AN ABSTRACT OF THE THESIS OF SOROOSH SOHANGIR, for the MASTER OF SCIENCE degree in COMPUTER SCIENCE, presented on 9 th November 2011, at Southern Illinois University Carbondale. TITLE: OPTIMIZED FEATURE SELECTION USING NEUROEVOLUTION OF AUGMENTING TOPOLOGIES (NEAT) MAJOR PROFESSOR: Dr. Shahram Rahimi Feature selection using the NeuroEvolution of Augmenting Topologies (NEAT) is a new approach. In this thesis an investigation had been carried out for implementation based on optimization of the network topology and protecting innovation through the speciation which is similar to what happens in nature. The NEAT is implemented through the JNEAT package and Utans method for feature selection is deployed. The performance of this novel method is compared with feature selection using Multilayer Perceptron (MLP) where Belue, Tekto, and Utans feature selection methods is adopted. According to unveiled data from this thesis the number of species, the training, accuracy and number of hidden neurons are notably improved as compared with conventional networks. For instance the time is reduced by factor of three.

Submitted by Emanuel Varela Cardoso (emanuel.varela@ufcg.edu.br) on 2018-11-09T17:51:39Z No. of bitstreams: 1 ALECIO SOARES SILVA – DISSERTAÇÃO (PPGMat) 2014.pdf: 873139 bytes, checksum: 9a35db2563d66eb36f4dabfe6e5cd45e (MD5) / Made available in DSpace on 2018-11-09T17:51:39Z (GMT). No. of bitstreams: 1 ALECIO SOARES SILVA – DISSERTAÇÃO (PPGMat) 2014.pdf: 873139 bytes, checksum: 9a35db2563d66eb36f4dabfe6e5cd45e (MD5) Previous issue date: 2014-08 / Capes / Neste trabalho consideramos o uso de algoritmo de Euclides com o intuito de aplicá-lo de uma forma interdisciplinar. Para atingir este objetivo construimos o conjunto dos números naturais, com base nos quatro axiomas de Peano e o conjunto dos inteiros por uma relação de equivalência específica. Além disto, fizemos um estudo de algumas propriedades aritméticas dos números inteiros, bem como do magnífico algoritmo de Euclides. Em seguida utilizamos este algoritmo como uma ferramenta para calcular o maximo divisor comum (MDC) de números inteiros e a partir do MDC estudamos a resolução de equações lineares diofantinas, as quais foram empregadas para fazer o balanceamento de Reações Quimicas. / In this work we consider the use of the Euclid’s algorithm in order to apply it in an interdisciplinary way. To achieve this we constructed the set of the natural numbers based on the four Peano axioms and the set of integers by a specific equivalence relation. Moreover, we have studied some arithmetic properties of integers, as well as the magnificent Euclidean algorithm. We then use this algorithm as a tool to calculate the Greatest Common Divisor (GCD) of integers and from this study the resolution of Diophantine linear equations, which were employed to do the balance of Chemical Reactions.

O Atracamento Molecular é uma importante ferramenta utilizada no descobrimento de novos fármacos. O atracamento com ligante flexível é um processo computacionalmente custoso devido ao número alto de graus de liberdade do ligante e da rugosidade do espaço de busca conformacional representando a afinidade entre o receptor e uma molécula ligante. O problema é definido como a busca pela solução de menor energia de ligação proteína-ligante. Considerando uma função suficientemente acurada, a solução ótima coincide com a melhor orientação e afinidade entre as moléculas. Assim, o método de busca e a função de energia são partes fundamentais para a resolução do problema. Muitos desafios são enfrentados para a resolução do problema, o tratamento da flexibilidade, algoritmo de amostragem, a exploração do espaço de busca, o cálculo da energia livre entre os átomos, são alguns dos focos estudados. Esta dissertação apresenta uma técnica baseada em um Algoritmo Genético de Chaves Aleatórias Viciadas, incluindo a discretização do espaço de busca e métodos de agrupamento para a multimodalidade do problema de atracamento molecular. A metodologia desenvolvida explora o espaço de busca gerando soluções diversificadas. O método proposto foi testado em uma seleção de complexos proteína-ligante e foi comparado com softwares existentes: AutodockVina e Dockthor. Os resultados foram estatisticamente analisados em termos estruturais. O método se mostrou eficiente quando comparado com outras ferramentas e uma alternativa para o problema de Atracamento Molecular. / Molecular Docking is a valuable tool for drug discovery. Receptor and flexible Ligand docking is a very computationally expensive process due to a large number of degrees of freedom of the ligand and the roughness of the molecular binding search space. A Molecular Docking simulation starts with a receptor and ligand unbounded structures and the algorithm tests hundreds of thousands of ligands conformations and orientations to find the best receptor-ligand binding affinity by assigning and optimizing an energy function. Despite the advances in the conception of methods and computational strategies for search the best protein-ligand binding affinity, the development of new strategies, the adaptation, and investigation of new approaches and the combination of existing and state-of-the-art computational methods and techniques to the Molecular Docking problem are clearly needed. We developed a Biased Random-Key Genetic Algorithm as a sampling strategy to search the protein-ligand conformational space. The proposed method has been tested on a selection of protein-ligand complexes and compared with existing tools AutodockVina and Dockthor. Compared with other traditional docking software, the proposed method has the best average Root-Mean-Square Deviation. Structural results were statistically analyzed. The proposed method proved to be efficient and a good alternative to the molecular docking problem.

