ALGORITMOS de Estimação de Distribuição (EDAs) compõem uma frente de pesquisa em Computação Evolutiva que tem apresentado resultados promissores para lidar com problemas complexos de larga escala. Nesse contexto, destaca-se o Algoritmo de Otimização Bayesiano (BOA) que usa um modelo probabilístico multivariado (representado por uma rede Bayesiana) para gerar novas soluções a cada iteração. Baseado no BOA e na investigação de algoritmos de detecção de estrutura de comunidades (para melhorar os modelos multivariados construídos), propõe-se dois novos algoritmos denominados CD-BOA e StrOp. Mostra-se que ambos apresentam vantagens significativas em relação ao BOA. O CD-BOA mostra-se mais flexível que o BOA, ao apresentar uma maior robustez a variações dos valores de parâmetros de entrada, facilitando o tratamento de uma maior diversidade de problemas do mundo real. Diferentemente do CD-BOA e BOA, o StrOp mostra que a detecção de comunidades a partir de uma rede Bayesiana pode modelar mais adequadamente problemas decomponíveis, reestruturando-os em subproblemas mais simples, que podem ser resolvidos por uma busca gulosa, resultando em uma solução para o problema original que pode ser ótima no caso de problemas perfeitamente decomponíveis, ou uma aproximação, caso contrário. Também é proposta uma nova técnica de reamostragens para EDAs (denominada REDA). Essa técnica possibilita a obtenção de modelos probabilísticos mais representativos, aumentando significativamente o desempenho do CD-BOA e StrOp. De uma forma geral, é demonstrado que, para os casos testados, CD-BOA e StrOp necessitam de um menor tempo de execução do que o BOA. Tal comprovação é feita tanto experimentalmente quanto por análise das complexidades dos algoritmos. As características principais desses algoritmos são avaliadas para a resolução de diferentes problemas, mapeando assim suas contribuições para a área de Computação Evolutiva / ESTIMATION of Distribution Algorithms represent a research area which is showing promising results, especially in dealing with complex large scale problems. In this context, the Bayesian Optimization Algorithm (BOA) uses a multivariate model (represented by a Bayesian network) to find new solutions at each iteration. Based on BOA and in the study of community detection algorithms (to improve the constructed multivariate models), two new algorithms are proposed, named CD-BOA and StrOp. This paper indicates that both algorithms have significant advantages when compared to BOA. The CD-BOA is shown to be more flexible, being more robust when using different input parameters, what makes it easier to deal with a greater diversity of real-world problems. Unlike CD-BOA and BOA, StrOp shows that the detection of communities on a Bayesian network more adequately models decomposable problems, resulting in simpler subproblems that can be solved by a greedy search, resulting in a solution to the original problem which may be optimal in the case of perfectly decomposable problems, or a fair approximation if not. Another proposal is a new resampling technique for EDAs (called REDA). This technique results in multivariate models that are more representative, significantly improving the performance of CD-BOA and StrOp. In general, it is shown that, for the scenarios tested, CD-BOA and StrOp require lower running time than BOA. This indication is done experimentally and by the analysis of the computational complexity of the algorithms. The main features of these algorithms are evaluated for solving various problems, thus identifying their contributions to the field of Evolutionary Computation

The HVAC system configuration is a conceptual design of the HVAC system, including the employed components, the topology of the airflow network, and the control strategy with set points. Selection of HVAC system configuration is normally done in the early stage of the design process. The configuration design, however, has significant impacts on the performance of the final system. This thesis describes the development of the design synthesis of optimal HVAC system configurations by Evolutionary Algorithm. In this research, the HVAC system configuration design synthesis has been formulated as an optimisation problem, in which, the component set of the configuration, the topology of the airflow network, and the control set points for the assumed supervisory control strategy, are the optimisation variables. Psychrometrics-based configuration model has been developed in order to evaluate the optimisation objective of minimising the annual energy consumption of the HVAC system. The optimisation is also subjected to a number of design constraints, including the connectivity of the topology, the performance limitations of the components, and the design requirements for the air-conditioned zones. The configuration synthesis problem is a multi-level optimisation problem. The topology depends on the set of selected components, whereas the search space of the control set points changes with the different components and topology. On the other hand, the performance of the configuration is assessed with its optimum operation; therefore the control set points have to be optimised for each configuration solution, before the optimum configuration can be identified. In this research, a simultaneous evolutionary approach has been developed. All optimisation variables of the configuration have been enwded into an integrated genotypic data structure. Evolutionary operators have also been developed to search the topological space (for the optimum topology) and parametric space (for the optimal control set points) at the same time. The performance of the developed approach has been validated with example optimisation problems. It is concluded that the implemented evolutionary algorithm has been able to find (near) optimum solutions for various design problems, though multiple trials may be required. The limitations of this approach and the direction of future development have been discussed.

