This work studies the generalizations of Zadeh’s Extension Principle through the change of the minimum function by another aggregation function. The mathematical details involved are studied in order to assure that real number functions are extended to functions of fuzzy numbers whose image only contains fuzzy numbers. The notion of φ-extension was introduced in order to formalize the definition of these extended mappings in a way to avoid inconsistencies with the definition of fuzzy number used. Some algebraic properties are explored aiming to understand how some characteristics of a real number function are preserved. Some applications of φ-extensions are also studied, like the extensions of binary operations of real numbers. Particularly, a new definition of the product operation of fuzzy numbers is used to define a generalization of ordered weighted averaging functions for fuzzy numbers with fuzzy numbers as weights.

The proliferation of intensive use of geolocation devices has significantly contributed to the considerable wear and tear on the battery life of these devices, resulting in difficulties in using certain applications. Faced with this issue, the present study introduces a comprehensive framework developed for the purpose of collecting, processing, and visualizing geolocation data on mobile devices, with a particular focus on smartphones. The system in question is composed of an Android library that enables the transmission of geolocation data while offering configuration options to enhance accuracy and reduce battery consumption. Additionally, a Java framework has been developed to receive and process this data, integrating with PostGIS extensions to ensure the acquisition of highly precise positions. Finally, a JavaScript library has been implemented to receive and display stored geolocations, providing a clear and intuitive understanding of underlying geographic patterns.

The increase in the complexity of fifth-generation (5G) mobile networks, caused by the high number of mobile devices, along with the increased demands with respect to new application requirements, requires the employment of management systems capable to provision 5G networks always best connected and best served. With this in mind, network operators employ Operation, Management, and Control (OMC) centers, an ecosystem
entailing different technologies and tools that interoperate to provide ultimate operations and management functions designed to keep the network alive with guaranteed levels agreed in SLA (Service Level Agreement) over time. An OMC is strategically constructed in a centralized environment facility and has the role of being able to respond to the occurrence of a "disaster", meaning events that potentially yield a certain level of network
service unavailability. As a way to guarantee some level of fault tolerance, redundant OMC instances must be adopted, where a backup OMC must assume control when a principal OMC instance fails. A Disaster Recovery System (DRS) has the goal of assigning which OMC instance takes the principal control inside a 5G ecosystem. To this, the DRS constantly monitors the 5G ecosystem in pursuit of detecting disaster events so that assigning a backup OMC to assume the principal control.

Based on the fact that DRS mainly operates in a reactive manner, meaning that the switching between OMCs is done after detecting a disaster occurrence, this master’s research devotes to exploiting Machine Learning techniques to efficiently control the assigning of multi-redundant OMCs in the control of 5G networks. To this end, the iDRS (Intelligent Disaster Recovery System) is introduced, which is based on predictive analyzes to act proactively in the attribution of OMCs control, in the hypothesis of provisioning an efficient and agile system to maintain 5G networks throughout their lifetime. 

Brain-Computer Interfaces (BCIs) enable interaction with computers through the user’s brain activity. In recent years, the BCIs have been gaining prominence due to technological advancements. In the context of BCIs, electroencephalography (EEG) is a common method for scanning brain activity. Among the available EEG devices, we can highlight the NeuroSky MindWave. This work presents Software designed to support BCI researchers in their studies, specifically those which use the MindWave as EEG device. This work details the usage context of the Software and provides the necessary documentation to understand its functioning. The Software tools are described, and all functionalities are presented to showcase the possibilities that can be achieved using the developed Software. Lastly, the Software is evaluated through a pilot study. With the execution of the study, it was possible to verify the functionality of the Software in a BCI research scenario.

Arduino embedded systems are used in teaching activities and applied in various automation scenarios. In addition to the literature review in articles, this research conducted a survey to investigate the tools used in the documentation and testing activities of the development of these systems. As a result of the research, the Guimarães framework is proposed for teaching the development of embedded systems with Arduino, which includes the execution of automated tests at the component and system levels, supported by documentation. In an application scenario, the development process was carried out using the Guimarães framework, which includes the development of requirement diagrams, system documentation using statecharts, the definition of test cases with analysis of the statechart’s path tree as a stopping criterion, documentation, and electronic hardware prototyping, development of software behavior and test cases using the Guimarães framework, and execution and analysis of the test cases.

The determination of hypocentral location uses techniques aimed at obtaining the spatial coordinates of the hypocenter from the temporal data observed at seismographic stations. Its applications range from the geological characterization of the Earth's crust to support for the prediction of new incidents and hydrocarbon exploration. In this context, this work proposes the application of \ac{olc}, hybridized with genetic operators of crossover, mutation, tournament, and elitism as an alternative to traditional meta-heuristics used for this problem. The \ac{olc} is an optimization algorithm inspired by the behavior of grey wolf packs, proposed by Seyedali Mirjalili in 2014, and has been applied in various areas, including optimization of complex problems, system design, and machine learning. The \ac{olc} is based on three main behaviors observed in wolf packs: hunting, searching, and communication. Through these behaviors, the algorithm is capable of efficiently exploring the search space, finding solutions close to the global optimum in optimization problems. The results of the proposed method were compared with other widely used meta-heuristics in hypocentral inversion, namely, \ac{gm}, \ac{ag}, and \ac{oep}. The criteria considered in the comparisons included Data Fitting, Cost of the Objective Function of Average Solutions, Computational Time, Average Solutions, Cost of the Objective Function in the Stabilization Period, Convergence, and Total Cost, and the tests were conducted in two different situations: on four synthetic seismic events and real seismic events. To evaluate the effectiveness of the method, a parametric statistical test was performed with \ac{af} and non-parametric tests of \ac{kw}, with post hoc test using \ac{db}. The results indicated that the \ac{olch} demonstrated superiority in the Cost of the Objective Function in the Stabilization Period on synthetic data compared to \ac{oep} and \ac{gm}, and to \ac{gm} in the Total Cost, while \ac{gm} was more efficient in computational time compared to \ac{olch}, and \ac{oep} outperformed \ac{olch} and \ac{gm} in convergence. In real data, costs of the objective function of average solutions, \ac{olch} was superior to \ac{oep}. In the cost of the objective function in the stabilization period, \ac{olch} also outperformed \ac{oep}. In total cost, \ac{olch} was superior to \ac{gm}, while \ac{ag} was better than \ac{gm}. In computational terms, \ac{gm} was superior to \ac{olch}, and in terms of convergence, \ac{oep} surpassed \ac{olch} and \ac{gm}. Thus, consolidating \ac{olch} as an innovative alternative in hypocentral location.

Copulas are functions that play an important role in probability theory. Since interval
probability takes into account the imprecision in the probability of some events, it is
likely that interval copulas have a relevant contribution to interval probability theory. This
article aims to introduce the definition and analysis of interval-valued copulas and their
properties. We pro- vide a condition for an interval-valued copula to be 1-Lipschitz and
from the interval-valued automorphisms we obtain the conjugate interval-valued copula
and some important inherited properties. We have seen that the Archimedean intervalvalued
copula, in most cases, has its behavior defined by its generative function. We also
show a condition for this generating function to generate an interval-valued copula and a
version of the Sklar’s theorem for representable interval-valued copulas.

Uma das principais atividades no desenvolvimento de software é o monitoramento, que desempenha um papel vital na verificação da implementação adequada de processos, identificação de erros e descoberta de oportunidades de melhoria. A Integração Contínua (CI) abrange um conjunto de práticas amplamente adotadas que aprimoram o desenvolvimento de software. No entanto, há indicações de que os desenvolvedores podem não monitorar adequadamente todas as práticas de CI. Nesta tese, mergulhamos profundamente no oceano do monitoramento das práticas de CI. Nosso objetivo é descobrir como esse monitoramento é conduzido, demonstrar as vantagens do monitoramento das práticas de CI e destacar os desafios que precisam ser superados. 

Em nosso primeiro estudo, analisamos o impacto de práticas específicas de CI no volume de Pull Requests e Issues relacionadas a bugs. Nossos resultados revelaram uma correlação positiva entre as práticas de CI e o aumento no número de merged pull requests. Nós também identificamos uma correlação significativa com o número de Issues relacionadas a bugs. Adicionalmente, nossos resultados sugerem que valores mais elevados de práticas de CI podem indicar uma melhor qualidade no processo de desenvolvimento.

Posteriormente, em nosso segundo estudo, investigamos a importância atribuída a essas práticas pelos desenvolvedores e o suporte ao monitoramento fornecido pelas ferramentas de CI mais populares. Descobrimos que geralmente os desenvolvedores monitoram apenas a cobertura e os metadados básicos da build (por exemplo, duração e status da build). Os desenvolvedores expressaram interesse em monitorar práticas de CI se tivessem oportunidade. Além disso, identificamos que vários dos serviços líderes de CI ainda possuem um suporte inicial para monitorar as práticas de CI.

Por fim, avaliamos o monitoramento em cenários reais, realizando um estudo de caso em três projetos de três organizações diferentes, no qual podemos verificar mais profundamente o interesse dos desenvolvedores pelo monitoramento de práticas de CI, seus benefícios, desafios e a evolução das práticas de CI durante um período de dois meses. O estudo de caso revelou que o monitoramento das práticas de CI oferece vários benefícios ao projeto e é pouco custoso de ser aplicado. Os participantes demonstraram um forte desejo de integrar dashboards de monitoramento de CI nos serviços de CI mais populares.

The use of radiotherapy for cancer treatment is essential for combating this disease. The challenge is to achieve the minimum dose prescribed for the tumor while avoiding exposure of healthy organs to radiation levels higher than the permitted limits. One of the main therapeutic approaches in this field is intensity-modulated teletherapy (IMRT). This study aimed to optimize the Beam Angle Optimization Problem and Fluency Map Optimization using metaheuristics. Three algorithms are presented: genetic algorithm, OWA-OMF memetic and multi-model memetic. All include the Ordered weighted averaging (OWA) operator. The multi-model memetic uses different OWA functions to determine the best fluence map for a solution. The algorithms were compared using a new quality indicator composed of the two new indices proposed in this study. Statistical tests were conducted to compare the effectiveness of these algorithms, revealing the superiority of the multi-model memetic algorithm over the others. With these algorithms, it was possible to find clinically viable solutions for most instances.

Our research presents a new approach to classifying macrocauses of crime, specifically
focusing on predicting and classifying the characteristics of ILVCs. Using a dataset
from Natal, Brazil, we experimented with five machine learning algorithms, namely
Decision Trees, Logistic Regression, Random Forest, SVC, and XGBoost. Our methodology
combines feature engineering, FAMD for dimensionality reduction, and SMOTE-NC for
data balancing. We achieved an average accuracy of 0.962, with a standard deviation of
0.016, an F1-Score of 0.961, with a standard deviation of 0.016, and an AUC ROC curve of
0.995, with a standard deviation of 0.004, using XGBoost. We validated our model using
the abovementioned metrics, corroborating their significance using the ANOVA statistical
method. Our work aligns with smart city initiatives, aiming to increase public safety and
the quality of urban life. The integration of predictive analysis technologies in a smart
city context provides an agile solution for analyzing macrocauses of crime, potentially
influencing the decision-making of crime analysts and the development of effective public
security policies. Our study contributes significantly to the field of machine learning applied
to crime analysis, demonstrating the potential of these techniques in promoting safer urban
environments. We also used the Design Science methodology, which includes a consistent
literature review, design iterations based on feedback from crime analysts, and a case
study, effectively validating our model. Applying the classification model in a smart city
context can optimize resource allocation and improve citizens’ quality of life through a
robust solution based on theory and data, offering valuable information for public safety
professionals.

This research addresses the problem of insufficient or inefficient requirements specification to meet informational needs in software development in compliance with the regulations and laws applicable to Digital Health systems. In particular, it highlights the importance of ensuring that legal requirements, such as those stipulated by the General Data Protection Law (GDPL), are integrated and implemented in the development process. The overall objective of this work is to develop a requirements specification artifact that complies with the GDPL in the context of agile development of digital health solutions. This artifact aims to establish a Requirements Engineering strategy that ensures legal compliance, facilitating its practical application in software development. In addition, the artifact should have an educational character, being useful for students and teachers, allowing a practical integration of the concepts of privacy and data protection in both the development and teaching processes.  The methodology of this research was divided into three parts: (1) the phase of preparing the extension proposal; (2) the phase of preparing the evaluation of the extension proposal; and (3) the phase of applying the evaluation of the proposal. In evaluating the proposal, an evaluation tool was developed for two profiles of participants (information technology professionals and teachers). We obtained a total of 24 responses to the forms. The results show that the adoption of the proposal to extend the requirements can be favorable, as it has the potential to help identify and effectively communicate the requirements appropriate to the GDPL to agile development teams, facilitating compliance with legal requirements and data protection. In addition, an important educational potential was identified in the proposal, as the participating teachers assessed that it can be used as a suitable tool for teaching about requirements and the GDPL in the classroom.

Por uma questão de flexibilidade, as linguagens de programação orientadas a objetos convencionais admitem valores nulos para referências. Na linguagem de programação Java, o uso de uma referência de objeto com um valor nulo causa o lançamento de uma exceção do tipo Null Pointer Exception (NPE), uma das causas mais frequentes de falhas em aplicações escritas nessa linguagem. A análise estática tem sido utilizada para inspecionar artefatos de software como código fonte ou código binário visando localizar a origem de faltas sem que seja necessário executar o programa de forma orientada a depuração. Apesar de sua eficácia, a análise estática baseia-se em um conjunto fixo e estático de regras que descrevem padrões de ocorrência de faltas e é conhecida por um número significativo de falsos positivos. Este estudo investiga como o uso de técnicas de Aprendizado de Máquina (AM) pode melhorar a precisão da detecção de faltas relacionadas a NPE por meio de análise estática, uma linha ainda inexplorada na literatura e na indústria de software. O objetivo principal é propor, implementar e avaliar uma abordagem baseada em classificação que enderece o problema da detecção de faltas relacionadas a NPE em código Java. As contribuições deste trabalho são: (i) uma análise de como as técnicas de AM podem ser usadas para detectar essas faltas por meio de análise estática e (ii) uma avaliação do desempenho das técnicas de AM em comparação às ferramentas tradicionais de análise estática.

