Desarrollo, implementación y análisis de un modelo cinemático local de velocidades para Costa Rica basado en la integración y homogenización de datos de estaciones GNSS continuas

El primer procesamiento estricto realizado con el software científico Bernese y contemplando las más estrictas normas de cálculo recomendadas internacionalmente, permitió obtener un campo puntual de alta exactitud, basado en la integración y estandarización de los datos de una red GPS ubicada en Costa Rica. Este procesamiento contempló un total de 119 semanas de datos diarios, es decir unos 2,3 años, desde enero del año 2009 hasta abril del año 2011, para un total de 30 estaciones GPS, de las cuales 22 están ubicadas en el territorio nacional de Costa Rica y 8 internaciones pertenecientes a la red del Sistema Geocéntrico para las Américas (SIRGAS). Las denominadas soluciones semilibres generaron, semana a semana, una red GPS con una alta exactitud interna definida por medio de los vectores entre las estaciones y las coordenadas finales de la constelación satelital. La evaluación semanal dada por la repetibilidad de las soluciones brindó en promedio errores de 1,7 mm, 1,4 mm y 5,1 mm en las componentes [n e u], confirmando una alta consistencia en estas soluciones. Aunque las soluciones semilibres poseen una alta exactitud interna, las mismas no son utilizables para fines de análisis cinemático, pues carecen de un marco de referencia. En Latinoamérica, la densificación del Marco Internacional Terrestre de Referencia (ITRF), está representado por la red de estaciones de operación continua GNSS de SIRGAS, denominada como SIRGAS-CON. Por medio de las denominadas coordenadas semanales finales de las 8 estaciones consideradas como vínculo, se refirió cada una de las 119 soluciones al marco SIRGAS. La introducción del marco de referencia SIRGAS a las soluciones semilibres produce deformaciones en estas soluciones. Las deformaciones de las soluciones semilibres son producto de las cinemática de cada una de las placas en las que se ubican las estaciones de vínculo. Luego de efectuado el amarre semanal a las coordenadas SIRGAS, se hizo una estimación de los vectores de velocidad de cada una de las estaciones, incluyendo las de amarre, cuyos valores de velocidad se conocen con una alta exactitud. Para la determinación de las velocidades de las estaciones costarricenses, se programó una rutina en ambiente MatLab, basada en una ajuste por mínimos cuadrados. Los valores obtenidos en el marco de este proyecto en comparación con los valores oficiales, brindaron diferencias promedio del orden de los 0,06 cm/a, -0,08 cm/a y -0,10 cm/a respectivamente para las coordenadas [X Y Z]. De esta manera se logró determinar las coordenadas geocéntricas [X Y Z]T y sus variaciones temporales [vX vY vZ]T para el conjunto de 22 estaciones GPS de Costa Rica, dentro del datum IGS05, época de referencia 2010,5. Aunque se logró una alta exactitud en los vectores de coordenadas geocéntricas de las 22 estaciones, para algunas de las estaciones el cálculo de las velocidades no fue representativo debido al relativo corto tiempo (menos de un año) de archivos de datos. Bajo esta premisa, se excluyeron las ocho estaciones ubicadas al sur de país. Esto implicó hacer una estimación del campo local de velocidades con solamente veinte estaciones nacionales más tres estaciones en Panamá y una en Nicaragua. El algoritmo usado fue el denominado Colocación por Mínimos Cuadrados, el cual permite la estimación o interpolación de datos a partir de datos efectivamente conocidos, el cual fue programado mediante una rutina en ambiente MatLab. El campo resultante se estimó con una resolución de 30' X 30' y es altamente constante, con una velocidad resultante promedio de 2,58 cm/a en una dirección de 40,8° en dirección noreste. Este campo fue validado con base en los datos del modelo VEMOS2009, recomendado por SIRGAS. Las diferencias de velocidad promedio para las estaciones usadas como insumo para el cálculo del campo fueron del orden los +0,63 cm/a y +0,22 cm/a para los valores de velocidad en latitud y longitud, lo que supone una buena determinación de los valores de velocidad y de la estimación de la función de covarianza empírica, necesaria para la aplicación del método de colocación. Además, la grilla usada como base para la interpolación brindó diferencias del orden de -0,62 cm/a y -0,12 cm/a para latitud y longitud. Adicionalmente los resultados de este trabajo fueron usados como insumo para hacer una aproximación en la definición del límite del llamado Bloque de Panamá dentro del territorio nacional de Costa Rica. El cálculo de las