The model of this thesis simulates a simple artificial eco-system in which evolving and learning agents try to survive by consuming balls of energy and surviving attacks by other agents. The study finds that the model indeed manages to evolve surviving, and in some cases very aggressive, agents. The thesis presents similar conclusions to that of the study of Polyworld by Yaeger [16]; that an evolving population only facilitates a need for complexity set by the world it evolves in and stagnates when the population has reached this level of complexity. If the populations are to evolve further, the world it lives in must first demand a higher level of complexity. Various problems with simulating artificial life are also discussed along with the more specific obstacles of simulating artificial life in Breve and NEST integrated. The physical world of the model is built in the Breve simulation environment and the neural networks are simulated in NEST through integrate-and-fire neurons and spike-timing dependent plasticity synapses.

Evolutionary algorithms (EAs) have been shown to be competent at solving complex, multimodal optimisation problems in applications where the search space is large and badly understood. EAs are therefore among the most promising classes of algorithms for solving the Protein Structure Prediction Problem (PSPP). The PSPP is how to derive the 3D-structure of a protein given only its sequence of amino acids. This dissertation defines, evaluates and shows limitations of simplified models for solving the PSPP. These simplified models are off-lattice extensions to the lattice HP model which has been proposed and is claimed to possess some of the properties of real protein folding such as the formation of a hydrophobic core. Lattice models usually model a protein at the amino acid level of detail, use simple energy calculations and are used mainly for search algorithm development. Off-lattice models usually model the protein at the atomic level of detail, use more complex energy calculations and may be used for comparison with real proteins. The idea is to combine the fast energy calculations of lattice models with the increased spatial possibilities of an off-lattice environment allowing for comparison with real protein structures. A hypothesis is presented which claims that a simplified off-lattice model which considers other amino acid properties apart from hydrophobicity will yield simulated structures with lower Root Mean Square Deviation (RMSD) to the native fold than a model only considering hydrophobicity. The hypothesis holds for four of five tested short proteins with a maximum of 46 residues. Best average RMSD for any model tested is above 6Å, i.e. too high for useful structure prediction and excludes significant resemblance between native and simulated structure. Hence, the tested models do not contain the necessary biological information to capture the complex interactions of real protein folding. It is also shown that the EA itself is competent and can produce near-native structures if given a suitable evaluation function. Hence, EAs are useful for eventually solving the PSPP.

The objective of this thesis was to investigate the possibility ofusing association rule algorithms to automatically generaterules for the output of a Parameter Tuning framework. Therules would be the basis for a recommendation to the user regardingwhich parameter space to reduce during experimentation.The parameter tuning output was generated by means ofan open source project (INPUT) example program. InPUT is atool used to describe computer experiment configurations in aframework independent input/output format. InPUT has adaptersfor the evolutionary algorithm framework Watchmakerand the tuning framework SPOT. The output was imported in Rand preprocessed to a format suitable for association rule algorithms.Experiments were conducted on data for which theparameter spaces were discretized in 2, 5, 10 steps. The minimumsupport threshold was set to 1% and 3% to investigatethe amount of rules over time. The Apriori and Eclat algorithmsproduced exactly the same amount of rules, and the top 5rules with regards to support were basically the same for bothalgorithms. It was not possible at the time to automatically distinguishinguseful rules. In combination with the many manualdecisions during the process of converting the tuning output toassociation rules, the conclusion was reached to not recommendassociation rules for enhancing the Parameter Tuningprocess.