[pt] Neste trabalho são analisados vários algoritmos de criptografia, entre eles, o DES (Data Encryption Standard), o algoritmo de chaves públicas RSA e o algoritmo de distribuição pública de chaves DH de Diffie- Hellman. São propostos então, três protocolos de segurança, para comunicação interativa e unidirecional em redes de comutação de pacotes. O primeiro protocolo foi proposto supondo-se que o algoritmo de criptografia utilizado pela rede, é o DH em conjunto com o DES. O segundo considera o uso somente do algoritmo DES para cifrar e decifrar a mensagem. O terceiro protocolo considera o uso do algoritmo RSA. As funções que poderão ser executadas em cada protocolo são: comunicação com privacidade, comunicação com autenticidade e comunicação assinada. Para o protocolo DH/DES propõe-se um modelo de unidade criptográfica para implementação das funções que o protocolo irá realizar. / [en] This work analyses several algoritms of cryptography; among them, the DES (Data Encryption Standard), the public key algorithm RSA and the public key distribution algorithm DH of Diffie-Hellman. Three security protocols are then proposed for interactive and one-way communication in packet switched networks. The sirst protocol was proposed with the assumption that the algorthm of cryptography, used by the network is the DH, in conection with the DES to encipher the message. The third protocol considers the use of the RSA. The functions that shall be executed in each protocol are: communication with privacy, communication with authenticity, and signed communication. For the protocol DH/DES a model of cryptographical unit is proposed for the implementation of the functions wich will be executed by the protocol.

[pt] A partir da forma geral do algoritmo de adaptação, é proposto o uso de mais uma variável de projeto, denominada janela. Esta janela tem por objetivo melhorar as características de convergência de algoritmos, cujos parâmetros são partes de estruturas IIR. A introdução das janelas é justificada heuristicamente e seu desempenho é avaliado por meio de diversas simulações de identificação de sistemas. Os resultados obtidos indicam aumentos significativos na velocidade de convergência (cerca de uma ordem de grandeza mais rápido do que os algoritmos atualmente mais usados), na precisão das estimativas dos parâmetros do problema e na robustez dos novos algoritmos (menor número de pólos instáveis durante a adaptação). Estes resultados foram observados em ambientes estacionários e não estacionários, com e sem ruído de medida e com ordem de identificação suficiente ou não. / [en] From the adaptation algorithm general form it is proposed the usage of another design variable, called Window. The goal of this Window is to improve the convergence characteristics of algorithms whose parameters are parts of IIR Structures. The introduction of the Window is heuristically justified and its performance is eventuated by several system identification simulations. The results achieved suggest significant increase in the convergence speed (about one order of magnitude faster than the currently most used algorithms), in the parameter estimation precision and in the new algorithm robusteness (fewer unstable poles during adaptation). These results were observed in sationary and non-stationary environments, with and without measurement noise and with sufficient identification order or not.