componentes del Polo de Euler por medio de una rutina programa en ambiente MatLab y aplicado a diferentes combinaciones de puntos no brindó mayores aportes a la definición física de este límite. La estrategia lo que confirmó fue simplemente la diferencia en la dirección de todos los vectores velocidad y no permitió reveló revelar con mayor detalle una ubicación de esta zona dentro del territorio nacional de Costa Rica. ABSTRACT The first strict processing performed with the Bernese scientific software and contemplating the highest standards internationally recommended calculation, yielded a precise field of high accuracy, based on the integration and standardization of data from a GPS network located in Costa Rica. This processing watched a total of 119 weeks of daily data, is about 2.3 years from January 2009 to April 2011, for a total of 30 GPS stations, of which 22 are located in the country of Costa Rica and 8 hospitalizations within the network of Geocentric System for the Americas (SIRGAS). The semi-free solutions generated, every week a GPS network with high internal accuracy defined by vectors between stations and the final coordinates of the satellite constellation. The weekly evaluation given by repeatability of the solutions provided in average errors of 1.7 mm 1.4 mm and 5.1 mm in the components [n e u], confirming a high consistency in these solutions. Although semi-free solutions have a high internal accuracy, they are not used for purposes of kinematic analysis, because they lack a reference frame. In Latin America, the densification of the International Terrestrial Reference Frame (ITRF), is represented by a network of continuously operating GNSS stations SIRGAS, known as SIRGAS-CON. Through weekly final coordinates of the 8 stations considered as a link, described each of the solutions to the frame 119 SIRGAS. The introduction of the frame SIRGAS to semi-free solutions generates deformations. The deformations of the semi-free solutions are products of the kinematics of each of the plates in which link stations are located. After SIRGAS weekly link to SIRGAS frame, an estimate of the velocity vectors of each of the stations was done. The velocity vectors for each SIRGAS stations are known with high accuracy. For this calculation routine in MatLab environment, based on a least squares fit was scheduled. The values obtained compared to the official values, gave average differences of the order of 0.06 cm/yr, -0.08 cm/yr and -0.10 cm/yr respectively for the coordinates [XYZ]. Thus was possible to determine the geocentric coordinates [XYZ]T and its temporal variations [vX vY vZ]T for the set of 22 GPS stations of Costa Rica, within IGS05 datum, reference epoch 2010.5. The high accuracy vector for geocentric coordinates was obtained, however for some stations the velocity vectors was not representative because of the relatively short time (less than one year) of data files. Under this premise, the eight stations located in the south of the country were excluded. This involved an estimate of the local velocity field with only twenty national stations plus three stations in Panama and Nicaragua. The algorithm used was Least Squares Collocation, which allows the estimation and interpolation of data from known data effectively. The algorithm was programmed with MatLab. The resulting field was estimated with a resolution of 30' X 30' and is highly consistent with a resulting average speed of 2.58 cm/y in a direction of 40.8° to the northeast. This field was validated based on the model data VEMOS2009 recommended by SIRGAS. The differences in average velocity for the stations used as input for the calculation of the field were of the order of +0.63 cm/yr, +0.22 cm/yr for the velocity values in latitude and longitude, which is a good determination velocity values and estimating the empirical covariance function necessary for implementing the method of application. Furthermore, the grid used as the basis for interpolation provided differences of about -0.62 cm/yr, -0.12 cm/yr to latitude and longitude. Additionally, the results of this investigation were used as input to an approach in defining the boundary of Panama called block within the country of Costa Rica. The calculation of the components of the Euler pole through a routine program in MatLab and applied to different combinations of points gave no further contributions to the physical definition of this limit. The strategy was simply confirming the difference in the direction of all the velocity vectors and not allowed to reveal more detail revealed a location of this area within the country of Costa Rica.