Multi-objective optimization is a growing field of interest for both theoretical and applied research, mostly due to the higher accuracy with which multi-objective problems (MOPs) model real- world scenarios. While single-objective models simplify real-world problems, MOPs can contain several (and often conflicting) objective functions to be optimized at once. This increased accuracy, however, comes at the expense of a higher difficulty that MOPs pose for optimization algorithms in general, and so a significant research effort has been dedicated to the development of approximate and heuristic algorithms. In particular, a number of proposals concerning the adaptation of evolutionary algorithms (EAs) for multi-objective problems can be seen in the literature, evidencing the interest they have received from the research community.This large number of proposals, however, does not mean that the full search power offered by multi- objective EAs (MOEAs) has been properly exploited. For instance, in an attempt to propose significantly novel algorithms, many authors propose a number of algorithmic components at once, but evaluate their proposed algorithms as monolithic blocks. As a result, each time a novel algorithm is proposed, several questions that should be addressed are left unanswered, such as (i) the effectiveness of individual components, (ii) the benefits and drawbacks of their interactions, and (iii) whether a better algorithm could be devised if some of the selected/proposed components were replaced by alternative options available in the literature. This component-wise view of MOEAs becomes even more important when tackling a new application, since one cannot antecipate how they will perform on the target scenario, neither predict how their components may interact. In order to avoid the expensive experimental campaigns that this analysis would require, many practitioners choose algorithms that in the end present suboptimal performance on the application they intend to solve, wasting much of the potential MOEAs have to offer.In this thesis, we take several significant steps towards redefining the existng algorithmic engineering approach to MOEAs. The first step is the proposal of a flexible and representative algorithmic framework that assembles components originally used by many different MOEAs from the literature, providing a way of seeing algorithms as instantiations of a unified template. In addition, the components of this framework can be freely combined to devise novel algorithms, offering the possibility of tailoring MOEAs according to the given application. We empirically demonstrate the efficacy of this component-wise approach by designing effective MOEAs for different target applications, ranging from continuous to combinatorial optimization. In particular, we show that the MOEAs one can tailor from a collection of algorithmic components is able to outperform the algorithms from which those components were originally gathered. More importantly, the improved MOEAs we present have been designed without manual assistance by means of automatic algorithm design. This algorithm engineering approach considers algorithmic components of flexible frameworks as parameters of a tuning problem, and automatically selects the component combinations that lead to better performance on a given application. In fact, this thesis also represents significant advances in this research direction. Primarily, this is the first work in the literature to investigate this approach for problems with any number of objectives, as well as the first to apply it to MOEAs. Secondarily, our efforts have led to a significant number of improvements in the automatic design methodology applied to multi-objective scenarios, as we have refined several aspects of this methodology to be able to produce better quality algorithms.A second significant contribution of this thesis concerns understanding the effectiveness of MOEAs (and in particular of their components) on the application domains we consider. Concerning combina- torial optimization, we have conducted several investigations on the multi-objective permutation flowshop problem (MO-PFSP) with four variants differing as to the number and nature of their objectives. Through thorough experimental campaigns, we have shown that some components are only effective when jointly used. In addition, we have demonstrated that well-known algorithms could easily be improved by replacing some of their components by other existing proposals from the literature. Regarding continuous optimization, we have conducted a thorough and comprehensive performance assessment of MOEAs and their components, a concrete first step towards clearly defining the state-of-the-art for this field. In particular, this assessment also encompasses many-objective optimization problems (MaOPs), a sub-field within multi-objective optimization that has recently stirred the MOEA community given its theoretical and practical demands. In fact, our analysis is instrumental to better understand the application of MOEAs to MaOPs, as we have discussed a number of important insights for this field. Among the most relevant, we highlight the empirical verification of performance metric correlations, and also the interactions between structural problem characteristics and the difficulty increase incurred by the high number of objectives.The last significant contribution from this thesis concerns the previously mentioned automatically generated MOEAs. In an initial feasibility study, we have shown that MOEAs automatically generated from our framework are able to consistently outperform the original MOEAs from where its components were gathered both for the MO-PFSP and for MOPs/MaOPs. The major contribution from this subset, however, regards continuous optimization, as we significantly advance the state-of-the-art for this field. To accomplish this goal, we have extended our framework to encompass approaches that are primarily used for this continuous problems, although the conceptual modeling we use is general enough to be applied to any domain. From this extended framework we have then automatically designed state-of- the-art MOEAs for a wide range of experimental scenarios. Moreover, we have conducted an in-depth analysis to explain their effectiveness, correlating the role of algorithmic components with experimental factors such as the stopping criterion or the performance metric adopted.Finally, we highlight that the contributions of this thesis have been increasingly recognized by the scientific community. In particular, the contributions to the research of MOEAs applied to continuous optimization are remarkable given that this is the primary application domain for MOEAs, having been extensively studied for a couple decades now. As a result, chapters from this work have been accepted for publication in some of the best conferences and journals from our field. / Doctorat en Sciences de l'ingénieur et technologie / info:eu-repo/semantics/nonPublished