Accurate forecasting of energy consumption can significantly contribute to improving distribution management and 

potentially contribute to controlling and reducing energy consumption rates. Advances in data-based computational

 techniques are becoming increasingly robust and popular as they achieve good accuracy in results. This study 

proposes the development of a model capable of classifying energy consumption profiles in the residential sector, 

using machine learning and transfer learning techniques. The application of Machine Learning (MA) techniques in 

energy production can indicate great potential for controlling and managing the production and distribution of 

electric energy, which can bring greater efficiency, improve production and optimize distribution. In this study, we 

combine AM techniques with the transfer of learning that is able to use pre-established knowledge in new contexts 

(knowledge bases), making the energy forecasting process more efficient and robust.

Big Data é um termo genérico que geralmente se refere a conjuntos de dados cujo tamanho cresce para além da capacidade dos métodos e ferramentas tradicionais de coletar, armazenar, processar e analisar dados em um tempo tolerável e utilizando recursos computacionais de forma razoável. Sistemas de Big Data (SBD) podem ser encontrados em diversas áreas, provendo insights e informações úteis a organizações e usuários. A complexidade e as características intrínsecas a esses sistemas requerem arquiteturas de software para satisfazer adequadamente requisitos funcionais e de qualidade. Arquiteturas de referência (ARs) são consideradas um ativo importante na construção de arquiteturas de software uma vez que elas promovem reuso de conhecimento e orientam seu desenvolvimento, padronização e evolução. Entretanto, muitas arquiteturas de referência para BDS ainda são produzidas utilizando uma abordagem ad-hoc sem seguir um processo sistematizado para seu projeto e avaliação. Este trabalho propõe a Modern Data Reference Architecture (MoDaRA), uma AR para SBD fundamentada em um processo sistemático que agrega prática da indústria e conhecimento acadêmico nesse domínio. O projeto da MoDaRA seguiu o ProSA-RA, um processo bem definido para guiar a definição de ARs, compreendendo fases como análise, síntese e avaliação arquitetural estruturadas sobre fontes de informação selecionadas. A MoDaRA foi avaliada considerando dois casos de uso da indústria e um checklist para avaliação de ARs adaptado a SBD.

Quasi-Nelson logic is a recently introduced generalization of Nelson's cons\-tructive logic with strong negation to a non-involutive setting. The present work proposes to study the logic of quasi-Nelson pocrims ($\mathbf{L}_{\mathrm{QNP}}$) and the logic of quasi-N4-lattices ($\mathbf{L}_{\mathrm{QN4}}$). This is done by means of an axiomatization via a finite Hilbert-style calculus. The principal question which we will address is whether the algebraic counterpart of a given fragment of quasi-Nelson logic (or class of quasi-N4-lattices) can be axiomatized abstractly by means of identities or quasi-identities. Our main mathematical tool in this investigation will the twist-algebra representation. Coming to the question of algebraiza\-bility, we recall that quasi-Nelson logic (as extensions of $\mathbf{FL_{ew}}$) is obviously algebraizable in the sense of Blok and Pigozzi. Furthermore, we showed the algebraizability of $\mathbf{L}_{\mathrm{QNP}}$ and $\mathbf{L}_{\mathrm{QN4}}$, which is BP-algebraizable with the set of defining identity $E(\alpha) := \{ \alpha \approx \alpha \to \alpha \}$ and the set of equivalence formula $\Delta(\alpha, \beta) := \{ \alpha \to \beta, \beta \to \alpha, \nnot \alpha \to \nnot \beta, \nnot \beta \to \nnot \alpha \}$. In this document, we register the achieved results up to the present moment and indicate a plan for the developments to be included in the final version of this thesis.

In recent years, consensus algorithms have become a fundamental part of fault-tolerant distributed systems. For many years, the widely disseminated consensus algorithm that served as the basis among the various applications was Paxos. Several other consensus algorithms and variations tolerated in Paxos have emerged in recent years, one of which is the Raft consensus algorithm. Raft came up with the initiative to make the understanding of the Paxos consensus mechanism simpler and more intuitive, both in the educational area and in practical implementations. In this way, it has become one of the most used algorithms in real implementations. A trend that computational systems have followed in recent years is the exchange of the use of batteries as an electrical power source for an approach that aims to collect energy from the environment for the computational operation of the device (Energy Harvester). This trend is mainly due to issues related to the environment, which suffer from the large amount of batteries coming from these types of devices and which will be discarded over the next few years. However, this approach of dealing through the collection of energy from the environment has some challenges that must be overcome, the main one being dealing with the dynamicity of the process of collecting energy from the environment in comparison with a system powered by an unlimited energy source by a certain period, which is the battery. To deal with this dynamicity of energy collection from the environment, an adopted measure is energy-driven computing, these are systems that are architecturally designed around the instability of energy collection from environments, so these systems seek adjust its operation according to the amount of energy being collected at that particular moment, aiming at reducing energy consumption in a period of savings in collection.

With this, this master's research aims to propose changes in the Raft consensus algorithm to satisfy its use in systems that collect energy from the environment, based on the concepts of energy-driven computing (Energy-Driven Computing). For this, it is necessary to bring a certain level of knowledge of the energy situation to the nodes (nodes) that are operating the Raft in a cluster, and through this knowledge of the current situation, each one satisfies its operations, observing the consumption of energy consumption , but without compromising the operation of the cluster as a whole.

Multidimensional fuzzy sets is a new extension of fuzzy sets on which the membership values of an element in the universe of discourse are increasingly ordered vectors on the set of real numbers in the interval [0, 1]. The main application of this type of set are the multi-criteria group decision making problems, in which, in the n-dimensional case, we have a set of situations, which are always evaluated by a fixed number n of experts. The multidimensional case is used when some of these experts refrain to evaluate some of these situations and, therefore, may be suitable for solving multi-criteria group decision making problems with incomplete information. This thesis aims to investigate fuzzy negations and fuzzy implications on the set of increasingly ordered vectors on [0, 1], i.e. on L∞ ([0, 1]), with respect to some partial order. In this thesis we study partial orders, giving special attention to admissible orders on L∞ ([0, 1]). In addition, some properties and methods to construct and generate such operators from fuzzy negations and fuzzy implications, respectively, are provided (in particular, a notion of ordinal sums of n-dimensional fuzzy negations and ordinal sums of multidimensional fuzzy negations will be proposed with respect to specific partial orders) and we demonstrate that an action of the group of automorphisms on fuzzy implications on L∞ ([0, 1]) preserves several original properties of the implication. Using a specific type of representable multidimensional fuzzy implication, we are able to generate a class of multidimensional fuzzy negations called natural m-negations. In the end, an application in decision-making problems is presented.

The critical nature of many complex software-intensive systems requires formal architecture descriptions for supporting automated architectural analysis regarding correctness properties. Due to the challenges of adopting formal approaches, many architects have preferred using notations such as UML, SysML, and their derivatives to describe the structure and behavior of software architectures. However, these semi-formal notations have limitations regarding the support for architectural analysis, particularly formal verification. This work investigates how to formally support SysML-based architecture descriptions to enable the formal verification of software architectures. As a result of this research, the main contribution is proposing a model-driven approach (MDD) that provides formal semantics to a SysML-based architectural language, SysADL, through a seamless transformation of SysADL architecture descriptions to the corresponding formal specifications in \piadl, a well-founded theoretically language based on the higher-order typed Pi-calculus. The proposal implementation involves the execution of a 4-phase process: (i) Model-to-Model (M2M) transformation of SysADL models into piadl model; (ii) Model-to-text (M2T) transformation of PI-ADL models into PI-ADL code; (iii) corresponding executable architecture generation, and architecture validation; and (iv) property verification. The work has other associated contributions to support the 4-phase process: (i) a denotational semantics to SysADL as a function of PI-ADL; (ii) a definition of a process to support the automated transformation of SysADL models into piadl models; (iii) The validation of the PI-ADL architecture generated by the MDD transformation to show that it is in accordance with the original SysADL architecture; and (iv) the verification of formal architectural properties using execution traces. The proposal was implemented and validated using a Flood Monitoring System architecture.

Natural Language Requirements Specification is a common type of documentation used in systems development to contain information about requirements. The quality of this information is important for the success of activities, systems development, and requirements management itself. However, Natural Language can lead to ambiguity and other defects that end up compromising Quality Criteria for information about requirements, such as understanding, clarity, completeness and others. A set of these structural defects were studied by Henning Femmer and named as Requirements Smells. Requirements Smells are anomalies related to problems in writing requirements using Natural Language. Furthermore, these anomalies can occur in any Natural Language Specification template. However, despite the literature containing an expressive amount of studies on the behavior and relationship of Requirements Smells with the quality of requirements, there are no studies on ways of systematized correction in Requirements Smells and how these corrections can help restore/achieve  Quality Criteria for requirements specified in Natural Language. The objective of this work is to develop a set of systematized corrections, entitled Refactoring Techniques, to correct Requirements Smells and help to achieve the Quality Criteria. The Refactoring Techniques will be developed following the structure suggested by Martin Fowler, one of the pioneer authors in code refactoring. Experiments will be carried out with participants who will review requirements and identify Requirements Smells and correct these requirements using the proposed Refactoring Techniques. For this, a set of 33 User Stories infected by Requirements Smells will be made available to users. At the end, they will answer questionnaires whose information will be analyzed and
understood. It is expected that the Refactoring Techniques developed can be effective in correcting and reaching the Quality Criteria and become another alternative solution for Requirements Smells correction.

Legal AI, the application of Artificial Intelligence (AI) in the legal domain, is a research field that comprises several dimensions and tasks of interest. As in other targeted application domains, one of the desired benefits is task automation, which increases the productivity of legal professionals and makes law more accessible to the general public. Text is an important data source in the legal domain, therefore Legal AI has a great interest in the Natural Language Processing advances. This thesis concerns the automation of the Legal Rhetorical Role Labeling (RRL), a task that assigns semantic functions to sentences in legal documents. Legal RRL is a relevant task because it finds information that is useful both by itself and for downstream tasks such as legal summarization and case law retrieval. There are several factors that make legal RRL a non-trivial task, even for humans: the heterogeneity of document sources, the lack of standards, the domain expertise required, and the subjectivity inherent in the task. These complicating factors and the large volume of legal documents justify the automation of the task. Such automation can be implemented as a sentence classification task, i.e. sentences are fed to a machine learning model that assigns a label or class to each sentence. Developing such models on the basis of Pre-trained Transformer Language Models (PTLMs) is an obvious choice, since PTLMs are the current state of the art for many NLP tasks, including text classification. Nevertheless, in this thesis we highlight two main problems with works that exploit PTLMs to tackle the Legal RRL task. The first one is the lack of works that address how to better deal with the idiosyncrasies of legal texts and the typically small size and imbalance of Legal RRL datasets. Almost all related works simply employ the regular fine-tuning strategy  to train models.

The second problem is the poor utilization of the intrinsic ability of PTLMs to exploit context, which hampers the performance of the models. This thesis aims to advance the current state of the art on the Legal RRL task by presenting three approaches devised to overcome such problems. The first approach relies on a data augmentation technique to generate synthetic sentence embeddings, thus increasing the amount of training data. The second approach makes use of positional data by combining sentence embeddings and positional embeddings to enrich the training data. The third approach, called Dynamically-filled Contextualized Sentence Chunks (DFCSC), specifies a way to produce efficient sentence embeddings by better exploiting the encoding capabilities of PTLMs. The studies in this thesis show that the first two approaches have a limited impact on the performance of the models. Conversely, models based on the DFCSC approach achieve remarkable results and are the best performers in the respective studies. Our conclusion is that the DFCSC approach is a valuable contribution to the state of the art of the Legal RRL task.

In this work, we introduce several contrapositivisation operators for fuzzy implications, we present a wide study of each of these operators with respect to the main properties routinely required by fuzzy implications, we prove that the classes of these contrapositivisators are invariant by automorphisms and present some conditions for the $N$-compatibility of the respective contrapositivisations, we propose some construction methods of classes of triangular norms (quasi-overlaps), triangular conorms (quasi-groupings) and aggregation functions from these contrapositivisators; we introduce the Min-Max contrapositivisation technique for fuzzy implications and some of its generalizations; we introduce four contrapositivisation operators for fuzzy coimplications so-called co-upper, co-lower, co-medium and co-($S$,$N$)- contrapositivisators, we characterize these operators from the point of view of the properties that are usually attributed to the fuzzy coimplications, we present sufficient conditions for the $N$-compatibility of the co-upper, co-lower, co-medium and co-($S$,$N$)- contrapositivisations, we show that the classes of co-upper, co-lower, co-medium and co-($S$,$N$)- contrapositivisators are closed under the action of automorphisms and we propose a construction method of triangular conorms from co-($S$,$N$)-contrapositivisators of ($T$,$N$)-coimplications and fuzzy negations; and finally, we propose the classes of interval-valued upper, lower and medium contrapositivisators and broadly characterize each one of them, we show that both classes are invariant by interval-valued automorphisms, we introduce the notion of $N$-compatibility for the respective interval-valued contrapositivisations and we prove that the best interval representations of real-valued upper, lower and medium contrapositivisators are, respectively, interval-valued upper, lower and medium contrapositivisators.