Doctor of Philosophy / Department of Mathematics / Nathan Albin / The concept of discrete $p$-modulus provides a general framework for understanding arbitrary families of objects on a graph. The $p$-modulus provides a sense of ``structure'' of the underlying graph, with different families of objects leading to different insight into the graph's structure. This dissertation builds on this idea, with an emphasis on the family of spanning trees and the underlying graph structure that spanning tree modulus exposes. This dissertation provides a review of the probabilistic interpretation of modulus. In the context of spanning trees, this interpretation rephrases modulus as the problem of choosing a probability mass function on the spanning trees so that two independent, identically distributed random spanning trees have expected overlap as small as possible. A theoretical lower bound on the expected overlap is shown. Graphs that attain this lower bound are called homogeneous and have the property that there exists a probability mass function that gives every edge equal likelihood to appear in a random tree. Moreover, any nonhomogeneous graph necessarily has a homogeneous subgraph (called a homogeneous core), which is shown to split the modulus problem into two smaller subproblems through a process called deflation. Spanning tree modulus and the process of deflation establish a type of hierarchical structure in the underlying graph that is similar to the concept of core-periphery structure found in the literature. Using this, one can see an alternative way of decomposing a graph into its hierarchical community components using homogeneous cores and a related concept: minimum feasible partitions. This dissertation also introduces a simple greedy algorithm for computing the spanning tree modulus that utilizes any efficient algorithm for finding minimum spanning trees. A theoretical proof of the convergence rate is provided, along with computational examples.

In modern applications, it is a big challenge that analyzing the order statistics about the most recent parts of the high-volume and high velocity stream data. There are some online quantile algorithms that can keep the sketch of the data in the sliding window and they can answer the quantile or rank query in a very short time. But most of them take the GK algorithm as the subroutine, which is not known to be mergeable. In this paper, we propose another algorithm to keep the sketch that maintains the order statistics over sliding windows. For the fixed-size window, the existing algorithms can’t maintain the correctness in the process of updating the sliding window. Our algorithm not only can maintain the correctness but also can achieve similar performance of the optimal algorithm. Under the basis of maintaining the correctness, the insert time and query time are close to the best results, while others can't maintain the correctness. In addition to the fixed-size window algorithm, we also provide the time-based window algorithm that the window size varies over time. Last but not least, we provide the window aggregation algorithm which can help extend our algorithm into the distributed system.

This thesis concerns the construction of a detailed, compartmental population balance model for twin-screw granulation using the stochastic weighted particle method. A number of new particle mechanisms are introduced and existing mechanisms augmented including immersion nucleation, coagulation, breakage, consolidation, liquid penetration, primary particle layering and transport. The model’s predictive power is assessed over a range of liquid-solid mass feed ratios using existing experimental data and is demonstrated to qualitatively capture key experimental trends in the physical characteristic of the granular product. As part of the model development process, a number of numerical techniques for the stochastic weighed method are constructed in order to efficiently solve the population balance model. This includes a new stochastic implementation of the immersion nucleation mechanism and a variable weighted inception algorithm that dramatically reduces the number of computational particles (and hence computational power) required to solve the model. Optimum operating values for free numerical parameters and the general convergence properties of the complete simulation algorithm are investigated in depth. The model is further refined though the use of distinct primary particle and aggregate population balances, which are coupled to simulate the complete granular system. The nature of this coupling permits the inclusion of otherwise computational prohibitive mechanisms, such as primary particle layering, into the process description. A new methodology for assigning representative residence times to simulation compartments, based on screw geometry, is presented. This residence time methodology is used in conjunction with the coupled population balance framework to model twin-screw systems with a number of different screw configurations. The refined model is shown to capture key trends attributed to screw element geometry, in particular, the ability of kneading elements to distribute liquid across the granular mass.