L'objectif de cette thèse est de concevoir un système automatique constitué d'une ou plusieurs caméras capables de détecter en trois dimensions un amalgame d'objets stockés dans un conteneur. Pour ceci, il est nécessaire de modéliser, de reconnaître et de localiser des formes dans une image. Dans un premier temps, Nous proposons une solution d'optimisation du calibrage de caméras. C'est une tâche essentielle pour récupérer des informations quantitatives sur les images capturées. Cette méthode nécessite des compétences spécifiques en matière de traitement d'image, ce qui n'est pas toujours le cas dans l'industrie. Nous proposons d'automatiser et d'optimiser le système d'étalonnage en éliminant la sélection des images par l'opérateur. Ensuite, nous proposons d'améliorer les systèmes de détection d'objets fins et sans motif. Enfin, nous proposons d'adapter des algorithmes évolutionnaires dans le but d'optimiser les temps de recherche. / The aim of this thesis is to design an automatic system involving one or several cameras capable of detecting in three dimensions a set of abjects placed in a bin. To do this, we must model, recognize and locate shapes in an image. First, we propose a solution to optimize the camera calibration system. This is an essential task for the retrieval of quantitative information about the captured images. However, the current methods require specific skills in image processing, which are not always available in industry. We propose to automate and optimize the calibration system by eliminating the selection of images by the operator. Second, we propose to improve the detection systems for thin and featureless abjects. Finally, we propose to adapt evolutionary algorithms to optimize search times.

The work deals with heterogeneous island models. The work designs and implements a new island model based on knowledge of homogeneous models of evolutionary algorithms. The model allows dynamic replanning of general computational methods. The work experimentally compares results of homogeneous and heterogeneous models.

Game playing is a relatively interesting task in the field of artificial intelligence in these days. The master thesis deals with general artificial intelligence which is capable of playing selected simple games based on information that is also avai- lable to the human player. Our selected games are 2048, Mario, racing simulator TORCS and Alhambra. All the information acquired by artificial intelligence is provided by games through an interface, therefore none of the models uses visual input. We use evolutionary approaches such as evolutionary algorithms, evolutio- nary strategy CMA and differential evolution applied to different types of neural networks. We are also dealing with deep reinforcement learning. We test these approaches and compare their results. 1

Absence epilepsy is a neurological disorder that commonly occurs in children. Some studies have shown that absence seizures predominantly originate either in the thalamus or the cerebral cortex. Some cerebellar nuclei (CN) neurons project to these brain areas, as explained further in Fig. 2.6 in Chapter 2. Also, some CN neurons have been observed to show modulation during the absence seizures. This indicates that they somehow participate in the seizure and hence are referred to as "participating neurons" in this thesis. In this research, I demonstrate how machine learning techniques and computer simulations can be applied to investigate the properties and the input conditions present in these participating neurons. My investigation found a sub-group of CN neurons, with similar interictal spiking activity, spiking activity between the seizures, that are most likely to participate in seizures. To investigate the input conditions present in the CN neurons that produce this type of interictal activity, I used a morphologically realistic conductance based model of an excitatory CN projection neuron [66] and optimised the input parameters to this model using an Evolutionary Algorithm (EA). The results of the EA revealed that these participating CN neurons receive a synchronous and bursting input from Purkinje cells and bursting input with long intervals(approx. 500ms) from mossy fibre. The same interictal activity can also be produced when the Purkinje cell input to the CN neuron is asynchronous. The excitatory input in this case also had long interburst intervals but there is a decrease in excitatory and inhibitory synaptic weight. Surprisingly, a slight change in these input parameters can change the interictal spiking pattern to an ictal spiking pattern, the spiking pattern observed during absence seizures. I also discovered that it is possible to prevent a participating CN neuron from taking part in the seizures by blocking the Purkinje cell input.