Bidirectional Helmholtz Machines

L'entraînement sans surveillance efficace et inférence dans les modèles génératifs profonds reste un problème difficile. Une approche assez simple, la machine de Helmholtz, consiste à entraîner du haut vers le bas un modèle génératif dirigé qui sera utilisé plus tard pour l'inférence approximative. Des résultats récents suggèrent que de meilleurs modèles génératifs peuvent être obtenus par de meilleures procédures d'inférence approximatives. Au lieu d'améliorer la procédure d'inférence, nous proposons ici un nouveau modèle, la machine de Helmholtz bidirectionnelle, qui garantit qu'on peut calculer efficacement les distributions de haut-vers-bas et de bas-vers-haut. Nous y parvenons en interprétant à les modèles haut-vers-bas et bas-vers-haut en tant que distributions d'inférence approximative, puis ensuite en définissant la distribution du modèle comme étant la moyenne géométrique de ces deux distributions. Nous dérivons une borne inférieure pour la vraisemblance de ce modèle, et nous démontrons que l'optimisation de cette borne se comporte en régulisateur. Ce régularisateur sera tel que la distance de Bhattacharyya sera minisée entre les distributions approximatives haut-vers-bas et bas-vers-haut. Cette approche produit des résultats de pointe en terme de modèles génératifs qui favorisent les réseaux significativement plus profonds. Elle permet aussi une inférence approximative amérliorée par plusieurs ordres de grandeur. De plus, nous introduisons un modèle génératif profond basé sur les modèles BiHM pour l'entraînement semi-supervisé.

Bidirectional Helmholtz Machines

L'entraînement sans surveillance efficace et inférence dans les modèles génératifs profonds reste un problème difficile. Une approche assez simple, la machine de Helmholtz, consiste à entraîner du haut vers le bas un modèle génératif dirigé qui sera utilisé plus tard pour l'inférence approximative. Des résultats récents suggèrent que de meilleurs modèles génératifs peuvent être obtenus par de meilleures procédures d'inférence approximatives. Au lieu d'améliorer la procédure d'inférence, nous proposons ici un nouveau modèle, la machine de Helmholtz bidirectionnelle, qui garantit qu'on peut calculer efficacement les distributions de haut-vers-bas et de bas-vers-haut. Nous y parvenons en interprétant à les modèles haut-vers-bas et bas-vers-haut en tant que distributions d'inférence approximative, puis ensuite en définissant la distribution du modèle comme étant la moyenne géométrique de ces deux distributions. Nous dérivons une borne inférieure pour la vraisemblance de ce modèle, et nous démontrons que l'optimisation de cette borne se comporte en régulisateur. Ce régularisateur sera tel que la distance de Bhattacharyya sera minisée entre les distributions approximatives haut-vers-bas et bas-vers-haut. Cette approche produit des résultats de pointe en terme de modèles génératifs qui favorisent les réseaux significativement plus profonds. Elle permet aussi une inférence approximative amérliorée par plusieurs ordres de grandeur. De plus, nous introduisons un modèle génératif profond basé sur les modèles BiHM pour l'entraînement semi-supervisé.

Kernel machines for epilepsy diagnosis via EEG signal classification: A comparative study

Objective: We carry out a systematic assessment on a suite of kernel-based learning machines while coping with the task of epilepsy diagnosis through automatic electroencephalogram (EEG) signal classification. Methods and materials: The kernel machines investigated include the standard support vector machine (SVM), the least squares SVM, the Lagrangian SVM, the smooth SVM, the proximal SVM, and the relevance vector machine. An extensive series of experiments was conducted on publicly available data, whose clinical EEG recordings were obtained from five normal subjects and five epileptic patients. The performance levels delivered by the different kernel machines are contrasted in terms of the criteria of predictive accuracy, sensitivity to the kernel function/parameter value, and sensitivity to the type of features extracted from the signal. For this purpose, 26 values for the kernel parameter (radius) of two well-known kernel functions (namely. Gaussian and exponential radial basis functions) were considered as well as 21 types of features extracted from the EEG signal, including statistical values derived from the discrete wavelet transform, Lyapunov exponents, and combinations thereof. Results: We first quantitatively assess the impact of the choice of the wavelet basis on the quality of the features extracted. Four wavelet basis functions were considered in this study. Then, we provide the average accuracy (i.e., cross-validation error) values delivered by 252 kernel machine configurations; in particular, 40%/35% of the best-calibrated models of the standard and least squares SVMs reached 100% accuracy rate for the two kernel functions considered. Moreover, we show the sensitivity profiles exhibited by a large sample of the configurations whereby one can visually inspect their levels of sensitiveness to the type of feature and to the kernel function/parameter value. Conclusions: Overall, the results evidence that all kernel machines are competitive in terms of accuracy, with the standard and least squares SVMs prevailing more consistently. Moreover, the choice of the kernel function and parameter value as well as the choice of the feature extractor are critical decisions to be taken, albeit the choice of the wavelet family seems not to be so relevant. Also, the statistical values calculated over the Lyapunov exponents were good sources of signal representation, but not as informative as their wavelet counterparts. Finally, a typical sensitivity profile has emerged among all types of machines, involving some regions of stability separated by zones of sharp variation, with some kernel parameter values clearly associated with better accuracy rates (zones of optimality). (C) 2011 Elsevier B.V. All rights reserved.