Learning Bayesian networks from data by the incremental compilation of new network polynomials

Probabilistic graphical models are a huge research field in artificial intelligence nowadays. The scope of this work is the study of directed graphical models for the representation of discrete distributions. Two of the main research topics related to this area focus on performing inference over graphical models and on learning graphical models from data. Traditionally, the inference process and the learning process have been treated separately, but given that the learned models structure marks the inference complexity, this kind of strategies will sometimes produce very inefficient models. With the purpose of learning thinner models, in this master thesis we propose a new model for the representation of network polynomials, which we call polynomial trees. Polynomial trees are a complementary representation for Bayesian networks that allows an efficient evaluation of the inference complexity and provides a framework for exact inference. We also propose a set of methods for the incremental compilation of polynomial trees and an algorithm for learning polynomial trees from data using a greedy score+search method that includes the inference complexity as a penalization in the scoring function.

Multi-dimensional classification of GABAergic interneurons with Bayesian network-modeled label uncertainty

Abstract Interneuron classification is an important and long-debated topic in neuroscience. A recent study provided a data set of digitally reconstructed interneurons classified by 42 leading neuroscientists according to a pragmatic classification scheme composed of five categorical variables, namely, of the interneuron type and four features of axonal morphology. From this data set we now learned a model which can classify interneurons, on the basis of their axonal morphometric parameters, into these five descriptive variables simultaneously. Because of differences in opinion among the neuroscientists, especially regarding neuronal type, for many interneurons we lacked a unique, agreed-upon classification, which we could use to guide model learning. Instead, we guided model learning with a probability distribution over the neuronal type and the axonal features, obtained, for each interneuron, from the neuroscientists’ classification choices. We conveniently encoded such probability distributions with Bayesian networks, calling them label Bayesian networks (LBNs), and developed a method to predict them. This method predicts an LBN by forming a probabilistic consensus among the LBNs of the interneurons most similar to the one being classified. We used 18 axonal morphometric parameters as predictor variables, 13 of which we introduce in this paper as quantitative counterparts to the categorical axonal features. We were able to accurately predict interneuronal LBNs. Furthermore, when extracting crisp (i.e., non-probabilistic) predictions from the predicted LBNs, our method outperformed related work on interneuron classification. Our results indicate that our method is adequate for multi-dimensional classification of interneurons with probabilistic labels. Moreover, the introduced morphometric parameters are good predictors of interneuron type and the four features of axonal morphology and thus may serve as objective counterparts to the subjective, categorical axonal features.

Multi-dimensional classification of GABAergic interneurons with Bayesian network-modeled label uncertainty

Interneuron classification is an important and long-debated topic in neuroscience. A recent study provided a data set of digitally reconstructed interneurons classified by 42 leading neuroscientists according to a pragmatic classification scheme composed of five categorical variables, namely, of the interneuron type and four features of axonal morphology. From this data set we now learned a model which can classify interneurons, on the basis of their axonal morphometric parameters, into these five descriptive variables simultaneously. Because of differences in opinion among the neuroscientists, especially regarding neuronal type, for many interneurons we lacked a unique, agreed-upon classification, which we could use to guide model learning. Instead, we guided model learning with a probability distribution over the neuronal type and the axonal features, obtained, for each interneuron, from the neuroscientists’ classification choices. We conveniently encoded such probability distributions with Bayesian networks, calling them label Bayesian networks (LBNs), and developed a method to predict them. This method predicts an LBN by forming a probabilistic consensus among the LBNs of the interneurons most similar to the one being classified. We used 18 axonal morphometric parameters as predictor variables, 13 of which we introduce in this paper as quantitative counterparts to the categorical axonal features. We were able to accurately predict interneuronal LBNs. Furthermore, when extracting crisp (i.e., non-probabilistic) predictions from the predicted LBNs, our method outperformed related work on interneuron classification. Our results indicate that our method is adequate for multi-dimensional classification of interneurons with probabilistic labels. Moreover, the introduced morphometric parameters are good predictors of interneuron type and the four features of axonal morphology and thus may serve as objective counterparts to the subjective, categorical axonal features.