The hybridization of metaheuristics is a topic that several researchers have studied due to its potential to produce more efficient heuristics than those based on a single technique. However, hybridization is not easy, as there are several ways to operationalize it. The task becomes even more challenging when three or more metaheuristic methods need to hybridize or when someone wants to add Mathematical Programming methods, thus creating matheuristics. Various methods have been proposed to hybridize metaheuristics, including some techniques that automate hybridization, such as multi-agent architectures. A few of these architectures use learning techniques, and an even smaller number deal with matheuristics. This work extends the capabilities of the Multi-agent Architecture for Metaheuristic Hybridization by including learning techniques and Mathematical Programming. The application of learning techniques is innovative, considering the agents' choice of heuristics to apply at different search stages. This work proposes a new form of hierarchical hybridization for Combinatorial Optimization problems with multiple decision levels. The algorithmic proposals are tested on the Traveling Car Renter with Passengers and the Cable Routing Problem in Wind Farms. These problems belong to the NP-hard class and require decision-making at multiple levels. In the case of the Traveling Car Renter with Passengers, there are three decision levels: route, car types, and customers' transport demand. Cable routing in wind farms requires decisions concerning the cable locations and the cable type used in each section. The experiments for the Traveling Car Renter with Passengers were conducted on three classes of instances, totaling ninety-nine test cases ranging from four to eighty cities, two to five vehicles, and ten to a hundred forty people requiring transportation. Experiments for The Cable Routing Problem in Wind Farms involved a set of two hundred instances. These instances are simulations of real situations developed in collaboration with domain experts. The approaches proposed in this work are compared to state-of-the-art algorithms for both problems.

Smart city application development platforms receive, store, process, and display large volumes of data from different sources and have several users, such as citizens, visitors, government, and companies. The underlying computing infrastructure to support these platforms must deal with the highly dynamic workload of the different applications, with simultaneous access from multiple users and sometimes working with many interconnected devices. Such an infrastructure typically encompasses cloud platforms for data storage and
computation, capable of scaling up or down according to the demands of applications. This thesis proposes an autonomic approach for autoscaling smart city platform infrastructures. The approach follows the MAPE-K control loop to dynamically adjust the infrastructure in response to workload changes. It supports scenarios where the number of processing requests is unknown a priori. The performance of the approach has been evaluated upon the computational environment that supports Smart Geo Layers (SGeoL), a platform for developing real-world smart city applications.

Fake News has been a big problem for society for a long time. It has been magnified,reaching worldwide proportions, mainly with the growth of social networks and instantchat platforms where any user can quickly interact with news, either by sharing, throughlikes and retweets or presenting hers/his opinion on the topic. Since this is a very fastphenomenon, it became humanly impossible to manually identify and highlight any fakenews. Therefore, the search for automatic solutions for fake news identification, mainlyusing machine learning models, has grown a lot in recent times, due to the variety oftopics as well as the variety of fake news propagated. Most solutions focus on supervisedlearning models, however, in some datasets, there is an absence of labels for most of theinstances. For this, the literature presents the use of semi-supervised learning algorithmswhich are able to learn from a few labeled data. Thus, this work will investigate the use ofsemi-supervised learning models for the detection of fake news, using as a case study theoutbreak of the Sars-CoV-2 virus, the COVID-19 pandemic.

Errors during the software development may give rise to flaws in the system that can cause important damages. One of the most important stages in the software development process is modelling the system architecture, possibly using software architecture description languages~(ADLs). The ADLs currently adopted by industry for software-intensive systems are mostly semi-formal and essentially based on SysML and specialized profiles. These ADLs allow describing the structure and the behavior of the system. Besides, it is possible to generate executable models or generate code in a target programming language and simulate its behaviour. This, however, does not constitute proof that the system is correct or safe. This work proposes a novel approach for empowering SysML-based ADLs with formal verification supported by model checking. It presents a CSP-based semantics to SysADL models. Furthermore, this work presents how correctness properties can be formally specified using CSP, and how the FDR4 refinement model-checker can verify these correctness properties. Finally, we present the new extension to SysADL studio that allows the automated transformation from SysADL architecture descriptions to CSP processes and the verification of important system correctness properties. The whole approach is illustrated via a case study, which is also part of this document. This case study demonstrates the usefulness of our approach in practice.

The Software-Defined Networking (SDN) paradigm has been widely used in diverse ecosystems as enabler for the management of heterogeneous administrative domains, extend programmable resources to intra-domain networks, or even compose cloud-native network architectures. On the other hand, while it can support the ability of next-generation networks to adapt to new protocols, SDN increases the scope of attack vectors to the network, resulting in several security issues. From this point of view, control applications running atop the SDN controller are responsible for establishing secure connections between the underlying node pairs. The secure exchange of cryptographic keys, so that two interconnected nodes can communicate securely over a public channel, represents a well-known challenge in symmetric cryptography systems field of research. The Diffie–Hellman (DH) and Advanced Encryption Standard (AES) stands to a widely adopted solution for exchanging cryptographic keys and encrypting traffic between nodes over untrusted networks. However, traditional cryptographic implementations impose high computational costs and key management risks, which can result to problems in the centralized control plane of the SDN network. This research sets out by exploring the Programming Protocol-independent Packet Processors (P4) paradigm, and proposes the dh-aes-p4 as the first solution for exchanging DH keys with AES adapted tailored to P4-based SDN devices. Although there exist similar cases in the literature, this work distinguishes itself as a new, low-cost, granular (based on network flows) and transparent alternative.

In distributed systems, the first step to establish communication with another device is to know its address, that is, to locate it. To locate services or resources automatically, discovery systems have been applied to diverse environments and usage contexts, from wireless sensor networks (WSNs) and peer-to-peer systems, to high processing clusters and cloud systems. The great diversity among usage contexts and application needs has led to the development of specialized discovery protocols, often incompatible with each other. This incompatibility prevents discovery across heterogeneous environments or protocols, restricting the services accessible to a given device. Therefore, to address these limitations, it is necessary to provide discovery solutions that integrate heterogeneous discovery environments and protocols. This in turn requires understanding: in which aspects these environments and protocols vary, and also what aspects they have in common. To address this issue, this paper presents a review of secondary studies from the literature which address service discovery and resource discovery across different environments, i.e. a tertiary study on the topic. Based on this review, a solution was developed to integrate the service discovery across heterogeneous discovery environments and protocols. A proof-of-concept of this solution was implemented, along with two discovery mechanisms: one to local service discovery and the other to decentralized discovery over the Internet. To evaluate the feasibility of the solution and analyze how these mechanisms interact with each other, a controlled experiment was conducted in a virtual network environment. Despite limitations and challenges that still remain, this research can contribute to the understanding of discovery systems, in what they have in common and in their points of variation, and move towards the “universal discovery” of services, which may allows the construction of new kinds of applications.

Approximate Computing is currently an emerging paradigm that seeks to replace some data accuracy with aspects such as performance and energy efficiency. At the software level, there are tools within this scope that apply some approximate computation techniques. However, these tools are limited in covering only some specific scope, applying only one of the techniques, and/or needing manual annotations on applications. The current state of the art still has open questions, such as wether application features influentiates the technique's choice; what would be the most appropriate technique for each particular context. Thus, this dissertation proposes the implementation of a tool that, according to the application profiling, chooses the most appropriate approximate computing technique to be applied. The tool uses the LLVM compilation infrastructure, where each step is implemented in the form of an LLVM Pass of code analysis or transformation. In addition to the Profiler, it was also implemented three approximate computing techniques and the experimental results show the technique chosen by the tool presents a balance between error rate and speedup.

Autism spectrum disorder (ASD) is a neuro-developmental disability marked by deficits in communicating and interacting with others. The standard
protocol for diagnosis is based on fulfillment of a descriptive criteria, which
does not establish precise measures and influence the late diagnosis. Thus,
new diagnostic approaches should be explored in order to better standardise
practices. The best case scenario would be to have a reliable automated
system that indicates the diagnosis with an acceptable level of assurance.
At the moment, there are no publicly available representative open-source
datasets with the main aim of this diagnosis. This work proposes a new methodology for collecting a Face Biometrics dataset with the aim to investigate the differences in facial expressions of ASD and Typical Developmental (TD) people. Thus, a new dataset of facial images was collected from YouTube videos, and computer vision-based techniques were used to extract image frames and filter the dataset. We have also performed initial experiments using classical supervised learning models as well as ensembles and managed to archive promising results.

Platforms for developing smart cities applications are responsible for providing various services to facilitate application development. Typically, such platforms manage a variety of applications, handle a large volume of data and serve a significant number of users that generate a high volume of requests. The large amount of requests often causes overload on the platform, degrading the quality of service provided to users. Also, as smart city platforms process requests for operations on large volumes of geographic data, it is important to monitor the databases to see if there are any limitations to processing large amounts of data in an acceptable time. Therefore, in this context, it is necessary to monitor the underlying computational infrastructure on which platforms for smart cities and applications are deployed, as well as monitor operations regarding access to geographic data stored in the databases used by the platforms. Aiming to address this problem, the objective of this work is to propose and implement a non-invasive strategy to enable the monitoring of platforms for smart cities, including the monitoring of the underlying infrastructure, as well as the operations directed to the databases. The proposed strategy is based on the aspect-oriented programming paradigm so that it is possible to monitor the computational infrastructure without the need to intervene on the implementation of the platform or generate coupling with respect to monitoring. This work also presents the implementation of the strategy and its instance in the monitoring of the Smart Geo Layers (SGeoL) platform, as well as an evaluation of the proposed monitoring strategy.

DevOps consiste em um conjunto de práticas que ajudam a lidar com conflitos entre as equipes de desenvolvimento e operação e busca garantir liberações de versões de software que sejam rápidas e confiáveis. O entendimento dessas práticas é essencial para a atuação de engenheiros de software na indústria. Neste sentido, a educação DevOps assume a tarefa vital de preparar os novos profissionais, através do ensino dessas práticas utilizando métodos de ensino adequados. O trabalho de pesquisa existente mostra que os métodos de ensino são úteis para os educadores desenvolverem e melhorarem seus cursos de DevOps. No entanto, há um número insuficiente de estudos investigando métodos de ensino na educação em DevOps. Nesta dissertação, realizamos dois estudos empíricos buscando compreender os métodos de ensino utilizados na educação em DevOps. No primeiro estudo, investigamos os métodos de ensino disponíveis na literatura. No segundo estudo, analisamos os métodos de ensino aplicados a partir de entrevistas com educadores do DevOps de cursos DevOps existentes. O objetivo do nosso trabalho é orientar novos educadores de DevOps a alcançar uma melhor experiência de ensino. Como resultado dos estudos, apresentamos um conjunto abrangente de 23 métodos de ensino, incluindo métodos de ensino tradicionais (aulas formais) bem como métodos de ensino menos usuais, tais como, aprendizado baseado em estúdio. Aprendizagem baseada em projetos e aprendizagem colaborativa foram os métodos de ensino mais recorrentes encontrados em ambos os estudos. A maioria destes métodos de ensino requerem uma maior interação entre educadores e alunos. Apresentamos também vinculações entre os métodos de ensino e os desafios. Estabelecemos tais vinculações durante o estudo I com base numa análise de estudos empíricos sobre métodos de ensino, enquanto as vinculações do estudo II vieram de uma análise de um estudo relacionado. Tais vinculações podem ajudar os educadores na seleção dos métodos de ensino do curso, onde o educador pode escolher os métodos que lidam com os desafios do seu contexto de ensino.

The aim of this thesis is to study a family of logics, comprised of Nelson’s logic S, constructive logic with strong negation N3, quasi-Nelson logic QN and quasi-Nelson implicative logic QNI. This is done in two ways. The first is by means of an axiomatisation via a Hilbert Calculus and the second is by studying some of the properties of the corresponding quasi-variety of algebras. The main contribution of the thesis is to prove that these logics fit within the theory of algebraisable logics. Making use of this result, the following are also proven. Regarding S, we introduced its first semantics, axiomatised by means of a finite Hilbert-style calculus, as well as established a version of the deduction theorem for it. Regarding QN and QNI, we showed that both are algebraisable with respect to the class of quasi-Nelson algebras and quasi-Nelson implication algebras, respectively; we showed that they are non-self-extensional; we showed how to obtain from them, by axiomatic extensions, other well-known logics, such as the {->, ~}-fragment of intuitionistic propositional logic, the {->, ~}-fragment of Nelson’s constructive logic with strong negation and the classical logic; and finally, we made explicit the quaternary term that guarantees that both QN and QNI satisfy the deduction theorem. Regarding N3, we study the role of the Nelson identity ((φ -> (φ -> ψ))∧(~ ψ -> (~ ψ -> φ)) = φ -> ψ) in establishing order-theoretic properties for its algebraic semantics. Moreover, we have studied the ⟨^, v, ~, ¬, 0, 1⟩-subreducts of quasi-Nelson algebras, and by making use of their twist representation, proved that this object-level correspondence can be stated as a categorical equivalence. Lastly, it is worth noting that QN I is the {->, ~}-fragment of QN , so some results concerning QNI may be easily extended to QN.

Ensembles of classifiers is a method in machine learning that consists in a collection of classifiers that process the same information and their output is combined in some manner. The process of classification is done in two main steps: the classification step and the combination step. In the classification step, each classifier processes the information and provides an output, in the combination step, the output of every classifier is combined, providing a single output. Although the combination step is extremely important, most works focus mostly on the classification step. Therefore, in this work, generalizations of the Choquet Integral will be proposed to be used as a combination method in ensembles of classifiers. The main idea is to allow a greater freedom of choice for functions in the integral, opening possibilities for otimization and using functions adequate to the data. Furthermore, a new notion of partial monotonicity is proposed, and consequently an alternative to the notion of pre-aggregation functions. Preliminary results that were obtained by the generalizations of the Choquet integral in the ensemble showed that they were capable of obtaining good results, having a superior performance to known methods in literature such as XGBoost, Bagging, among others. Furthermore, the generalizations that used the proposed aggregation functions had good performance when compared to other classes of functions, such as Copulas and Overlaps. 