Support vector machines for detecting age-related changes in running kinematics

Age-related changes in running kinematics have been reported in the literature using classical inferential statistics. However, this approach has been hampered by the increased number of biomechanical gait variables reported and subsequently the lack of differences presented in these studies. Data mining techniques have been applied in recent biomedical studies to solve this problem using a more general approach. In the present work, we re-analyzed lower extremity running kinematic data of 17 young and 17 elderly male runners using the Support Vector Machine (SVM) classification approach. In total, 31 kinematic variables were extracted to train the classification algorithm and test the generalized performance. The results revealed different accuracy rates across three different kernel methods adopted in the classifier, with the linear kernel performing the best. A subsequent forward feature selection algorithm demonstrated that with only six features, the linear kernel SVM achieved 100% classification performance rate, showing that these features provided powerful combined information to distinguish age groups. The results of the present work demonstrate potential in applying this approach to improve knowledge about the age-related differences in running gait biomechanics and encourages the use of the SVM in other clinical contexts. (C) 2010 Elsevier Ltd. All rights reserved.

Normal Versus Pathological Voice Signals Using Wavelet Analysis and Least Squares Support-Vector Machines

State of Sao Paulo Research Foundation (FAPESP)

Remote control of CNC machines using the CyberOPC communication system over public networks

During the last few years, the evolution of fieldbus and computers networks allowed the integration of different communication systems involving both production single cells and production cells, as well as other systems for business intelligence, supervision and control. Several well-adopted communication technologies exist today for public and non-public networks. Since most of the industrial applications are time-critical, the requirements of communication systems for remote control differ from common applications for computer networks accessing the Internet, such as Web, e-mail and file transfer. The solution proposed and outlined in this work is called CyberOPC. It includes the study and the implementation of a new open communication system for remote control of industrial CNC machines, making the transmission delay for time-critical control data shorter than other OPC-based solutions, and fulfilling cyber security requirements.

Comparison Between Voltage Control Structures of Synchronous Machines

This paper compares the behaviour of two different control structures of automatic voltage regulators of synchronous machines equipped with static excitation systems. These systems have a fully controlled thyristor bridge that supplies DC current to the rotor winding. The rectifier bridge is fed by the stator terminals through a step-down transformer. The first control structure, named ""Direct Control"", has a single proportional-integral (PI) regulator that compares stator voltage setpoint with measured voltage and acts directly on the thyristor bridge`s firing angle. This control structure is usually employed in commercial excitation systems for hydrogenerators. The second structure, named ""Cascade Control"", was inspired on control loops of commercial DC motor drives. Such drives employ two PIs in a cascade arrangement, the external PI deals with the motor speed while the internal one regulates the armature current. In the adaptation proposed, the external PI compares setpoint with the actual stator voltage and produces the setpoint to the internal PI-loop which controls the field current.

Formal derivation of finite state machines for class testing

Previous work on generating state machines for the purpose of class testing has not been formally based. There has also been work on deriving state machines from formal specifications for testing non-object-oriented software. We build on this work by presenting a method for deriving a state machine for testing purposes from a formal specification of the class under test. We also show how the resulting state machine can be used as the basis for a test suite developed and executed using an existing framework for class testing. To derive the state machine, we identify the states and possible interactions of the operations of the class under test. The Test Template Framework is used to formally derive the states from the Object-Z specification of the class under test. The transitions of the finite state machine are calculated from the derived states and the class's operations. The formally derived finite state machine is transformed to a ClassBench testgraph, which is used as input to the ClassBench framework to test a C++ implementation of the class. The method is illustrated using a simple bounded queue example.

