Learning Bayesian Networks with Thousands of Variables

We present a method for learning Bayesian networks from data sets containing thousands of variables without the need for structure constraints. Our approach is made of two parts. The first is a novel algorithm that effectively explores the space of possible parent sets of a node. It guides the exploration towards the most promising parent sets on the basis of an approximated score function that is computed in constant time. The second part is an improvement of an existing ordering-based algorithm for structure optimization. The new algorithm provably achieves a higher score compared to its original formulation. Our novel approach consistently outperforms the state of the art on very large data sets.

Learning Bounded Tree-Width Bayesian Networks via Sampling

Learning Bayesian networks with bounded tree-width has attracted much attention recently, because low tree-width allows exact inference to be performed efficiently. Some existing methods [12, 14] tackle the problem by using k-trees to learn the optimal Bayesian network with tree-width up to k. In this paper, we propose a sampling method to efficiently find representative k-trees by introducing an Informative score function to characterize the quality of a k-tree. The proposed algorithm can efficiently learn a Bayesian network with tree-width at most k. Experiment results indicate that our approach is comparable with exact methods, but is much more computationally efficient.

Predicting impact damage, residual strength and crashworthiness of composite structures

The development of the latest generation of wide-body carbon-fibre composite passenger aircraft has heralded a new era in the utilisation of these materials. The premise of superior specific strength and stiffness, corrosion and fatigue resistance, is tempered by high development costs, slow production rates and lengthy and expensive certification programmes. Substantial effort is currently being directed towards the development of new modelling and simulation tools, at all levels of the development cycle, to mitigate these shortcomings. One of the primary challenges is to reduce the extent of physical testing, in the certification process, by adopting a ‘certification by simulation’ approach. In essence, this aspirational objective requires the ability to reliably predict the evolution and progression of damage in composites. The aerospace industry has been at the forefront of developing advanced composites modelling tools. As the automotive industry transitions towards the increased use of composites in mass-produced vehicles, similar challenges in the modelling of composites will need to be addressed, particularly in the reliable prediction of crashworthiness. While thermoset composites have dominated the aerospace industry, thermoplastics composites are likely to emerge as the preferred solution for meeting the high-volume production demands of passenger road vehicles. This keynote presentation will outline recent progress and current challenges in the development of finite-element-based predictive modelling tools for capturing impact damage, residual strength and energy absorption capacity of thermoset and thermoplastic composites for crashworthiness assessments.

Learning Bayesian Networks with Bounded Tree-Width via Guided Search

Bounding the tree-width of a Bayesian network can reduce the chance of overfitting, and allows exact inference to be performed efficiently. Several existing algorithms tackle the problem of learning bounded tree-width Bayesian networks by learning from k-trees as super-structures, but they do not scale to large domains and/or large tree-width. We propose a guided search algorithm to find k-trees with maximum Informative scores, which is a measure of quality for the k-tree in yielding good Bayesian networks. The algorithm achieves close to optimal performance compared to exact solutions in small domains, and can discover better networks than existing approximate methods can in large domains. It also provides an optimal elimination order of variables that guarantees small complexity for later runs of exact inference. Comparisons with well-known approaches in terms of learning and inference accuracy illustrate its capabilities.

NMR in the characterization of heavy residual procedural streams

The main objective of this work was to monitor a set of physical-chemical properties of heavy oil procedural streams through nuclear magnetic resonance spectroscopy, in order to propose an analysis procedure and online data processing for process control. Different statistical methods which allow to relate the results obtained by nuclear magnetic resonance spectroscopy with the results obtained by the conventional standard methods during the characterization of the different streams, have been implemented in order to develop models for predicting these same properties. The real-time knowledge of these physical-chemical properties of petroleum fractions is very important for enhancing refinery operations, ensuring technically, economically and environmentally proper refinery operations. The first part of this work involved the determination of many physical-chemical properties, at Matosinhos refinery, by following some standard methods important to evaluate and characterize light vacuum gas oil, heavy vacuum gas oil and fuel oil fractions. Kinematic viscosity, density, sulfur content, flash point, carbon residue, P-value and atmospheric and vacuum distillations were the properties analysed. Besides the analysis by using the standard methods, the same samples were analysed by nuclear magnetic resonance spectroscopy. The second part of this work was related to the application of multivariate statistical methods, which correlate the physical-chemical properties with the quantitative information acquired by nuclear magnetic resonance spectroscopy. Several methods were applied, including principal component analysis, principal component regression, partial least squares and artificial neural networks. Principal component analysis was used to reduce the number of predictive variables and to transform them into new variables, the principal components. These principal components were used as inputs of the principal component regression and artificial neural networks models. For the partial least squares model, the original data was used as input. Taking into account the performance of the develop models, by analysing selected statistical performance indexes, it was possible to conclude that principal component regression lead to worse performances. When applying the partial least squares and artificial neural networks models better results were achieved. However, it was with the artificial neural networks model that better predictions were obtained for almost of the properties analysed. With reference to the results obtained, it was possible to conclude that nuclear magnetic resonance spectroscopy combined with multivariate statistical methods can be used to predict physical-chemical properties of petroleum fractions. It has been shown that this technique can be considered a potential alternative to the conventional standard methods having obtained very promising results.

Avaliaçao do potencial energético da biomassa florestal residual na ZIF de Arade-Alte/ São Bartolomeu de Messines

Dissertação de mest., Energias Renováveis e Gestão de Energia, Faculdade de Ciências e Tecnologia, Univ. do Algarve, 2011

Environmental risk factors for the initial acquisition of Pseudomonas aeruginosa in young children with cystic fibrosis

Thesis (Ph.D.)--University of Washington, 2013

New methods for meta-analysis under a fixed effects framework, with frequentist and Bayesian estimation.

Thesis (Ph.D.)--University of Washington, 2015