We present a state-based Fuzzy model called Reversal Fuzzy Switch Graphs (RFSG). This model enables the activation or deactivation of edges as well as the updating of fuzzy values from the action of aggregation functions, whenever a transition occurs between the states. The Fuzzy feature of RFSGs allows you to model uncertainties whereas th activation and deactivation of edges allow the simulation of dynamic aspects of access to system states. When more than one aggregation function is used in this process, we have the Reversal Fuzzy Reactive Graphs (RFRG).
In addition, we propose some operations that are based on aggregation functions (unions, intersections, Cartesian product and extension). We also present the relationship between the RFRGs and the usual Fuzzy graphs together with a notion for simulation and bisimulation. We also introduce the concept of homomorphism between RFSGs and a modal logic for verifying properties of systems that are modeled by such graphs.

In the Smart Cities environment, applications development and execution face important challenges related to 1) Big Data Concept: The huge amount of processed and stored data with various data sources and data types; 2) Multi-domains: the many involved domains (Economy, Traffic, Health, Security, Agronomy, etc); and 3) Multiples processing methods like Data Flow, Batch Processing, Services, and Microservices. In facing these challenges, many platforms, middlewares, and architectures have been proposed to run applications in the Smart City’s environment. Despite all the progress already made, the vast majority of solutions have not met the functional requirements of Application Development, Application Deployment, and Application Runtime. Some studies point out that in a universe of 97 platforms, only 20.6% met the functional requirements of Application Development, Application Deployment, and Application Runtime. And, when those requirements are related to Scalability (non-functional), this number goes to 0.01%. Due to the lack of solutions that explore these requirements, all these concerns on Smart City’s Application Developing are passed on to the various stakeholders. For example, while Service Providers are concerned with: How to measure, charge, deploy, scale-up or scale-down, and execute to efficiently use the Computing Infrastructure, for developers it is important to know: How to implement, execute, scale application components, where to store their data, and where to deploy (Cloud, Fog, or Edge). In this work, we seek to outline and answer some of these questions. To this, we propose to build an evolutionary model for organizing and executing applications in the context of Smart Cities, the Smart City Application Architectural Model (Sapparchi). Sapparchi is an integrated architectural model for Smart Cities applications that define multi-processing levels (at the moment can support Edge, Fog, and Cloud). In addition, solutions are presented for monitoring, deploying, and scaling applications deployed at Cloud, Fog, and Edge levels. Finally, we present the Sapparchi middleware platform for developing, deploying, and running applications in the smart city environment with a focus on self-scaling and multi-processing computacional levels (From Cloud to Edge)

The logistics research field has observed the growing use of electric vehicles in different branches, including passenger transport. The PCVEP is a logistics problem that mixes elements of the well-known Traveling Salesman Problem (PCV), the Traveling Salesman with Passengers Problem (PCVP) and the Traveling Time Electric Salesman Problem (PCVEJT), in addition to also considering, restrictions for increasing or recharging travel autonomy and the fact that the autonomy is sensitive to the number of passengers loaded in the vehicle, all intrinsic to the problem itself and which further hamper the ability to solve the problem. The accomplishment of this work consisted of the research and the study of Problems of Routing of Electric Vehicles (PRVE) and of problems that address issues of ridesharing. In the first moment, a bibliographic survey of the works that addressed and solved the issues mentioned above was carried out for the proper formulation and description of the PCVEP. Therefore, because it is an unprecedented problem, a bank of Euclidean instances was created for the problem, a random part and an adapted part of TSPLIB. As solving methods for PCVEP, a set of heuristic, naive and hybridized algorithms was developed, for the proper anchoring of the experiments. Yet another set of meta-heuristic algorithms was developed for PCVEP, a randomized greedy procedure, improved through a search in descending variable neighborhood, and a colony algorithm of multi-ant ants, with ants that admit in their solving process , characteristics specific to PCVEP, such as charging stations, distance between locations and passenger loading. PCVEP addresses and solves PRVE, bringing up important sustainability issues, promoting the minimization of greenhouse gas emissions into the atmosphere, reducing traffic in large cities, and also encouraging socialization among people.

The advances in integration capacity of the chips allowed the emergence of systems with several processing cores, with networks-on-chip becoming the main paradigm in the communication between elements of multi-processed systems. Several proposals have emerged in order to meet mainly restrictions of average latency, area, energy consumption. The projects also cover the network architecture, with the generation of topologies that provide optimized performance for specific applications. This work proposes a heuristic for the generation of fault-tolerant topologies capable of delivering real-time and non-real-time packets via an alternative path within the network in the event of a link failure. Exploration always starts from a regular mesh-2D topology and seeks fault-tolerant topologies that are able to deliver as many packages as possible on time. The development of the implementation is based on the NOC42 simulator, making it capable of working with irregular topologies, real-time packages and a routing algorithm based on a routing table.

Handwritten signature is one of the oldest and most accepted biometric authentication methods for human identity establishment in society. With the popularisation of computers and, consequently, computational biometric authentication systems, the signature was chosen for being one of the biometric traits of an individual that is likely to be relatively unique for every person. However, when dealing with biometric data, including signature data, problems related to high dimensional space, can be generated.  Among other issues, irrelevant, redundant data and noise are the most significant, as they result in a decreased of identification accuracy. Thus, it is necessary to reduce the space by selecting the smallest set of features that contain the most discriminative features, increasing the accuracy of the system. In this way, our proposal in this work is to analyse the impact of feature selection on individuals identification accuracy based on the handwritten online signature. For this, we will use two well-known online signature databases: SVC2004 and xLongSignDB. For the feature selection process, we have applied two filter and one wrapper methods. Then, the resulted datasets are evaluated by classification algorithms and validated with a clustering technique. Besides, we have used a statistical test to corroborate our conclusions. Experiments presented satisfactory results when using a smaller number of features which are more representative, showing that we reached an average accuracy of over 98\% for both datasets which were validated with the clustering methods, which achieved an average accuracy over 80\% (SVC2004) and 70\% (xLongSignDB).

Embedded systems, especially context-aware systems, whose behaviour is determined
by information constantly obtained by different kinds of sensors, can be very hard to be
tested. That happens due to the nature of their input data, which can be hard to
replicate, and also because of their limited resources. Thus, software testing techniques
that may work for “common” software can be insufficient for this kind of system. Tools
created to support the testing activity of embedded systems are often limited to unit
tests, and avoid having to deal with data received from sensors, which is actually the
foundation of context-aware systems. To support software testing on context-aware
systems, this work proposes (i) an approach to collect and document the variation of
context data captured by sensors over time, (ii) the concept of context driven testing
and (iii) the development of a tool called Context Based Testing framework (ConBaT
framework) to help collecting context data and creating context driven tests for Arduino
systems.

Every year, millions of people are affected by chronic wounds around the world. The wound treatment process is costly and requires nursing professionals to develop activities during patient care, among them: the correct identification and classification of wound tissues. In this way, this work proposes to build a hybrid computational system of two configurations to support the treatment of wounds. The first configuration uses a mobile application to perform the capture, segmentation and classification of the images of wounds. The other configuration has a client-server architecture, the images are captured and segmented in the application and sent, through the Internet, to the web server, which is responsible for classifying the tissue of the wounds. For this, some studies were carried out in the literature, namely: review of scientific articles, applications similar to the one proposed in this work, methods for assessing the efficiency of computer systems and wound classification algorithms. Thus, this work uses the method of segmentation and classification of wound images developed by Marques et al. \cite{marques2018ulcer} to build the hybrid computational system. An evaluation questionnaire was developed to be applied to nursing specialists at Hospital Universitário Onofre Lopes in order to assess the technical quality of segmentation and classification of wounds. Thus, by means of a \ textit {survey}, we sought to answer the following research question: are the execution time and energy expenditures acceptable by nurses in both configurations of the system?

Brazilian Courts have been working in virtualisation of judicial processes since this century's rise, leading to a revolution in
relations, services and labour force. A huge volume of data has been produced and computational techniques have been an intimate ally to keeping business processes under control and delivering services as juridical clients expect. However,although there is a misunderstanding that automation solutions are always ’intelligent’, which in most cases, it is not true, there has never been any discussion about the use of intelligent solutions for this end as well as any issues related with automatic predicting and decision making using historical data in context. One of the problems that has already come to light is the bias in judicial datasets around the world. Thus, this work will focus on evaluating,applying and understanding resources based on fine and parameter tuning, with the end of better using machine learning techniques when working on judicial systems, and, therefore,raising the discussion related to secondary issues. We have used a real dataset of judicial sentences (Além da Pena), applying supervised and unsupervised learning models and our results point to the accuratedetection of gender bias.

The growth in the number of User Equipment (UE) in the need for the development of new mobile network technologies is needed in order to meet the demand for traffic required. With this, the concept of Mobile Networks of Fifth Generation (5G) emerges with innumerable paradigms and approaches in order to advance on how the current internet service needs. However, mobility management in mobile networks is faced with the need to be resolved, one of which is the efficiency of handover decision mechanisms. Numerous numbers have been documented in the literature in order to obtain the best mobility decisions. Among the various mobility availability algorithms, according to solutions based on Multi-Attribute (Multiple Attribute Decision Making - MADM), they are considered the most robust due to their efficiency with regard to multimedia traffic on mobile networks. Most evaluations of MADM-based methods documented in the literature do not have an analysis of data on the subjective aspects related to user perception/satisfaction with regard to mobility models, which is a fundamental condition for assessing the Quality of Experience (Quality of Experience - QoE) in multimedia applications applied in this scenario. Therefore, this work offers a wide review of MADM methods, the quality-oriented mobility decision process.

The use of Networks-on-Chip (NoCs) in the communication infrastructure of multiprocessor systems (MPSoCs) has become a standard due to its scalability and support for parallel communications. These architectures allow the execution of applications formed by different tasks that communicate with each other, and the support for this communication has a fundamental role in the system's performance. IPNoSys (Integrated Processing NoC System) is an unconventional architecture, with its own execution model, developed to exploit this NoC communication structure as a high-performance processing system. In the scenario of conventional computer networks, there was a convergence towards the use of the Software-Defined Network (SDN) paradigm, and a central component controls the network, which has an overview of the network and is programmable to change the network configuration to adapt to the specifics of the application or the needs of the programmer. Some works propose the use of the SDN paradigm in NoCs in order to create more flexible architectures. Thus, SDNoCs has a simpler communication infrastructure but is connected to a programmable controller that manages the network's functioning. This work aims to present an architecture based on the IPNoSys execution model but using SDN concepts to provide network control. IPNoSys III is an NoC with a 2D mesh topology, which contains a communication unit and four processing cores on each node, with memory access, that executes packets in the IPNoSys format. An SDN controller, connected to all nodes, has an overview and manages the network to execute the routing algorithm and map tasks according to performance objectives. As a proof of concept, we developed a programming and simulation environment for this architecture in SystemC, and the evaluations performed show the operation and benefits obtained through the use of an SDN controller.

Data clustering consists of grouping similar objects according to some characteristic. In literature, there are several clustering algorithms, among which stands out the Fuzzy C-Means (FCM), one of the most discussed algorithms, being used in different applications. Although it is a simple and easy to manipulate clustering method, the FCM requires as its initial parameter the number of clusters. Usually, this information is unknown, beforehand and this becomes a relevant problem in the data cluster analysis process. Moreover, the design of the FCM algorithm strongly depends on the selection of the initial centers of the clusters. In general, the selection of the initial set of centers is random, which may compromise the performance of the FCM and, consequently, of the cluster analysis process. In this context, this work proposes a new methodology to determine the number of clusters and the set of initial centers of the partial algorithms, using the FCM algorithm and some of its variants as a case study. The idea is to use a subset of the original data to define the number of clusters and determine the set of initial centers through a method based on mean type functions. With this new methodology, we intend to reduce the side effects of the clusters definition phase, possibly speeding up the processing time and decreasing the computational cost. To evaluate the proposed methodology, different cluster validation
indices will be used to evaluate the quality of the clusters obtained by the FCM algorithms and some of its variants, when applied to different databases.

The popularity of the Android platform can be attributed to their ability to run apps that leverage the many capabilities of mobile devices. Android applications are mostly written in Java, however, they are very different from standard Java applications, with different abstractions, multiple entry points, and also have a different form of communication between components. These differences in the structure of Android applications have had negative effects on the user experience in terms of low robustness. In terms of robustness, the exception handling mechanism for the Android platform has two main problems: (1) the Terminate ALL" approach and (2) a lack of a holistic view on exceptional behavior. Exception handling is strongly related to program robustness. In addition to robustness, energy consumption and performance are other non-functional requirements that need to be taken into account during development. These three requirements can directly affect the quality of the user experience and the quality of the functioning of the applications. In this context this work proposes a general methodology to efficient engineering of Android applications and an EHM called DroidEH to support the methodology and to improve the robustness of Android applications. Studies have been carried out to understand the impact of exception handling on the robustness and energy consumption of Android applications. The evaluation of the methodology showed that it is satisfactory in achieving the objective of allowing the developer to make decisions taking into account these non-functional requirements and to determine through the trade-o between these requirements, dierent operation modes that can be implemented in the application using the DroidEH. Furthermore, it was observed that the use of DroidEH in applications can enhance its robustness.

