Bernstein polynomials in element-free Galerkin method

In the recent decades, meshless methods (MMs), like the element-free Galerkin method (EFGM), have been widely studied and interesting results have been reached when solving partial differential equations. However, such solutions show a problem around boundary conditions, where the accuracy is not adequately achieved. This is caused by the use of moving least squares or residual kernel particle method methods to obtain the shape functions needed in MM, since such methods are good enough in the inner of the integration domains, but not so accurate in boundaries. This way, Bernstein curves, which are a partition of unity themselves,can solve this problem with the same accuracy in the inner area of the domain and at their boundaries.

Modelos sin malla en simulación numérica de estructuras aeroespaciales

La presente Tesis Doctoral aborda la introducción de la Partición de Unidad de Bernstein en la forma débil de Galerkin para la resolución de problemas de condiciones de contorno en el ámbito del análisis estructural. La familia de funciones base de Bernstein conforma un sistema generador del espacio de funciones polinómicas que permite construir aproximaciones numéricas para las que no se requiere la existencia de malla: las funciones de forma, de soporte global, dependen únicamente del orden de aproximación elegido y de la parametrización o mapping del dominio, estando las posiciones nodales implícitamente definidas. El desarrollo de la formulación está precedido por una revisión bibliográfica que, con su punto de partida en el Método de Elementos Finitos, recorre las principales técnicas de resolución sin malla de Ecuaciones Diferenciales en Derivadas Parciales, incluyendo los conocidos como Métodos Meshless y los métodos espectrales. En este contexto, en la Tesis se somete la aproximación Bernstein-Galerkin a validación en tests uni y bidimensionales clásicos de la Mecánica Estructural. Se estudian aspectos de la implementación tales como la consistencia, la capacidad de reproducción, la naturaleza no interpolante en la frontera, el planteamiento con refinamiento h-p o el acoplamiento con otras aproximaciones numéricas. Un bloque importante de la investigación se dedica al análisis de estrategias de optimización computacional, especialmente en lo referente a la reducción del tiempo de máquina asociado a la generación y operación con matrices llenas. Finalmente, se realiza aplicación a dos casos de referencia de estructuras aeronáuticas, el análisis de esfuerzos en un angular de material anisotrópico y la evaluación de factores de intensidad de esfuerzos de la Mecánica de Fractura mediante un modelo con Partición de Unidad de Bernstein acoplada a una malla de elementos finitos. ABSTRACT This Doctoral Thesis deals with the introduction of Bernstein Partition of Unity into Galerkin weak form to solve boundary value problems in the field of structural analysis. The family of Bernstein basis functions constitutes a spanning set of the space of polynomial functions that allows the construction of numerical approximations that do not require the presence of a mesh: the shape functions, which are globally-supported, are determined only by the selected approximation order and the parametrization or mapping of the domain, being the nodal positions implicitly defined. The exposition of the formulation is preceded by a revision of bibliography which begins with the review of the Finite Element Method and covers the main techniques to solve Partial Differential Equations without the use of mesh, including the so-called Meshless Methods and the spectral methods. In this context, in the Thesis the Bernstein-Galerkin approximation is subjected to validation in one- and two-dimensional classic benchmarks of Structural Mechanics. Implementation aspects such as consistency, reproduction capability, non-interpolating nature at boundaries, h-p refinement strategy or coupling with other numerical approximations are studied. An important part of the investigation focuses on the analysis and optimization of computational efficiency, mainly regarding the reduction of the CPU cost associated with the generation and handling of full matrices. Finally, application to two reference cases of aeronautic structures is performed: the stress analysis in an anisotropic angle part and the evaluation of stress intensity factors of Fracture Mechanics by means of a coupled Bernstein Partition of Unity - finite element mesh model.

Approximate entropy of network parameters

We study the notion of approximate entropy within the framework of network theory. Approximate entropy is an uncertainty measure originally proposed in the context of dynamical systems and time series. We first define a purely structural entropy obtained by computing the approximate entropy of the so-called slide sequence. This is a surrogate of the degree sequence and it is suggested by the frequency partition of a graph. We examine this quantity for standard scale-free and Erdös-Rényi networks. By using classical results of Pincus, we show that our entropy measure often converges with network size to a certain binary Shannon entropy. As a second step, with specific attention to networks generated by dynamical processes, we investigate approximate entropy of horizontal visibility graphs. Visibility graphs allow us to naturally associate with a network the notion of temporal correlations, therefore providing the measure a dynamical garment. We show that approximate entropy distinguishes visibility graphs generated by processes with different complexity. The result probes to a greater extent these networks for the study of dynamical systems. Applications to certain biological data arising in cancer genomics are finally considered in the light of both approaches.

