a promising basis for dealing with Big Data, because it naturally deals with the problem of large data size. An adaptation of ELM to Big Data problems, and a corresponding toolbox (published and freely available) are described in chapter 4. An adaptation includes an iterative solution of ELM which satisfies a limited computer memory constraints and allows for a convenient parallelization. Other tools are GPU-accelerated computations and support for a convenient huge data storage format. The chapter also provides two real-world examples of dealing with Big Data using ELMs, which present other problems of Big Data such as veracity and velocity, and solutions to them in the particular problem context.

No description available.

No description available.

No description available.

Thesis (Ph.D.) OGI School of Science & Engineering at OHSU, June 2008. / Includes bibliographical references (leaves 138-145).

Thesis (M.S.)--Oregon State University, 2004. / Printout. Includes bibliographical references (leaves 45-46). Also available on the World Wide Web.