An fMRI Normative Database for Connectivity Networks Using One-Class Support Vector Machines

The application of functional magnetic resonance imaging (fMRI) in neuroscience studies has increased enormously in the last decade. Although primarily used to map brain regions activated by specific stimuli, many studies have shown that fMRI can also be useful in identifying interactions between brain regions (functional and effective connectivity). Despite the widespread use of fMRI as a research tool, clinical applications of brain connectivity as studied by fMRI are not well established. One possible explanation is the lack of normal pattern, and intersubject variability-two variables that are still largely uncharacterized in most patient populations of interest. In the current study, we combine the identification of functional connectivity networks extracted by using Spearman partial correlation with the use of a one-class support vector machine in order construct a normative database. An application of this approach is illustrated using an fMRI dataset of 43 healthy Subjects performing a visual working memory task. In addition, the relationships between the results obtained and behavioral data are explored. Hum Brain Mapp 30:1068-1076, 2009. (C) 2008 Wiley-Liss. Inc.

The impact of functional connectivity changes on support vector machines mapping of fMRI data

Functional magnetic resonance imaging (fMRI) is currently one of the most widely used methods for studying human brain function in vivo. Although many different approaches to fMRI analysis are available, the most widely used methods employ so called ""mass-univariate"" modeling of responses in a voxel-by-voxel fashion to construct activation maps. However, it is well known that many brain processes involve networks of interacting regions and for this reason multivariate analyses might seem to be attractive alternatives to univariate approaches. The current paper focuses on one multivariate application of statistical learning theory: the statistical discrimination maps (SDM) based on support vector machine, and seeks to establish some possible interpretations when the results differ from univariate `approaches. In fact, when there are changes not only on the activation level of two conditions but also on functional connectivity, SDM seems more informative. We addressed this question using both simulations and applications to real data. We have shown that the combined use of univariate approaches and SDM yields significant new insights into brain activations not available using univariate methods alone. In the application to a visual working memory fMRI data, we demonstrated that the interaction among brain regions play a role in SDM`s power to detect discriminative voxels. (C) 2008 Elsevier B.V. All rights reserved.

Comparação de desempenho entre a formulação direta do Método dos Elementos de Contorno com Funções Radiais e o Método dos Elementos Finitos em problemas de Poisson e Helmholtz

O presente trabalho objetiva avaliar o desempenho do MECID (Método dos Elementos de Contorno com Interpolação Direta) para resolver o termo integral referente à inércia na Equação de Helmholtz e, deste modo, permitir a modelagem do Problema de Autovalor assim como calcular as frequências naturais, comparando-o com os resultados obtidos pelo MEF (Método dos Elementos Finitos), gerado pela Formulação Clássica de Galerkin. Em primeira instância, serão abordados alguns problemas governados pela equação de Poisson, possibilitando iniciar a comparação de desempenho entre os métodos numéricos aqui abordados. Os problemas resolvidos se aplicam em diferentes e importantes áreas da engenharia, como na transmissão de calor, no eletromagnetismo e em problemas elásticos particulares. Em termos numéricos, sabe-se das dificuldades existentes na aproximação precisa de distribuições mais complexas de cargas, fontes ou sorvedouros no interior do domínio para qualquer técnica de contorno. No entanto, este trabalho mostra que, apesar de tais dificuldades, o desempenho do Método dos Elementos de Contorno é superior, tanto no cálculo da variável básica, quanto na sua derivada. Para tanto, são resolvidos problemas bidimensionais referentes a membranas elásticas, esforços em barras devido ao peso próprio e problemas de determinação de frequências naturais em problemas acústicos em domínios fechados, dentre outros apresentados, utilizando malhas com diferentes graus de refinamento, além de elementos lineares com funções de bases radiais para o MECID e funções base de interpolação polinomial de grau (um) para o MEF. São geradas curvas de desempenho através do cálculo do erro médio percentual para cada malha, demonstrando a convergência e a precisão de cada método. Os resultados também são comparados com as soluções analíticas, quando disponíveis, para cada exemplo resolvido neste trabalho.

Low-cost vibration sensors: tendencies and applications in condition monitoring of machines and structures

Trabalho Final de Mestrado para obtenção do grau de Mestre em Engenharia Mecânica