Overal reaction rate and stoichiometry for thermal decomposition of hydrazine

An asymptotic analysîs of the Eberstein-Glassman kinetic mechanlsm for the thermal décomposition of hydrazine is carried out. It is shown that at températures near 800°K and near 1000°K,and for hydrazine molar fractions of the order of unity, 10-2 the entire kinetics reduces to a single, overall reaction. Characteristic times for the chemical relaxation of ail active, intermediate species produced in the décomposition, and for the overall reaction, are obtained. Explicit expressions for the overall reaction rate and stoichiometry are given as functions of température, total molar concentration (or pressure)and hydrazine molar fraction. Approximate, patched expressions can then be obtained for values of température and hydrazine molar fraction between 750 and 1000°K, and 1 and 10-3 respectively.

A dynamic programming approach to the formulation and solution of finite element equations

A method for formulating and algorithmically solving the equations of finite element problems is presented. The method starts with a parametric partition of the domain in juxtaposed strips that permits sweeping the whole region by a sequential addition (or removal) of adjacent strips. The solution of the difference equations constructed over that grid proceeds along with the addition removal of strips in a manner resembling the transfer matrix approach, except that different rules of composition that lead to numerically stable algorithms are used for the stiffness matrices of the strips. Dynamic programming and invariant imbedding ideas underlie the construction of such rules of composition. Among other features of interest, the present methodology provides to some extent the analyst's control over the type and quantity of data to be computed. In particular, the one-sweep method presented in Section 9, with no apparent counterpart in standard methods, appears to be very efficient insofar as time and storage is concerned. The paper ends with the presentation of a numerical example

Dynamic topic-based adaptation of language models: a comparison between different approaches

This paper presents a dynamic LM adaptation based on the topic that has been identified on a speech segment. We use LSA and the given topic labels in the training dataset to obtain and use the topic models. We propose a dynamic language model adaptation to improve the recognition performance in "a two stages" AST system. The final stage makes use of the topic identification with two variants: the first on uses just the most probable topic and the other one depends on the relative distances of the topics that have been identified. We perform the adaptation of the LM as a linear interpolation between a background model and topic-based LM. The interpolation weight id dynamically adapted according to different parameters. The proposed method is evaluated on the Spanish partition of the EPPS speech database. We achieved a relative reduction in WER of 11.13% over the baseline system which uses a single blackground LM.

Strong Connectivity in Directed Hypergraphs and its Application to the Atomic Decomposition of Ontologies