The use of dynamic selection techniques, for attributes or members of an ensemble,
has appeared in several works in the literature as a mechanism to increase the accuracy of rating
ensembles. Individually, each of these techniques has already shown the benefits of using it. The
objective of this work is to improve the efficiency of the classifier ensembles through the use of
 dynamic selection techniques for the definition of structure of these systems. With that, it will be
possible to explore the use of these two techniques integrated in the classification of an instance,
making each instance be classified using its own subset of attributes and classifiers. When used in
an integrated manner, due to the use of the two dynamic processes, it is believed that the complete
system has a long execution time. Aiming to overcome this disadvantage in its use, where the
complete dynamic system will be used only in certain instances. Thus, some instances would be
classified using all the dynamic system, while the other instances would be classified using only a single classifier. In other words, some instances may not require a level high complexity of the classification system. For these instances, a classifier will be used. In this way, the dynamic ensemble will only be used in instances considered difficult to classify. Initial results showed that the integration of these two dynamic techniques obtained promising results in terms of accuracy. Finally, these results were not significantly affected by the addition of the decision criterion, which generated a very significant reduction in the total processing time of the system.

In the last few decades, the number of people living in cities has grown exponentially. This scenario imposes several challenges to the management of the city, since the services offered to the population (transportation, security, health, electricity supply, etc.) need to be scaled up quickly to support an increasing number of inhabitants. The realization of the concept of smart cities emerged as a promising solution to face the various challenges resulting from urban growth. Smart city environments are characterized by the presence of a myriad of applications that aim to facilitate city management, contributing to the provision of more efficient services and, consequently, improving the quality of life of citizens. However, developing such applications is not a trivial task. In many cases, developers need to meet several complex requirements to be implemented. In addition, to allow contextualization and correlation of information produced in the city, they need to be enriched with geographical information that represents the urban space. In this sense, smart city platforms play a fundamental role in achieving this environment. Such platforms provide high-level services that can be easily reused by developers to leverage application development. In this perspective, this work presents Smart Geo Layers (SGEOL), a scalable platform for developing applications for smart cities. In addition to allowing the integration of urban data with geographic information, SGEOL offers facilities for: i) management of context data, ii) integration of heterogeneous data, iii) semantic support; iv) data analysis and visualization; v) support for data security and privacy, etc. This work also presents experiences of real use of SGEOL in different scenarios, as well as results of computational experiments that evaluate its performance and scalability.

A fuzzy graph is a fuzzy relation between the elements of a set, they are ideal for modeling uncertain data about these sets. The fuzzy graphs appear frequently in the literature, among them, stands out the fuzzy graph of Rosenfeld, based on fuzzy sets of
Zadeh, and its extensions, such as: interval-valued fuzzy graphs, bi-polar fuzzy graphs and m-polar fuzzy graphs. The applications of these concepts are vast: cluster analysis, pattern classification, database theory, social science, neural networks, decision analysis, among others. As well as fuzzy graphs, studies on admissible orders and their extensions are frequent. Originally, admissible orders were introduced in the context of interval-valued fuzzy sets by H. Bustince et al. and since then they have been widely investigated. Recently, this notion has been studied in other types of fuzzy sets, such as interval-valued intuitionistic fuzzy sets, hesitant fuzzy sets, multidimensional fuzzy sets and n-dimensional fuzzy sets. In this context, this work proposes to extend the fuzzy graph of Rosenfeld to interval-valued n-dimensional fuzzy graphs, based on n-dimensional fuzzy sets, as well as, for the admissible interval-valued n-dimensional fuzzy graphs, that we equip with an admissible ordered semi-vector space. We present some methods to generate admissible orders in the n-dimensional fuzzy set and the concept of n-dimensional aggregation functions with respect to an admissible order. We extend the concept of ordered semi-vector space in a semi-field of non-negative real numbers to an arbitrary weak semi-field. We define in a set of admissible interval n-dimensional fuzzy graphs the concept of ordered semi-vector space, whit this, we introduced in this set the concept of admissible interval n-dimensional fuzzy graphs aggregation function. Several properties of these concepts were investigated, in addition to presenting some applications.

Currently, blockchain has been widely used to store decentralised and secure transactions involving cryptocurrency (for instance, Bitcoin and Ethereum solutions). On the other hand, smart city applications are concerned about how data and services can be safely stored and shared. In this regard, this research investigates the use of blockchain and pinpoints a set of essential requirements to meet the needs of blockchain for the context of smart cities. Base on that, a platform named Orthus was proposed to support the use of blockchain in smart city initiatives focused on scalability. This master’s thesis demonstrated a case study about how to use the proposed platform in the context of the Natal Smart City Initiative, in Brazil, to handle land registration. Moreover, it also compares this platform with other implementations that use blockchain in different domains. Finally, this research confirms that use of blockchain technology has much to contribute to smart city solutions once it enables the creation of solutions in distributed networks being able to meet the demand of the entire population.

The information available in crash reports has been used to understand the root cause of bugs and improve the overall quality of systems. Nonetheless, crash reports often lead to a huge amount of information, being necessary to apply techniques that aim to consolidate the crash report data into groups, according to a set of well-defined criteria. In this dissertation, we contribute with customization of rules that automatically find and group correlated crash reports (according to their stack traces) in the context of large scale web-based systems. We select and adapt some approaches described in the literature about crash report grouping and suspicious file ranking of crashing the system. Next, we design and implement a software tool to identify and rank buggy files using stack traces from crash reports. We use our tool and approach to identify and rank buggy files—that is, files that are most likely to contribute to a crash and thus need a fix.

We evaluate our approach comparing two sets of classes and methods: the classes (methods) that developers changed to fix a bug and the suspected buggy classes (methods) that are present in the stack traces of the correlated crash reports. Our study provides new pieces of evidence of the potential use of crash report groups to correctly indicate buggy classes and methods present in stack traces. For instance, we successfully identify a buggy class with recall varying from 61.4% to 77.3% and precision ranging from 41.4% to 55.5%, considering the top 1, top 3, top 5, and top 10 suspicious buggy files identified and ranked by our approach. The main implication of our approach is that developers can locate and fix the root cause of a crash report considering a few classes or methods, instead of having to review thousands of assets.

The development of Internet of Things (IoT) applications facing important issues such as the inherent device heterogeneity. in terms of capabilities, computing power, network protocols, and energy requirements. To address this challenge, IoT middleware platforms have been proposed to abstract away the specificities of such devices, promoting interoperability among them, and easing application development. One of these proposals is FIWARE, an open, generic platform developed in the European Community to leverage the development of Future Internet applications. Given a set of FIWARE components required for a specific application under development, their deployment and configuration can be made either manually or using a container-based approach. However, setting up an environment composed by the main FIWARE components might sometimes not be a trivial process. This work proposes FIWARE-Lab@RNP, a Web virtual laboratory for prototyping and experimenting applications based on the FIWARE platform. The main concern of FIWARE-Lab@RNP is enabling the use of FIWARE resources through the Internet in a transparent way, thus relieving users from the need of deploying and operating a FIWARE instance on their development or owned environment. The virtual laboratory provides functionalities for easily creating, configuring, and managing instances of FIWARE components, devices, context entities, and services while attempting to minimize the learning curve regarding these tasks.

The use of soft biometrics as an auxiliary tool on user identification is already well known.
It is not, however, the only use possible for biometric data, as such data can be adequate to
get low level information from the user that are not only related to his identity. Gender,
hand-orientation and emotional state are some examples, which it can be called softbiometrics.
It is very common to find work using physiologic modalities for soft-biometric
prediction, but the behavioural data is often neglected. Two possible behavioural modalities
that are not often found in the literature are keystroke dynamics and handwriting
signature, which can be seen used alone to predict the users gender, but not in any kind of
combination scenario. In order to fill this space, this study aims to investigate whether the
combination of those two different biometric modalities can impact the gender prediction
accuracy, and how this combination should be done.

In the Scrum Framework, the Product Owner (PO) takes the central role within the development process, being responsible for communicating between the customer and the developers. In this intermediation, he manages the Product Backlog, which maintains a list of items to be developed, corresponding to the customer's needs. In this sense, the academy has explored the challenges of the PO, mainly in the planning activities where, in this context, the decision making is seen as his most important task. However, the lack of structured information that can support their choices, makes them, many times, make wrong decisions or omit this responsibility. In Goal-Oriented Requirements Engineering, the requirements are described from the stakeholders organizational goals and, according to the literature, their definition can bring several benefits in terms of information organization capacity. Most Scrum projects use user stories to specify requirements and, although they contain the definition of the goal, she is not evidented in the development process. That said, this work aims to provide a presentation of the organizational information, inherent to the desired product or service, in a provision that justifies and guides the decision making of the PO. To this end, an artifact, the Goals Driven Product Backlog, was proposed, which seeks to highlight the goals and their relationships with user stories. The evaluative study carried out found evidence that the artifact provides more structured information to the PO and, consequently, contributes to his decision making.

Speech is a crucial ability for humans to interact and communicate.
Speech-based technologies are becoming more popular with speech interfaces,
real-time translation, and budget healthcare diagnosis.  Thus, this work aims
to explore an important but under-investigated topic on the field: multilingual
speech recognition.  We employed three languages: English, Brazilian
Portuguese, and Mandarin. To the best of our knowledge, those three languages
were not compared yet.  The objectives are to explore Brazilian Portuguese in
comparison with the other two more well-investigated languages, by verifying
speaker recognition robustness in multilingual environments, and further
investigate fuzzy methods. We have performed an analysis for text-independent
speaker identification on closed-set using log-Energy, 13-MFCCs, Deltas, and
Double Deltas with four classifiers.  The closed-set text-independent speaker
identification results indicated that this problem presents some robustness on
multilingual environments, since adding a second language, it degrades the
accuracy by 5.45\%, and 5.32\% for a three language dataset using an SVM
classifier.

Smart cities are composed of several interconnected systems, designed to promote better management of urban and natural resources in cities, thus contributing to improving the quality of life of citizens. Data is of great importance for smart cities, as they significantly contribute to the strategic decision-making process for urban space. However, such a scenario is typically characterized by the high heterogeneity of data sources making the search for significant information more complex. To deal with these characteristics, ontologies have been used in conjunction with Linked Data to semantically represent information, infer new information from existing data and effectively integrate connected information from different sources. This scenario requires a data management strategy that includes efficient mechanisms to support information filtering and knowledge discovery. In this context, this work proposes a search component to semantic data based on the representation of georeferenced information in smart cities through ontologies and linked data. The proposed solution was applied to geo-referenced educational data from a city to infer new non-explicit information from existing data and relationships.

Digital forensics is a branch of computer science that uses computational techniques to analyze criminal evidence with greater speed and accuracy. In the context of the Brazilian justice system, during a criminal investigation, forensic specialists extract, decode, and analyze the evidence collected to allow the prosecutor to make legal demands for a prosecution. These experts have a very short time to analyze to find criminal evidence can take a long time. To solve this problem, this paper proposes ARTEMIS (A micRoservice archiTecturE for imagesin criMe evIdenceS or Microservice Architecture for images in criminal evidence) an architecture for classifying large amounts of image files present in evidence using open source software. The image classification module contains some pre-trained classifiers, considering the need of foren-ses analysts from the MPRN (Rio Grande do Norte Public Ministry). Models were built to identify specific types of objects with for example: firearms, ammunition, Brazilian ID cards, text documents, cell phone screen captures enudez. The results obtained show that the system obtained good precision in most cases. This is extremely important in the context of this research, where false positives should be avoided in order to save analysts' work time. In addition, the proposed architecture was able to accelerate the process of evidence analysis.

Continuous Integration (CI) is the practice of automating and improving the frequency of code integration (e.g., daily builds). CI is often considered one of the key elements involved to support agile software teams. It helps to reduce the risks in software development by automatically building and testing a project codebase, which allows the team to fix broken builds immediately. The adoption of CI can help development teams to assess the quality of software systems. The potential benefits of adopting CI have brought the attention of researchers to study its advantages empirically. Previous research has studied the impact of adopting CI on diverse aspects of software development. Despite the valuable advancements, there are still many assumptions in the community that remains empirically unexplored.

Our work empirically investigates the software quality outcomes and their relationship with the adoption of CI. This thesis provides a systematic literature mapping that presents a broad knowledge of how practitioners and researchers recognize the CI practice to affect software product-related aspects. Additionally, we improve some assumptions by performing two empirical studies that aim to answer the following open questions: (i) Does the adoption of CI share a relationship with the evolution of test code? (ii) The adherence to CI best practices is related to the degree of code quality? Finally, we present a pioneering study that goes beyond the correlation tests to investigate the estimated causal effect of CI adoption and its impact on automated tests. Thereby, we apply a causal inference using directed acyclic graphs and probabilistic methods to determine the causal effect of CI in automated tests. Our results suggest that, despite the CI adoption trade-offs, it is likely to be associated with
improvements in software quality. Additionally, it employs a considerable positive causal effect on the volume of automated tests.

 Aggregations are functions that have the ability to combine multiple objects into a single object of the same nature. Minimum, maximum, weighted average and arithmetic mean, are examples of aggregations frequently used in everyday life which have several possibilities for applications. However, when working with aggregations, such as those mentioned above, the objects in question are always real numbers. There are almost no studies in the literature that portray these aggregations when objects are complex numbers. This is due to the fact that to introduce some aggregations, the objects involved need to be provided with a total order relation. The Graduated Complex Numbers (NCG), proposed by the author, was recently applied in the performance evaluation of classification algorithms. The method required the comparison  of complex graduated numbers to achieve that the notion of local order is proposed and consequently the concept of local aggregation is developed. Two applications of such approach are provided.