Searching for a consensus similarity function for generalized trapezoidal fuzzy numbers

There is controversy regarding the use of the similarity functions proposed in the literature to compare generalized trapezoidal fuzzy numbers since conflicting similarity values are sometimes output for the same pair of fuzzy numbers. In this paper we propose a similarity function aimed at establishing a consensus. It accounts for the different approaches of all the similarity functions. It also has better properties and can easily incorporate new parameters for future improvements. The analysis is carried out on the basis of a large and representative set of pairs of trapezoidal fuzzy numbers.

Predicting EQ-5D from the Parkinson's disease questionnaire PDQ-8 using multi-dimensional Bayesian network classifiers

The impact of the Parkinson's disease and its treatment on the patients' health-related quality of life can be estimated either by means of generic measures such as the european quality of Life-5 Dimensions (EQ-5D) or specific measures such as the 8-item Parkinson's disease questionnaire (PDQ-8). In clinical studies, PDQ-8 could be used in detriment of EQ-5D due to the lack of resources, time or clinical interest in generic measures. Nevertheless, PDQ-8 cannot be applied in cost-effectiveness analyses which require generic measures and quantitative utility scores, such as EQ-5D. To deal with this problem, a commonly used solution is the prediction of EQ-5D from PDQ-8. In this paper, we propose a new probabilistic method to predict EQ-5D from PDQ-8 using multi-dimensional Bayesian network classifiers. Our approach is evaluated using five-fold cross-validation experiments carried out on a Parkinson's data set containing 488 patients, and is compared with two additional Bayesian network-based approaches, two commonly used mapping methods namely, ordinary least squares and censored least absolute deviations, and a deterministic model. Experimental results are promising in terms of predictive performance as well as the identification of dependence relationships among EQ-5D and PDQ-8 items that the mapping approaches are unable to detect

Micromechanical basis for shear strength of rock discontinuities

The theoretical basis for evaluating shear strength in rock joints is presented and used to derive an equation that governs the relationship between tangential and normal stress on the joint during situations of slippage between the joint faces. The dependent variables include geometric dilatancy, the instantaneous friction angle, and a parameter that considers joint surface roughness. The effect roughness is studied, and the aforementioned formula is used to analyse joints under different conditions. A mathematical expression is deduced that explains Barton's value for the joint roughness coefficient (JRC) according to the roughness geometry. In particular, when the Hoek and Brown failure criterion is used for a rock in the contact with the surface roughness plane, it is possible to determine the shear strength of the joint as a function of the relationship between the uniaxial compressive strength of the wall with the normal stress acting on the wall. Finally, theoretical results obtained for the geometry of a three-dimensional joint are compared with those of the Barton's formulation

Decision boundary for discrete Bayesian network classifiers

Bayesian network classifiers are a powerful machine learning tool. In order to evaluate the expressive power of these models, we compute families of polynomials that sign-represent decision functions induced by Bayesian network classifiers. We prove that those families are linear combinations of products of Lagrange basis polynomials. In absence of V -structures in the predictor sub-graph, we are also able to prove that this family of polynomials does indeed characterize the specific classifier considered. We then use this representation to bound the number of decision functions representable by Bayesian network classifiers with a given structure.