Los hipergrafos dirigidos se han empleado en problemas relacionados con lógica proposicional, bases de datos relacionales, linguística computacional y aprendizaje automático. Los hipergrafos dirigidos han sido también utilizados como alternativa a los grafos (bipartitos) dirigidos para facilitar el estudio de las interacciones entre componentes de sistemas complejos que no pueden ser fácilmente modelados usando exclusivamente relaciones binarias. En este contexto, este tipo de representación es conocida como hiper-redes. Un hipergrafo dirigido es una generalización de un grafo dirigido especialmente adecuado para la representación de relaciones de muchos a muchos. Mientras que una arista en un grafo dirigido define una relación entre dos de sus nodos, una hiperarista en un hipergrafo dirigido define una relación entre dos conjuntos de sus nodos. La conexión fuerte es una relación de equivalencia que divide el conjunto de nodos de un hipergrafo dirigido en particiones y cada partición define una clase de equivalencia conocida como componente fuertemente conexo. El estudio de los componentes fuertemente conexos de un hipergrafo dirigido puede ayudar a conseguir una mejor comprensión de la estructura de este tipo de hipergrafos cuando su tamaño es considerable. En el caso de grafo dirigidos, existen algoritmos muy eficientes para el cálculo de los componentes fuertemente conexos en grafos de gran tamaño. Gracias a estos algoritmos, se ha podido averiguar que la estructura de la WWW tiene forma de “pajarita”, donde más del 70% del los nodos están distribuidos en tres grandes conjuntos y uno de ellos es un componente fuertemente conexo. Este tipo de estructura ha sido también observada en redes complejas en otras áreas como la biología. Estudios de naturaleza similar no han podido ser realizados en hipergrafos dirigidos porque no existe algoritmos capaces de calcular los componentes fuertemente conexos de este tipo de hipergrafos. En esta tesis doctoral, hemos investigado como calcular los componentes fuertemente conexos de un hipergrafo dirigido. En concreto, hemos desarrollado dos algoritmos para este problema y hemos determinado que son correctos y cuál es su complejidad computacional. Ambos algoritmos han sido evaluados empíricamente para comparar sus tiempos de ejecución. Para la evaluación, hemos producido una selección de hipergrafos dirigidos generados de forma aleatoria inspirados en modelos muy conocidos de grafos aleatorios como Erdos-Renyi, Newman-Watts-Strogatz and Barabasi-Albert. Varias optimizaciones para ambos algoritmos han sido implementadas y analizadas en la tesis. En concreto, colapsar los componentes fuertemente conexos del grafo dirigido que se puede construir eliminando ciertas hiperaristas complejas del hipergrafo dirigido original, mejora notablemente los tiempos de ejecucion de los algoritmos para varios de los hipergrafos utilizados en la evaluación. Aparte de los ejemplos de aplicación mencionados anteriormente, los hipergrafos dirigidos han sido también empleados en el área de representación de conocimiento. En concreto, este tipo de hipergrafos se han usado para el cálculo de módulos de ontologías. Una ontología puede ser definida como un conjunto de axiomas que especifican formalmente un conjunto de símbolos y sus relaciones, mientras que un modulo puede ser entendido como un subconjunto de axiomas de la ontología que recoge todo el conocimiento que almacena la ontología sobre un conjunto especifico de símbolos y sus relaciones. En la tesis nos hemos centrado solamente en módulos que han sido calculados usando la técnica de localidad sintáctica. Debido a que las ontologías pueden ser muy grandes, el cálculo de módulos puede facilitar las tareas de re-utilización y mantenimiento de dichas ontologías. Sin embargo, analizar todos los posibles módulos de una ontología es, en general, muy costoso porque el numero de módulos crece de forma exponencial con respecto al número de símbolos y de axiomas de la ontología. Afortunadamente, los axiomas de una ontología pueden ser divididos en particiones conocidas como átomos. Cada átomo representa un conjunto máximo de axiomas que siempre aparecen juntos en un modulo. La decomposición atómica de una ontología es definida como un grafo dirigido de tal forma que cada nodo del grafo corresponde con un átomo y cada arista define una dependencia entre una pareja de átomos. En esta tesis introducimos el concepto de“axiom dependency hypergraph” que generaliza el concepto de descomposición atómica de una ontología. Un modulo en una ontología correspondería con un componente conexo en este tipo de hipergrafos y un átomo de una ontología con un componente fuertemente conexo. Hemos adaptado la implementación de nuestros algoritmos para que funcionen también con axiom dependency hypergraphs y poder de esa forma calcular los átomos de una ontología. Para demostrar la viabilidad de esta idea, hemos incorporado nuestros algoritmos en una aplicación que hemos desarrollado para la extracción de módulos y la descomposición atómica de ontologías. A la aplicación la hemos llamado HyS y hemos estudiado sus tiempos de ejecución usando