This work presents a new variant of the Traveling Car Renter Problem not yet described in the literature, denominated the Traveling Car Renter with Passengers. This problem provides a set of cities, a set of vehicles and a set of passengers. The salesman's tour can be done using dierent vehicles, i.e., the problem encompasses the process of vehicles' rental and delivery. In the proposed model, the variant of the Traveling Car Renters Problem is merged with Ridesharing aspects. That is, in cities there may be passengers interested in traveling to a certain destination and willing to share costs with the salesman while they are aboard on the vehicle. The objective of the problem is to determine, in a graph, the lowest Hamiltonian cycle considering the vehicles' exchanges and  he shipments along the tour. The problem is made up of several interlinked decisions: the sequence of visited cities, the order of used cars, the cities where the cars must be rented and/or delivered and the passengers' boarding schema. A general analysis of the problem is described to justify its complexity. Two formulations of mixed integer programming are proposed. These formulations are linearized using two dierent techniques, resulting in four linear
models. These models are implemented in two solvers and validated based on instances of the problem, which in turn are based on instances of the Traveling Car Renter Problem. In addition, two naive heuristics and a metaheuristic to solve the problem. Comparative computational experiments and performance tests are performed on a set of 54 instances. The results obtained are compared and the conclusions are reported.

The virtualization of network functions is a technology that proposes to decouple network functions, traditionally allocated on specialized hardware, so that making them available as software elements executing at general-purpose servers premises. Such flexibility allows offering network services running over cloud infrastructures and facilitates enforcing network policies based on the chaining of different functions, through which a targeting traffic must be subjected. The variation in services demand will require the resource management attribute of elasticity to tackle performance goals, adjusting the computational resources of the functions to suit both the new projected demand  and operating costs so as to avoid provisioning beyond the need. Traditionally, reactive threshold-based approaches afford elasticity function, to the cost of exponentially increasing response times as resources run out. Recent work suggest proactive elasticity approaches harnessing the combination of machine learning methods that allow anticipating decisions and adapting resources to the projected demand, as much as possible. Such adequacy is crucial for the success of a proactive elasticity solution, in the perspective to enable assertive scaling decisions to respond with agility and precision to variations in demand, as well as to contribute with the balance of cost and performance objectives. This doctoral thesis presents ARENA, a proactive elasticity mechanism for autoscaling virtualized network functions driven by demand prediction based on machine learning to maximize the assertiveness of horizontal and vertical dimensioning decisions.

The aging of the world's population and the growth in the number of people with ch-
ronic diseases have increased expenses with medical care. Thus, the use of technological

solutions, including Internet of Things-based solutions, has been widely adopted in the
medical eld to improve the patients' health. In this context, approaches based on Cloud

Computing have been used to store and process the information generated in these soluti-
ons. However, using Cloud can create delays that are intolerable for medical applications.

Thus, the Fog Computing paradigm emerged as an alternative to overcome this problem,
bringing computation and storage closer to the data sources. However, managing medical
data stored in Fog is still a challenge. Moreover, characteristics of privacy, condentiality,
and interoperability need to be considered in approaches that aim to explore this problem.
So, this work denes a Fog Computing-based software architecture designed to provide

patient-centered management of medical records. This architecture uses Blockchain tech-
nology to provide the necessary privacy features. This thesis also describes a case study

that analyzed the requirements of privacy, condentiality, and interoperability in a Home
Care scenario. Finally, the performance behavior related to access to data managed in the
proposed architecture was analyzed in the mentioned scenario.

As a method of trying to extrapolate the Church’s thesis, using the ideas of the fuzzy sets presented by Zadeh, the fuzzy automaton theory emerges in the late 1960s, as an extension of the finite automata theory, adding the possibility of computing with some level of uncertainty. Over the years, due to the maturation of the extensions of the fuzzy sets, different generalizations of fuzzy automata started to emerge in the literature, such as interval-valued fuzzy automata, intuitionist fuzzy automata, etc. Fuzzy automata, in addition to being the fundamental part of the fuzzy computation theory, also present a relative success in practical applications, mainly in the field of pattern recognition, through uncertainty modeling. This work presents a new generalization of fuzzy automata based on the definitions of typical hesitant fuzzy sets (which we will call typical hesitant fuzzy automata), as well as the motivation for this generalization and for bringing to the domain nof computing the possibility of working with uncertainties and also being able to work with hesitation. Therefore, this new generalization aims to enable new ways to face problems that were not easily modeled before, using only uncertainties. Besides, we will show ways to apply this new type of automata in the fields of digital image processing and data classification.

Computer games has been consolidated as a favourite activity in the last years. Although such games were created to promote
competition and promote self-improvement, there are some recurrent issues. One that has received the least amount of attention so far is the problem of "account sharing" which is when a player shares his/her account with more experienced players in order to progress in the game. The companies running those games tend to punish this behaviour, but this specific case is hard to identify. Since, the popularity of machine learning has never been higher, the aim of this study is to better understand how biometric data from online games behaves, to understand how the choice of character impacts a player and how different algorithms perform when we vary how frequently a sample is collected. The experiments showed through the use of statistic tests how consistent a player can be even when he/she changes characters or roles, what are the impacts of more training samples, how the tested
machine learning algorithms are affected by how often we collect our samples, and how dimensionality reduction techniques, such as Principal Component Analysis affect our data, all providing more information about how this state of art game database works.

Game jams are events where people form teams with the purpose of developing functional game prototypes in a collaborative way and under thematic and time constraints. Although such events have grown in popularity, there is still little evidence in the scientific literature on how collaboration occurs in this type of event, and especially, what are the factors that facilitate such collaboration. This study aims to understand how facilitating aspects support collaboration in game development through jams. To this aim, a study was conducted during the Global Game Jam 2018 and 2019. Our findings describe the main practices during the event and what strategies are used by teams in order to facilitate the collaboration. In addition, mentors and experienced jammers are recognized as the main coordinators of these tasks, acting as mediators between those teams. This work also contributes to understanding and characterizing the collaborative work in jams. With collected and validated evidence from experts, we suggest a catalog of practices that support the interaction between teams in this type of event. These techniques may be applied in similar contexts to facilitate collaboration between teams.

The advances of integrated circuit fabrication technology, allowed by the reduction of transistors size, make possible the creation of multiprocessed complex systems inside a single chip, named Multiprocessors System of Chip (MPSoC). To allow communication among the different cores, one of the main communication model used currently is the Network on Chip (NoC) which shows more scalability than traditional bus solution.  In spite of the expected potential performance improvement due to task parallelism, to achieve a real performance improvement it is necessary an efficient available resource management in the system, such as processing cores and communication links. This management is related, among other aspects, to how tasks are mapped in the MPSoC cores and which NoC channels are used to provide communication routes for tasks. In the literature, these aspects are handled individually, even though, they are highly correlated. Based on this principle, in this work, it is presented a mathematical formulation of the Mapping and Routing Problem (PMR) that combines both problems. Additionally, in order to find optimized mapping solutions, this work presents three proposals of static mapping and one of the dynamic mapping.In the static mapping context, it is presented bioinspired hybrid optimization strategies (Genetic, Memetic and Transgenetic). These strategies present a general approach to find mapping solution and internally, it is used an exact routing fitness function evaluation based on the proposed model.In the dynamic mapping context, it is proposed an algorithm that uses the Transgenetic operator to provide the task allocation by demand at run time. The algorithms were implemented and its results were simulated using a NoC simulation tool. In order to compare to state of art, three algorithms from literature were also implemented and simulated. The results show that approaches able to capture more deeply the features of the architecture are more efficient. More specifically, the Transgenetic algorithm presents the best results for latency and energy. Furthermore, it was possible to use the Transgenetic aspects to propose a dynamic solution that can be used when the system does not know the application behavior.

The Beam Angle Optimization (BAO) and the Fluency Map Optimization (FMO) are two problems that arise from the planning of radiation therapy for cancer treatments. One of the main modes of radiation therapy is the IMRT (Intensity Modulated Radiation Therapy) which consists in using computer-controlled linear accelerators to deliver precise radiation doses to the tumor. The IMRT aims at finding a balance between exposing the region of the tumor and, at the same time, preventing, as much as possible, healthy tissues that surround the tumor to receive radiation. This study presents linear and quadratic programming models  for the FMO and  single and multiobjective algorithms that use those models for the BAO problem.  This study also reports the results of computational experiments on a set of real instances.

The design of efficient elasticity control mechanisms for dynamic resource allocation is crucial to increase the efficiency of future cloud-network sliced networks. Current elasticity control mechanisms proposed for cloud or network-slice environments only consider cloud or network resources. Cloud-network sliced networks will obtain substantial gains from these mechanisms only if they consider both cloud and network resources in an integrated fashion. Moreover, they must enable a fast enough orchestration of these resources but preserving the performance and isolation of the cloud-network slices. In this work, we propose a elasticity control approach to dynamically control and manage the orchestration of resources in cloud-network slice-defined systems. A prototype was developed, and its preliminary results suggest that the proposing approach is a viable solution to provide elasticity in environments defined by cloud-network slices.

With the rising popularity of IoT, several platforms have been proposed for supporting the development of IoT applications. KNoT and FIWARE are examples of open-source platforms with a complementary purpose. While KNoT is a gateway-based middleware to embed connectivity into devices and route messages between them and applications, FIWARE provides a rich ecosystem with standardized APIs for developing IoT applications. Aiming at combining the KNoT capability of integrating a plethora of devices with the high-level abstractions provided by the FIWARE platform, this work presents the KNoT-FI environment. It integrates KNoT and FIWARE towards easing the development of IoT applications and seamlessly using capabilities of devices with or without native Internet connection through the FIWARE advanced interfaces. This work also presents a validation of KNoT-FI in the development of a real-world application that automatically manages lighting, temperature, and ambient sound in smart buildings.

Smarter cities are largely adopting data infrastructure and analysis to improve decision making for public safety issues. Although traditional hotspot policing methods have shown benefits in reducing crime, previous studies suggest that the adoption of predictive techniques can produce more accurate estimates for future crime concentration. In this work we propose a framework to generate future hotspots using spatiotemporal features and other geographic information from OpenStreetMap. We implemented an open source Python-package called predspot to support efficient hotspots prediction following the steps suggested in the framework. To evaluate the predictive approach against the traditional methodology implemented by Natal’s police department, we compared two crime mapping methods (KGrid and KDE) and two efficient machine learning algorithms (Random Forest and Gradient Boosting) in twelve crime scenarios, considering burglary, violent and drugs crimes. The results indicate that our predictive approach estimate hotspots 1.6-5.1 times better than the analysts baseline. A feature importance analysis were extracted from the models to account with how much the selected variables helped the predictions and to discuss the modelling strategy we conducted.

Time-bounded collaborative events, such as hackathons, game jams, among others, have become quite popular in recent years and have been drawing the attention of the scientific community. Existing research has studied various aspects of these events, while the post-event, i.e. what occurs after these events, has received little attention. In addition, existing studies are very limited to events of the same nature, such as civic hackathons, industrial or academic, as well as for game jams, where the existing works contemplate only one type of such events. Research on the phenomena surrounding the post-event, on a wider granularity, is still scarce. In this study, we address this gap by presenting the results of an exploratory study, featuring analyses of two large hackathons and a global game jam. Most participants in these events expressed their intention to continue with the social relations that were developed in them. However, they indicated that there is no planning to continue the development of the projects developed, although there is interest in working on other projects with the team formed at the event. The main objective is to validate and deepen these questions, bringing contributions to the positive and negative aspects of the discontinuity of these projects and the maintenance of the social bonds formed during the events.

The tourism sector is one of the most relevant economic activity in nowdays. In this way, it is important invest in different approaches to create a great experience during visitors trips in one destination. In a context of Smart Cities, the ideia of Smart Destination appears as one solution to improve the tourism experience using techonlogy to support visitors in one Smart City. The proposed study creates an approach to support a Smart Tourism Destination to create a better trip planning based on photos from social medias. The research aims to create recommendation to single or group of tourists using techniques of image classification and fuzzy inference to map tourists preferences. Through the fuzzy inference system and using the tourism experts knowledge inside a recommendation system, the proposed approach is able to create personalized recommendations using attractions from one Smart Destination

The realization of innovative use cases involving data communication has challenged the established communication networking approaches. This thesis aims to evolve current WLAN Wi-Fi systems with shared access facilities to meet the challenges of efficiently increasing the massive demand for mobile data in 5G use cases, namely UDN (Ultra-Dense Networking) and URLLC (Ultra-reliable and low-latency Communications). New complementary aspects of emerging technologies in 5G systems, namely, virtualization of network functions [7], computational fog [8], software-defined networking, and others, were investigated in the prospect to create a unique and innovative framework for Wi-Fi WLAN-sharing systems. This thesis proposes the WISE framework (WLAN slIcing SErvice), which applies network slicing technique for the first time in WLAN-sharing enabled Wi-Fi CPEs (Consumer Premise Equipments) located closer to users and things, in perspective to offer differentiated network services featuring isolation, independence, and increased performance. Moreover, the WISE framework exploits fog computing technology as a way to expand the computational capabilities of off-the-shelf CPEs to afford running part of the functionalities featuring the target framework.

In machine learning, the data preprocessing has the aim to improve the data quality, through to analyze and to identify of problems in it. So, the machine learning technique will receive the data of a good quality. The feature selection is one of the most important pre-processing phases. Its main aim is to choose the best subset that represents the dataset, aiming to reduce the dimensionality and to increase the classifier performance. There are different features selection approaches, on of them is the Dynamic Feature Selection. The Dynamic Feature Selection selects the best subset of attributes for each instance, instead of only one subset for a full dataset. After to select a more compact data representation, the next step in the classification is to choose the model to classify the data. This model can be composed by a single classifier or by a system with multiples classifiers, known as Ensembles classifier. These systems to combine the output to obtain a final answer for the system. For these systems to get better performance than a single classifier it is necessary to promote diversity between the components of the system. So, it is necessary that the base classifiers do not make mistakes for the same patterns. For this, the diversity is considered one of the most important aspects to use ensembles. The aim of the work is to use the Dynamic Feature Selection in Ensembles systems. To this,  three versions were developed to adapt this feature selection and to create diversity between the classifiers of the ensemble. The versions were compared using different selection rates in an ensemble with five classifiers. After this, the best version was tested with different ensemble sizes.