Today’s aerospace industry continues to exploit liquid hydrocarbon fossil fuels. Motivated by operational considerations, continued availability and cost, this is likely to be the case for many years, despite the obvious environmental concerns. The interplay of liquid atomisation, spray vaporisation and the combustion process are intricately linked. However, the physical process of fuel injection and its atomisation into tiny droplets prior to combustion remains poorly understood. Because atomisation governs the size of the fuel droplets, and therefore their subsequent evaporation rate, adjusting the injection sequence is of paramount importance and will have far-reaching repercussions on many aspects of the combustion process, for example pollutant formation. In the context of jet engines, kerosene is usually injected in its liquid form via an airblast-type fuel injector. A coflowing high-speed airstream destabilises the liquid fuel, which is thus sprayed into fine droplets into the combustion chamber. The prediction of this phenomenon for various operating conditions relevant to the aeronautical industry requires a deeper understanding of the mechanisms involved in the interaction of the two fluids. A key element in predicting the complex behaviour of spray formation and evolution in jet engines is accurate modelling of fuel atomisation. Atomisation represents one of the key challenges that remains to be undertaken to make predictive computational simulations possible. However, the inherent multi-physics and multi-scale nature of this process limits numerical investigations. Thanks to the steady progress in computer power and Computational Fluid Dynamics (CFD) methods, computational modelling of injection systems emerges as a promising tool that can drive the design of future devices. This research project sets out to investigate the atomisation process in detail, in particular in providing physical insight into the fundamental physics of the phenomenon, in conjunction with an analysis on wetting behaviours and liquid droplet tracking. High-fidelity numerical simulations are performed using a novel in-house state-of-the-art multiphase flow modelling capability, RCLSFoam. The performance of the numerical scheme is demonstrated on typical two-dimensional and three-dimensional benchmark test cases relevant to both multiphase flow modelling and atomisation, and validated against other computational methods. An informed and systematic qualitative assessment of the topological variations of the phase interface during primary atomisation of a liquid film is made through dynamical analysis, while investigating an extensive domain of operating conditions at ambient and aero-engine injection conditions relevant to industry. This analysis demonstrated the influence of shear-driven instabilities on the atomisation process. The shear stress and difference in inertia between liquid and gas are observed to play a significant role in the atomisation process. In addition, the key physical mechanisms and their competing effects have been mapped out in order to predict the evolution of the process according to the operating conditions of the injection system. The proposed cartography gathers four different atomisation mechanisms. In particular, for sufficiently high liquid injection speeds, three-dimensional wave modes were observed to co-exist (the “3-D wave mode” regime). For very low liquid flow rates, accumulated liquid at the atomising edge undergoes deformation by which droplets are generated (the “accumulation” regime). For an increasing gas injection speed and a fixed liquid velocity, the effects of surface tension were observed to result in the generation of streamwise ligaments only, which tend to pair up (the “ligament-merging” regime). Finally, “vortex action” is another observed mechanism by which the liquid film is fragmented. Overall, this research project culminated in (i) the study of dynamic wetting behaviours, with the implementation and validation against experimental data of the Kistler dynamic contact model; and (ii) the demonstration of an algorithm for droplet capture and subsequent post-processing analysis of the droplet characteristics.

Made available in DSpace on 2014-06-11T19:35:40Z (GMT). No. of bitstreams: 0 Previous issue date: 2008-12-19Bitstream added on 2014-06-13T20:46:42Z : No. of bitstreams: 1 pinho_af_dr_guara.pdf: 1635741 bytes, checksum: d8dc3d0b8a67932941a332b122ed1672 (MD5) / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) / Métodos de otimização combinados com a simulação computacional a eventos discretos têm sido utilizados em diversas aplicações na manufatura. Entretanto, estes métodos possuem baixo desempenho, em relação ao tempo computacional, ao manipularem mais de uma variável de decisão. Desta forma, o objetivo desta tese de doutorado é propor um método para otimização de modelos de simulação a eventos discretos com maior eficiência em relação ao tempo de processamento quando comparado a uma ferramenta comercial conhecida. Cabe ressaltar que a qualidade da variável de resposta não será alterada, ou seja, o método proposto manterá a eficácia das soluções encontradas. Será mostrado que a melhora neste desempenho é obtida através de uma melhor percepção do comportamento dos parâmetros existentes nos algoritmos genéticos, em especial o parâmetro tamanho da população. A comparação entre o método desenvolvido com a ferramenta de otimização existente no mercado se dará através de uma metodologia já consolidada disponível na literatura. As conclusões serão apresentadas comprovando a eficácia do método proposto. / Optimization methods combined with discrete events simulation have been used in many manufacturing applications. However, these methods have poor performance considering the computational time, when manipulating more than one decision variable. In this way, the aim of this thesis is to propose a method for optimizing discrete events simulation models with higher efficiency in relation to the processing time when compared to a known commercial tool. Besides, the optimization quality will not be altered, i. e., the proposed method will keep the effectiveness of the achieved solutions. It will be shown that the performance improvement is obtained by means of a better perception of the behavior of all parameters presented in the genetic algorithms, particularly the population size parameter. The comparison between the developed method and the optimization tool will be accomplished by means of a consolidated methodology available in the simulation literature. The conclusions will be presented proving the effectiveness of the developed method.