una selección de ontologías muy conocidas del área biomédica, la mayoría disponibles en el portal de Internet NCBO. Los resultados de la evaluación muestran que los tiempos de ejecución de HyS son mucho mejores que las aplicaciones más rápidas conocidas. ABSTRACT Directed hypergraphs are an intuitive modelling formalism that have been used in problems related to propositional logic, relational databases, computational linguistic and machine learning. Directed hypergraphs are also presented as an alternative to directed (bipartite) graphs to facilitate the study of the interactions between components of complex systems that cannot naturally be modelled as binary relations. In this context, they are known as hyper-networks. A directed hypergraph is a generalization of a directed graph suitable for representing many-to-many relationships. While an edge in a directed graph defines a relation between two nodes of the graph, a hyperedge in a directed hypergraph defines a relation between two sets of nodes. Strong-connectivity is an equivalence relation that induces a partition of the set of nodes of a directed hypergraph into strongly-connected components. These components can be collapsed into single nodes. As result, the size of the original hypergraph can significantly be reduced if the strongly-connected components have many nodes. This approach might contribute to better understand how the nodes of a hypergraph are connected, in particular when the hypergraphs are large. In the case of directed graphs, there are efficient algorithms that can be used to compute the strongly-connected components of large graphs. For instance, it has been shown that the macroscopic structure of the World Wide Web can be represented as a “bow-tie” diagram where more than 70% of the nodes are distributed into three large sets and one of these sets is a large strongly-connected component. This particular structure has been also observed in complex networks in other fields such as, e.g., biology. Similar studies cannot be conducted in a directed hypergraph because there does not exist any algorithm for computing the strongly-connected components of the hypergraph. In this thesis, we investigate ways to compute the strongly-connected components of directed hypergraphs. We present two new algorithms and we show their correctness and computational complexity. One of these algorithms is inspired by Tarjan’s algorithm for directed graphs. The second algorithm follows a simple approach to compute the stronglyconnected components. This approach is based on the fact that two nodes of a graph that are strongly-connected can also reach the same nodes. In other words, the connected component of each node is the same. Both algorithms are empirically evaluated to compare their performances. To this end, we have produced a selection of random directed hypergraphs inspired by existent and well-known random graphs models like Erd˝os-Renyi and Newman-Watts-Strogatz. Besides the application examples that we mentioned earlier, directed hypergraphs have also been employed in the field of knowledge representation. In particular, they have been used to compute the modules of an ontology. An ontology is defined as a collection of axioms that provides a formal specification of a set of terms and their relationships; and a module is a subset of an ontology that completely captures the meaning of certain terms as defined in the ontology. In particular, we focus on the modules computed using the notion of syntactic locality. As ontologies can be very large, the computation of modules facilitates the reuse and maintenance of these ontologies. Analysing all modules of an ontology, however, is in general not feasible as the number of modules grows exponentially in the number of terms and axioms of the ontology. Nevertheless, the modules can succinctly be represented using the Atomic Decomposition of an ontology. Using this representation, an ontology can be partitioned into atoms, which are maximal sets of axioms that co-occur in every module. The Atomic Decomposition is then defined as a directed graph such that each node correspond to an atom and each edge represents a dependency relation between two atoms. In this thesis, we introduce the notion of an axiom dependency hypergraph which is a generalization of the atomic decomposition of an ontology. A module in the ontology corresponds to a connected component in the hypergraph, and the atoms of the ontology to the strongly-connected components. We apply our algorithms for directed hypergraphs to axiom dependency hypergraphs and in this manner, we compute the atoms of an ontology. To demonstrate the viability of this approach, we have implemented the algorithms in the application HyS which computes the modules of ontologies and calculate their atomic decomposition. In the thesis, we provide an experimental evaluation of HyS with a selection of large and prominent biomedical ontologies, most of which are available in the NCBO Bioportal. HyS outperforms state-of-the-art implementations in the tasks of extracting modules and computing the atomic decomposition of these ontologies.