On Algebras for Interval-Valued Fuzzy Logic

Machine learning is a field of artficial inteligence that is dedicated to the study and
development of computational techniques which obtain knowledge through acumulated
experiences. According to the nature of information provided, machine learning was inicially
divided into two types: supervised and unsupervised learning. In supervised learning,
the data used in training have labels, while in the unsupervised learning the instances
to be trained have no labels. Over the years the academic community started studying
the third type of learning that is regarded as the middle ground between supervised and
unsupervised learning and is known as semi-supervised learning. In this type of learning,
most of training set labels are unknown, but there is a small part of data that has known
labels.The semi-supervised learning is attractive because of its potential to use labeled
and unlabelled data to achieve better performance than supervised learning. This paper
consists of a study in the field of semi-supervised learning and implements changes on
the self-training and co-training algorithms. In the literature, it´s common to develop
researches that change the structure of these algorithms, however, none of them propose
some variation in the rate of inclusion of new instances in the labeled data set, which is
the main purpose of this work. In order to achieve this goal, three methods are proposed:
FlexCon-G, FlexCon e FlexCon-C. The main diference between this methods is: 1) In the
way that they perform the calculation of a new value for the minimum confidence rate
to include new patterns and 2) The strategy used to choose a label of each instance. In
order to evaluate the proposed methods, we will performed experimentations on 30 datasets
with diversified characteristics. The obtained results indicate that the three proposed
methods perform better than original self-training and co-training methods in most cases.

Nowadays, a variety of mobile devices are available and accessible to the general population, making it an indispensable item for communication and use of various services. In this same direction, these devices have become quite useful in several areas of expertise, including the medical field. With the integration of these devices and applications, it is possible to perform preventive work, helping to combat outbreaks and even prevent epidemics. According to the World Health Organization (2017), malaria is one of the most lethal infectious diseases in the world, mainly in the region of sub-Saharan Africa, while in Brazil it is more frequent the occurrence of cases in the Amazon region. For the diagnosis of malaria it is essential to have trained and experienced technicians to identify the species and phases of the disease, a crucial part to define the ideal dosages of administering medication to patients. In this work, we propose a low-cost malaria diagnosis system using mobile devices, where some segmentation, digital image processing, and convolutional neural networks techniques are applied to perform cell counting, parasitemia estimation, and Plasmodium parasite classification in the species P.falciparum and P.vivax in the trophozoite step. A prototype with 3D parts and electronic automation was proposed to perform the scanning and imaging of blood slides to integrate with the mobile system and perform the on-site diagnosis, without the need for changing microscopic equipment, thus, based on the premise of low cost. A 93% accuracy was obtained in a convolutional neural network train model. In view of this, it is possible to break barriers of accessibility in countries with few resources in the use of diagnostic tools and screening of diseases.

Since arising fuzzy set theory many extensions proposals have been given, one of them is the n-dimensional fuzzy set in which its elements are tuples of size n whose components are values in [0, 1], ordered in increasing form, called n-dimensional intervals. Generally, these sets are used to develop tools that aid in modeling situations involving decision-making where given a problem and an alternative, each n-dimensional interval represents the opinion of n specialists on the degree to which an alternative meets a given criterion or attribute for this problem. However, this approach is not able to deal with situations in which a particular expert can, for example, refrain from any decision-making criteria, and therefore, we would have in the same problem coexisting n-dimensional intervals with different values of n or where the set of specialists changes for each pair alternative/attribute. Thus, we need a new fuzzy set extension in which its elements (intervals) can have any dimensions. In this work, we present the concept of multidimensional fuzzy sets as a generalization of the n-dimensional fuzzy sets in which the elements can have different dimensions. We also present a way to generate comparisons (ordering) of these elements of different dimensions, discuss conditions under which these sets have lattice structure and introduce the concepts of admissible orders, multidimensional aggregation functions and fuzzy negations on multidimensional fuzzy sets. In addition, we deepen studies on ordinal sums of fuzzy negations.

The Traveling Car Renter Salesman Problem, or simply Traveling Car Renter Problem (CaRS), is a generalization of the Traveling Salesman Problem (TSP) where the tour can be decomposed into contiguous paths that are traveled by different rented cars. The objective is to construct a minimal cost Hamiltonian circuit, considering the penalty paid for changing cars in the tour.  This penalty is the cost of returning a car to the city where it was rented. CaRS is classified as an NP-hard problem. This work studies the CaRS version classified as: complete, total, unrestricted, with no repetition, free and symmetric. This research  is focused on hybrid procedures that combine metaheuristics and methods based on Linear Programming (LP). The following methods were investigated: scientific algorithms (ScA), evolutionary algorithms (EA), variable neighborhood descent (VND), adaptive local search (ASLP) and a new variant of ALSP called iterated adaptive local search (IALSP). The following techniques are proposed to deal with CaRS:  ScA+ALSP, EA+IALSP, ScA+IALSP and ScA+VND+IALSP. A mixed integer programming model is proposed for CaRS which was used in the ALSP and IALSP. Non-parametric tests were used to compare the algorithms within a set of instances from the literature.

Open Data initiative has been gaining strength in recent times, with increasing participation of public institutions. However, there are still many challenges that need to be overcome when deciding to open data. This negatively affects the quality and effectiveness of publications. Therefore, the objective of this work is to establish a process that help brazilian public institutions to open their data, systematizing the necessary tasks and phases. For this, we carried out a systematic mapping of the literature, in order to discover strategies, best practices, challenges and difficulties that exist in the field.

The internet has enabled the creation of an immense global data space, that can be accessed in the form of web pages.
However, web pages are ideal for presenting content to human beings, but not to be interpreted by machines.
In addition, it becomes difficult to relate the information stored in the databases behind these pages.
From this came the Linked Data, a set of good practices for relating and publishing data data.

The standard format recommended by Linked Data for storing and publishing related data is RDF.
This format uses triples in the form (subject, predicate, object) to stabilish relationships between the data.
A triplestore can be easily visualized as a graph, so queries are made by defining paths in the graph.
SPARQL, the standard query language for RDF graphs, supports the definition of paths using regular expressions.
However, regular expressions have reduced expressiveness, insufficient for some desirable queries.
In order to overcome this problem, some studies have proposed the use of context-free grammars to define the paths.

We present an algorithm for evaluating context-free path queries in graphs inspired by top-down parsing techniques.
Given a graph and a query defined over a context-free grammar, our algorithm identifies pairs of vertices linked by paths that form words of the language generated by the grammar.
We show that our algorithm is correct and demonstrate other important properties of it.
It presents cubic worst-case runtime complexity in terms of the number of vertices in the graph.
We implemented the proposed algorithm and evaluated its performance with RDF databases and synthetic graphs to confirm its efficiency.

Currently, the cities depend considerable on information systems. A large part of these systems, regardless of the form of public or private management, was developed using technologies and concepts that are now considered outdated. Moreover, because they were not designed to communicate with other systems in an interoperable way, many of these systems in the cities are isolated and non-standardized solutions. In contrast, the dynamism demanded by companies, government and, mainly, population presupposes the union of these systems, working in an integrated and interoperable way. This interoperability is critical to achieving the efficiency and effectiveness of the use of expected resources in an smart city. Furthermore, the union between these systems can bring previously unimaginable results, when compared to the results acquired by each isolated system. These characteristics refer to the concept of System of Systems, which is a set of complex, independent, heterogeneous systems that have their own purposes and collaborate with others to achieve common goals. The interaction between different systems made possible by a SoS is more than the sum of the systems involved, since it allows a SoS to offer new functionalities that are not provided by any of the systems operating alone. Based on the above mentioned characteristics, this paper proposes Mandala, a SoS-centric middleware that enables interoperability between information systems in smart cities. The goal is to make the heterogeneity of the systems involved transparent, providing an environment of integration and interoperation of information systems.

The Multi-objective Spanning Tree Problem (MSTP) is an NP-hard extension of the Minimum Spanning Tree (MST). Once the MTSP models several real-world problems in which conflicting objectives need to be optimized simultaneously, it has been extensively studied in the literature and several exact and heuristic algorithms were proposed for it. Besides, over the last years, researchs have showed the considerable performance of algorithms that combine various metaheuristic strategies. They are called hybrid algorithms and previous works successfully applied them to several optimization problems. In this work, five new hybrid algorithms are proposed for two versions of the MSTP: three for the bi-objective version (BiST) based on Pareto dominance and two for the many-objective version based on the ordered weighted average operator (OWA-ST). This research hybridized elements from various metaheuristics. Computational experiments investigated the potential of the new algorithms concerning computational time and solution quality. The results were compared to the state-of-the-art.

Educational digital games has been an area of research that has increased with the years throughout the international scene. This fact has been due to the potential that the games have of fun, immersion and stimulation to the learning of natural and personalized way. Thus, one of the great challenges found in this area is the creation of games that meet the contents proposed to be taught or practiced. In this sense, the procedural content generation has emerged as an area that can assist the development of educational games. The procedural content generation (PCG) deals with the automatic creation of contents such as textures, sounds, objects and, in the context of this work, the level’s generation. Thus, PCG contributes to the creation of new levels without the need for human intervention. With this, the research seeks to make use of PCG technique to create levels of educational games that require the player to achieve certain pedagogical goals throughout the game. For this, we propose a generation approach in three stages: (i) generation of the basic structure of the level (ex: only floors and walls); (ii) The generation of elements related to the pedagogical objectives of the level; (iii) complete the remainder of the level with enemies and other scenario elements. In this way, it can be used to create different challenges and scenarios so that a student can practice certain content, since whenever a challenge is completed, a new challenge can be generated for the student. In this way, this approach will be investigated using the grammar-based PCG technique. Therefore, we seek to verify if the technique in conjunction with the proposed approach assists in the generation of contents and creates them effectively, evaluating their quality and functionalities with elementary students.

With the advancement of hydrocarbon exploration, the oil industry has been searching for ways to minimize exploratory risks, with one of these ways being the improvement of the used tools. There are three steps in this exploration: the seismic data acquisition, the seismic processing and seismic interpretation. This work is part of the seismic processing, more specifically of one of its stages, the seismic velocity analysis, which aims to find the seismic velocity field that offers reliable earth subsurface models through known algorithms of velocity analysis. One of the objectives of this work is the creation of tools to facilitate this velocity analysis by implementing these algorithms so that they work integrated in a single platform of analysis. Another point that this advance brought, was the considerable increase in the volume of seismic data acquired, which led to an increasing need of computer processing power. Given this need, we present a methodology for velocity analysis using GPUs and its results, showing the viability of using it to accelerate Geophysics algorithms, particularly algorithms for velocity analysis. Finally, case studies will be presented, showing the performance results of the algorithms in CPU and GPU versions.

Software traceability, known for the relationship between any software engineering artifacts, brings great advantages to the development process. The information it provides helps in decision making in the face of a change, better understanding of the artifact, reusability, maintenance, forecasting of costs and deadlines, among others. With environments increasingly adept at agile methodologies, with client side-by-side giving constant feedbacks, a adequacy to these requested changes has been a common practice during system development. And, in order for changes to be made safely, traceability information helps in making decisions, with a goal in which the change does not bring inconsistencies, introduce errors, and generate system failures.
Some project management tools support traceability elements. However, with several data that can be provided with such a practice, it is difficult to interpret them, especially when they are presented only textually. Knowing that the visualization of information brings the possibility of an analysis with large volumes of data in a fast and clear way, offering a safer decision making and, allowing to discover information previously unseen, it is possible to verify in the literature techniques of visualization of information of traceability. However, such techniques require information in addition to these data, need to consider the pillars of information exposed in the academy (problematic, what, when and who to view) to have an adequate visualization.
With this purpose, this work performs interviews in the industry to respond to the pillars of information considered in the proposal of a visualization. Then, an analysis based on Grounded Theory is done on the data collected. Then, in the assembled context of traceability, defined profiles, needs and problems described, and artifacts generated in agile environments, the existing information visualizations are studied in the bibliography.
As a result, a discussion and suggestion of appropriate visualization for traceability information are made based on the suggestions in the literature and data collected from the interview. Later, with Heuristics created, an evaluation of the project management tools that integrate the platform of hosting and versioning of data Github is made, to see if they provide the noticed visualization of traceability information.

Recent years have been marked by a growing advance in embedded computing, sensoring

technologies and connected devices. Such an advance has had a signicant and expressive

impact on innovative paradigms such as the Internet of Things (IoT), which believes that

intelligent objects capable of connecting in the network can cooperate among each other

to achieve a common goal. Such growth has leveraged supplier initiatives to produce

protocols and communication standards that would enable such cooperation, however,

the considerable diversity of devices and consequently protocols that have emerged have

made this process dicult, creating numerous challenges, including heterogeneity and

interoperability. These challenges have made the IoT application development process a

complex and costly task, since the capabilities of these protocols and standards aimed at

discovering the devices on the network, communication among them, have become quite

specic for each device, forcing the developer to create complex integration strategies to

deal with this limitation. In this way, this work proposes a textit framework that will seek

to simplify the process of development of IoT applications through device virtualization, so

that heterogeneous aspects connected to devices will be abstracted by this virtualization,

and common operations of protocols such as discovery of devices and communication with

them will be abstracted through a common interface between them, integrating them and

reducing the impacts of the heterogeneous characteristics.

This paper seeks to model and examine solutions to the Profitable Tour Problem with Passengers and Time Constraints (PTP-TR). The work proposes a mathematical model for the problem, exact solution algorithm and metaheuristics for the solution approximation. In order to operationalize the computational experiment necessary to the present research and because it is a model not described in the literature, test instances were also created. The work performs a computational experiment to evaluate the performance of mathematical modeling and delineate the ability to approximate metaheuristic algorithms for the problem. Finally, it describes the schedule for the masters defense and how the problem can be developed in future works.

In the literature, many problems are pointed out regarding the process of Requirements Engineering. Recent research shows that software development environments face many challenges ranging from requirement elicitation to validation. The challenges listed in the literature are part of topics within the academy in the lecture of Requirements Engineering. Those challenges impact on product quality and may compromise the continuity of a project. Therefore, we believe that maybe there is a deficit in the teaching of the lecture that impacts on the industry, besides a possible lack of parallelism in both contexts. Concerning that scenario, this work lists methodologies and activities that change the traditional method of teaching related to Requirements Engineering. The activities focus on interpreting solutions and writing requirements. For that, it was necessary to perform a systematic review of the literature in order to identify how the lecture is taught. Besides that, we did a survey directed to professors and industry aiming to identify the state of the lecture and difficulties within the area in the country. It was verified that professors and industry face many challenges. The industry challenges may be a consequence of academy teaching. It is necessary to get to know the challenges before they impact on the job market, which means that they need to be identified already in the academy.  From the results that we got, it was concluded that, indeed, it is essential to overcome the challenges presented still in the academy. There is Also a need for more practical activities and new approaches in the classroom. On the other hand, in the industry, we recommend that they collaborate with the academy. In this way, once the industry demands are identified, the academy can provide, for the future professionals, a formation based on expected skills.

One of the main problems of machine learning algorithms is the dimensionality problem. With the rapid growth of complex data in real-world scenarios, attribute selection becomes a mandatory pre-processing step in any application to reduce data complexity and computational time. Based on this, several works were produced to develop efficient methods to accomplish this task. Most attribute selection methods select the best attributes based on some specific criteria. In addition, recent studies have successfully constructed models to select attributes considering the particularities of the data, assuming that similar samples should be treated separately. Although some progress has been made, a poor choice of a single algorithm or criterion to assess the importance of attributes, and the arbitrary choice of attribute numbers made by the user can lead to poor analysis. In order to overcome some of these issues, this paper presents the development of some two strands of automated attribute selection approaches. The first are fusion methods of multiple attribute selection algorithms, which use ranking-based strategies and classifier committees to combine attribute selection algorithms in terms of data (Data Fusion) and decision (Fusion Decision) algorithms, allowing researchers to consider different perspectives in the attribute selection step. The second method (PF-DFS) brings an improvement of a dynamic selection algorithm (DFS) using the idea of Pareto frontier multiobjective optimization, which allows us to consider different perspectives of the relevance of the attributes and to automatically define the number of attributes to select . The proposed approaches were tested using more than 15 actual and artificial databases and the results showed that when compared to individual selection methods such as the original DFS itself, the performance of one of the proposed methods is notably higher. In fact, the results are promising since the proposed approaches have also achieved superior performance when compared to established dimensionality reduction methods, and by using the original data sets, showing that the reduction of noisy and / or redundant attributes may have a positive effect on the performance of classification tasks.

The emergence of new digital technologies come with challenging technical and business requirements. The traditional approach to provide computational infrastructure to application workloads, which relies on in-house management of hardware, does not present technical and cost-effective attributes to deliver high-performance, reliability and scalability. As the biggest technologic paradigm shift in the history of humanity, cloud computing allows diverse deployment and service model alternatives, suitable to diverse requirements, such as security, latency, computational performance, availability and cost. Therefore, numerous companies distribute thousands of clouds worldwide, creating an equitable market through competition, where players create unique features to differentiate from competitors. Consequently, in the consumer side, picking a vendor tipically translates into vendor lock-in, a situation where the applications heavily depend on the vendor’s approach of exposing features, making it difficult to switch between vendors whenever convenient or to support complex scenarios across multiple distributed heterogeneous clouds, such as federation. An immediate work-around for users is to pick cloud solutions that implement standards or post-facto open source platforms, such as OpenStack, which are assumed to provide native interoperability between installations. In the industry, however, OpenStack proves that the lack of interoperability is a real concern even between its deployments, due the high flexibility and complexity of supported use cases. Therefore, this investigation documents intra-platform interoperability, as in OpenStack, presenting in detail the Python client library created by the community to abstract deployment differences, counting with numerous and significant contributions from the author. Afterwards, an extensive validation of that library is performed across one testing and five production clouds from different vendors worldwide, because despite the fact the library is extensively used by the community, it had never been formally validated. The validation unveiled bugs, functionality and documentation gaps. Since the OpenStack intra-platform interoperability had never been documented in the literature, a systematic literature review followed, allowing a deep comparison of the state of the art of vendor lock-in taxonomy and approaches in opposition to that library, presenting its advantages, disadvantages and recommendations for users. Lastly, the suggestions for future work include support for multiple programming languages and the adoption of the client library as a standard for inter-platform interoperability.

The use of Automatic Code Generators for Formal Methods not only minimizes efforts on the implementation of Software Systems, as also reduces the chance of existing errors on the execution of such Systems. These tools, however, can themselves have faults on their source codes that causes errors on the generation of Software Systems, and thus verification of such tools is encouraged. This PhD thesis aims at creating and developing a strategy to verify JCircus, an automatic code generator from a large subset of Circus to Java. The interest in Circus comes from the fact that it allows the specification of concurrent and state-rich aspects of a System in a straightforward manner. The strategy of verification consists on the following steps: (1) extension of the existing operational semantics to Circus and proof that it is sound with respect to the existing denotational semantics of circus in the Unifying Theories of Programming (UTP), a framework that allows proof and unification of different theories; (2) development and implementation of a strategy that refinement-checks the generated code by JCircus, through a toolchain that encompasses a Labelled Predicate Transition System (LPTS) Generator for Circus and a Model Generator that inputs this LPTS and generates an Oracle that uses the Java Pathfinder code model-checker that refinement-checks the generated code by JCircus. Combined with coverage-based testing techniques, we envisage improving the reliability of the Code Generation from Circus to Java.

In the literature it is quite common to find problems that need efficient mechanisms in accomplishing 
the task of combining entries of the same nature in a value of the same type as the inputs. The
aggregation functions are quite efficient in the accomplishment of this work, being able to be used, 
for example, to model the connectives of the fuzzy logic and also in problems of decision making. An
important family of aggregations, belonging to the middle class of functions, was introduced by Yager 
in 1988, who called them ordered weighted averaging functions (OWA). These functions are a kind of
weighted average, whose weights are not associated with the particular inputs, but their respective
magnitudes, that is, the importance of an input is determined by their value. More recently, it has 
been found that non-aggregate class functions may also be able to combine inputs, such as pre-
aggregations and mixture functions, which may not satisfy the mandatory monotonicity condition for
aggregation functions. Thus, the objective of this work is to present a detailed study on aggregations
and preaggregations, in order to provide a good theoretical basis in an area that has a wide possibility
of applications. We present a detailed study of generalized mixing functions - GM, which extend the
Yager OWA functions, and propose some ways to generalize the GM functions: limited generalized mixing
functions and dynamic ordered weighted averaging functions.

Several researches have pointed the hybridization of metaheuristics as an eective way to deal with combinatorial optimization problems. Hybridization allows the combination of dierent techniques, exploiting the strengths and compensating the weakness of each of them. MAHM is a promising adaptive framework for hybridization of metaheuristics, originally designed for single objective problems. This framework is based on the concepts of Multiagent Systems and Particle Swarm Optimization. In this study we propose an extension of MAHM to the bi-objective scenario. The proposed framework is called BOMAHM. To adapt MAHM to the bi-objective context, we redene some concepts such as particle position and velocity. In this study the proposed framework is applied to the biobjective Symmetric Travelling Salesman Problem. Four methods are hybridized: PAES, GRASP, NSGA2 and Anytime-PLS. Experiments with 11 bi-objective instances were performed and the results show that BO-MAHM is able to provide better non-dominated sets in comparison to the ones obtained by algorithms existing in literature as well as hybridized versions of those algorithms proposed in this work.

Pattern classification techniques can be considered the most important activitie in pattern
recognition area where aims assing a unknown sample test to a class. Generally individual
classifiers haven’t good recognition rates compared to multiple classifiers. Thus ensemble
of classifiers can be used to increase the accuracy of classification systems. Ensemble
systems provide good recognition rates when the classifiers members of the ensemble
system have uncorrelated errors in different sub-spaces of the problem; This characteristic
is measured by diversity measures. In this context, the present thesis explores ensemble
systems using dynamic selection. Ao contrário de comitês que utilizam seleção estática,
em comitês de classificadores utilizando seleção dinâmica, para cada padrão de teste
estima-se o nível de competência de cada classificador de um conjunto inicial. Apenas
os classificadores mais competentes são selecionados para classificar o padrão de teste.
O presente trabalho objetiva explorar, avaliar e propor métodos para seleção dinâmica
de classificadores baseando-se em medidas de diversidade. Unlike emseble sysetm using
static selection, in ensembles using dynamic selection for each test pattern is estimated
the competence level for the initial set of classifiers. Only the most relevant classifiers are
selected to classify the test pattern. This paper aims to explorer, evaluate and propose
methods for ensemble systems based on diversity measures. To achieve this goal, several
ensemble systems in the literature using dynamic selection are exploited, as well as hybrid
versions of them are proposed in order to quantify, by experiments, the influence of diversity
measure among classifiers members in ensemble systems. Therefore the contribution of this
work is empirically elucidate the advantages and disadvantages of using diversity measures
in dynamic selection of classifiers.

With the increase of the size on the data sets used in classication systems, selecting
the most relevant attribute has become one of the main tasks in pre-processing phase.
In a dataset, it is expected that all attributes are relevant. However, this is not always
veried. Selecting a set of attributes of more relevance aids decreasing the size of the data
without aecting the performance, or even increase it, this way achieving better results
when used in the data classication. The existing features selection methods elect the
best attributes in the data base as a whole, without considering the particularities of
each instance. The Unsupervised-based Feature Selection, proposed method, selects the
relevant attributes for each instance individually, using clustering algorithms to group
them accordingly with their similarities. This work performs an experimental analysis
of dierent clustering techniques applied to this new feature selection approach. The
clustering algorithms k-Means, DBSCAN and Expectation-Maximization (EM) were used
as selection method. Analyzes are performed to verify which of these clustering algorithms
best ts to this new Feature Selection approach. Thus, the contribution of this study is to
present a new approach for attribute selection, through a Semidynamic and a Dynamic
version, and determine which of the clustering methods performs better selection and get
a better performance in the construction of more accurate classifiers.

The Traveling Salesman with Passengers and High Occupancy Problem is a version of the classic TSP where the salesman is the driver of a vehicle who shares travels’ expenses with passengers. Besides shared expenses, the driver also benefits from discounts of the high-occupancy vehicle lanes, i.e. traffic lanes in which high occupancy vehicles are exempted from tolls. This  paper presents a succinct mathematical model for this problem and an algorithm based on Simulated Annealing and Variable Neighborhood Search metaheuristics. The results of the heuristic algorithm are compared with the optimal solutions obtained by an exact algorithm.This work addresses the study of this novel combinatorial optimization problem, going from the relationship it draws with other ones widely covered by the literature, by the means of a revision of related work, until the conception of artificial test cases to fulfill the purpose of serving as comparison subjects to the experimental algorithms developed to solve it.

As our society becomes more and more dependent of software systems the demand robustness requirements increases. The exception handling mechanism is one of the most used techniques to enable the development of robust software systems develop. The exception handling policy comprises the set of rules that specify how exceptions should be thrown and handled inside a system. But usually the policy is not explicitly defined. As a consequence, it becomes a challenge for developers to create the exception handling code according to it. This work proposes an Eclipse plug-in, called ExceptionPolicyExpert, to guide the developer on how to implement this kind of code by checking policy violations and providing recommendations to developers concerning how to exceptions should be handled and signaled. In order to support the creation of such tool, we performed a qualitative study, using Grounded Theory techniques, to understand which are the main challenges that the developers have during the implementation of the exception handling code. This study showed that most of the developers did not receive any instructions regarding the exception handling policy and they often handle exceptions in a wrong way. Therefore, the proposed tool aims to provide information to developer regarding the exception handling policy integrated to the IDE - helping him/her to develop exception handling code and preventing policy violations. The tool evaluation showed that the tool helps the developer to make decisions when implementing the exception handling code.

The Internet of Things (IoT) paradigm encompasses a hardware and software infrastructure that connects physical devices, known as things, to the digital world. It is estimated that in 2020 there will be nearly 100 billion of connected IoT devices whose data and services will be used for building a myriad of applications. However, developing applications in the IoT context is not a trivial task. Given the large number and variety of devices involved, these applications need to be built and executed in a scalable way to support a large number of connected devices as well as to store and process the huge amount of data they produce. Additionally, applications in this context also need to deal with several different protocols. In this context, middleware platforms have emerged as promising solutions to facilitate application development. These platforms offer standardized interfaces for access to devices, abstracting the developers of details of communication via network, protocols and data formats used by various devices. In this perspective, this work presents the EcoCIT platform, a scalable middleware platform that provides support for the integration of IoT devices to the Internet, as well as the development and execution of IoT applications with scalability requirements through the use of scalable technologies and on-demand computing services provided by cloud computing platforms.