GTH-UPM system for search on speech evaluation

This paper describes the GTH-UPM system for the Albayzin 2014 Search on Speech Evaluation. Teh evaluation task consists of searching a list of terms/queries in audio files. The GTH-UPM system we are presenting is based on a LVCSR (Large Vocabulary Continuous Speech Recognition) system. We have used MAVIR corpus and the Spanish partition of the EPPS (European Parliament Plenary Sessions) database for training both acoustic and language models. The main effort has been focused on lexicon preparation and text selection for the language model construction. The system makes use of different lexicon and language models depending on the task that is performed. For the best configuration of the system on the development set, we have obtained a FOM of 75.27 for the deyword spotting task.

Parallel Computer Vision Algorithms for Graphics Processing Units

La evolución de los teléfonos móviles inteligentes, dotados de cámaras digitales, está provocando una creciente demanda de aplicaciones cada vez más complejas que necesitan algoritmos de visión artificial en tiempo real; puesto que el tamaño de las señales de vídeo no hace sino aumentar y en cambio el rendimiento de los procesadores de un solo núcleo se ha estancado, los nuevos algoritmos que se diseñen para visión artificial han de ser paralelos para poder ejecutarse en múltiples procesadores y ser computacionalmente escalables. Una de las clases de procesadores más interesantes en la actualidad se encuentra en las tarjetas gráficas (GPU), que son dispositivos que ofrecen un alto grado de paralelismo, un excelente rendimiento numérico y una creciente versatilidad, lo que los hace interesantes para llevar a cabo computación científica. En esta tesis se exploran dos aplicaciones de visión artificial que revisten una gran complejidad computacional y no pueden ser ejecutadas en tiempo real empleando procesadores tradicionales. En cambio, como se demuestra en esta tesis, la paralelización de las distintas subtareas y su implementación sobre una GPU arrojan los resultados deseados de ejecución con tasas de refresco interactivas. Asimismo, se propone una técnica para la evaluación rápida de funciones de complejidad arbitraria especialmente indicada para su uso en una GPU. En primer lugar se estudia la aplicación de técnicas de síntesis de imágenes virtuales a partir de únicamente dos cámaras lejanas y no paralelas—en contraste con la configuración habitual en TV 3D de cámaras cercanas y paralelas—con información de color y profundidad. Empleando filtros de mediana modificados para la elaboración de un mapa de profundidad virtual y proyecciones inversas, se comprueba que estas técnicas son adecuadas para una libre elección del punto de vista. Además, se demuestra que la codificación de la información de profundidad con respecto a un sistema de referencia global es sumamente perjudicial y debería ser evitada. Por otro lado se propone un sistema de detección de objetos móviles basado en técnicas de estimación de densidad con funciones locales. Este tipo de técnicas es muy adecuada para el modelado de escenas complejas con fondos multimodales, pero ha recibido poco uso debido a su gran complejidad computacional. El sistema propuesto, implementado en tiempo real sobre una GPU, incluye propuestas para la estimación dinámica de los anchos de banda de las funciones locales, actualización selectiva del modelo de fondo, actualización de la posición de las muestras de referencia del modelo de primer plano empleando un filtro de partículas multirregión y selección automática de regiones de interés para reducir el coste computacional. Los resultados, evaluados sobre diversas bases de datos y comparados con otros algoritmos del estado del arte, demuestran la gran versatilidad y calidad de la propuesta. Finalmente se propone un método para la aproximación de funciones arbitrarias empleando funciones continuas lineales a tramos, especialmente indicada para su implementación en una GPU mediante el uso de las unidades de filtraje de texturas, normalmente no utilizadas para cómputo numérico. La propuesta incluye un riguroso análisis matemático del error cometido en la aproximación en función del número de muestras empleadas, así como un método para la obtención de una partición cuasióptima del dominio de la función para minimizar el error. ABSTRACT The evolution of smartphones, all equipped with digital cameras, is driving a growing demand for ever more complex applications that need to rely on real-time computer vision algorithms. However, video signals are only increasing in size, whereas the performance of single-core processors has somewhat stagnated in the past few years. Consequently, new computer vision algorithms will need to be parallel to run on multiple processors and be computationally scalable. One of the most promising classes of processors nowadays can be found in graphics processing units (GPU). These are devices offering a high parallelism degree, excellent numerical performance and increasing versatility, which makes them interesting to run scientific computations. In this thesis, we explore two computer vision applications with a high computational complexity that precludes them from running in real time on traditional uniprocessors. However, we show that by parallelizing subtasks and implementing them on a GPU, both applications attain their goals of running at interactive frame rates. In addition, we propose a technique for fast evaluation of arbitrarily complex functions, specially designed for GPU implementation. First, we explore the application of depth-image–based rendering techniques to the unusual configuration of two convergent, wide baseline cameras, in contrast to the usual configuration used in 3D TV, which are narrow baseline, parallel cameras. By using a backward mapping approach with a depth inpainting scheme based on median filters, we show that these techniques are adequate for free viewpoint video applications. In addition, we show that referring depth information to a global reference system is ill-advised and should be avoided. Then, we propose a background subtraction system based on kernel density estimation techniques. These techniques are very adequate for modelling complex scenes featuring multimodal backgrounds, but have not been so popular due to their huge computational and memory complexity. The proposed system, implemented in real time on a GPU, features novel proposals for dynamic kernel bandwidth estimation for the background model, selective update of the background model, update of the position of reference samples of the foreground model using a multi-region particle filter, and automatic selection of regions of interest to reduce computational cost. The results, evaluated on several databases and compared to other state-of-the-art algorithms, demonstrate the high quality and versatility of our proposal. Finally, we propose a general method for the approximation of arbitrarily complex functions using continuous piecewise linear functions, specially formulated for GPU implementation by leveraging their texture filtering units, normally unused for numerical computation. Our proposal features a rigorous mathematical analysis of the approximation error in function of the number of samples, as well as a method to obtain a suboptimal partition of the domain of the function to minimize approximation error.

Source terms for calculations of vaporizing and burning fuel sprays with non-unity Lewis numbers in gases with temperature-dependent thermal conductivities

Liquid-fueled burners are used in a number of propulsion devices ranging from internal combustion engines to gas turbines. The structure of spray flames is quite complex and involves a wide range of time and spatial scales in both premixed and non-premixed modes (Williams 1965; Luo et al. 2011). A number of spray-combustion regimes can be observed experimentally in canonical scenarios of practical relevance such as counterflow diffusion flames (Li 1997), as sketched in figure 1, and for which different microscalemodelling strategies are needed. In this study, source terms for the conservation equations are calculated for heating, vaporizing and burning sprays in the single-droplet combustion regime. The present analysis provides extended formulation for source terms, which include non-unity Lewis numbers and variable thermal conductivities.