<p>With the increasing availability of data and the rise of networked systems such as autonomous vehicles, drones, and smart girds, the application of data-driven, machine learning methods with multi-agents systems have become an important topic. In particular, reinforcement learning has gained a lot of popularity due to its similarities with optimal control, with the potential of allowing us to develop optimal control systems using only observed data and without the need for a model of a system's state dynamics. In this thesis work, we explore the application of reinforcement learning with multi-agents systems, which is known as multi-agent reinforcement learning (MARL). We have developed algorithms that address some challenges in the cooperative setting of MARL. We have also done work on better understanding the convergence guarantees of some known multi-agent reinforcement learning algorithms, which combine reinforcement learning with distributed consensus methods. And, with the aim of making MARL better suited to real-world problems, we have also developed algorithms to address some practical challenges with MARL and we have applied MARL on a real-world problem.</p> <p>In the first part of this thesis, we focus on developing algorithms to address some open problems in MARL. One of these challenges is learning with output feedback, which is known as partial observability in the reinforcement learning literature. One of the main assumptions of reinforcement learning in the singles agent case is that the agent can fully observe the state of the plant it is controlling (we note the “plant" is often referred to as the “environment" in the reinforcement learning literature. We will use these terms interchangeably). In the single agent case this assumption can be reasonable since it only requires one agent to fully observe its environment. In the multi-agent setting, however, this assumption would require all agents to fully observe the state and furthermore since each agent could affect the plant (or environment) with its actions, the assumption would also require that agent's know the actions of other agents. We have also developed algorithms to address practical issues that may arise when applying reinforcement learning (RL) or MARL on large-scale real-world systems. One such algorithm is a distributed reinforcement learning algorithm that allows us to learn in cases where the states and actions are both continuous and of large dimensionality, which is the case for many real-world applications. Without the ability to handle continuous states and actions, many algorithms require discretization, which with high dimensional systems can become impractical. We have also developed a distributed reinforcement learning algorithm that addresses data scalability of RL. By data scalability we mean how to learn from a very large dataset that cannot be efficiently processed by a single agent with limited resources.</p> <p>In the second part of this thesis, we provide a finite-sample analysis of some distributed reinforcement learning algorithms. By finite-sample analysis, we mean we provide an upper bound on the squared error of the algorithm for a given iteration of the algorithm. Or equivalently, since each iteration uses one data sample, we provide an upper bound of the squared error for a given number of data samples used. This type of analysis had been missing in the MARL literature, where most works on MARL have only provided asymptotic results for their proposed algorithms, which only tells us how the algorithmic error behaves as the number of samples used goes to infinity. </p> <p>The third part of this thesis focuses on applications with real-world systems. We have explored a real-world problem, namely transactive energy systems (TES), which can be represented as a multi-agent system. We have applied various reinforcement learning algorithms with the aim of learning an optimal control policy for this system. Through simulations, we have compared the performance of these algorithms and have illustrated the effect of partial observability (output feedback) when compared to full state feedback.</p> <p>In the last part we present some other work, specifically we present a distributed observer that aims to address learning with output feedback by estimating the state. The proposed algorithm is designed so that we do not require a complete model of state dynamics, and instead we use a parameterized model where the parameters are estimated along with the state.</p>

Made available in DSpace on 2016-08-17T14:52:53Z (GMT). No. of bitstreams: 1 Reginaldo Isaias Braga.pdf: 1510667 bytes, checksum: 681b1cdce694568db4239625f302fc05 (MD5) Previous issue date: 2004-07-02 / This thesis deals with implementation of a railway control interlocking simulator whose functional requirements can be adaptive, that is, modified, complemented or same redefined, through an approach based on Finite State Machines. Depending on its requirements, a controller of rail traffic can assume simple or critics characteristics in terms of safety or fault tolerance. The modeling based on the Finite State Machine approach provides the conditions for the adaptability of some functional requirements. This characteristic is emphasized better when we used automated tools of definition of state machines, which allow the easy integration with the remaining of the software application. A detailed discussion about safety's considerations in design of mission and safety critical systems is presented. Emphasis is given in the subjects related to the safety software. A methodology and tool (SpecTRM-RL) that implements it, derived of recent researches in safety software, is shown. These considerations are fundamental in the design and requirements analysis in railway s interlocking control systems. These kinds of systems can be classified as mission and safety critical and they are progressively migrating for relay based controllers to microprocessor-based controllers where in the whole or partly safety's requirements are implemented by software. Fault tolerance techniques either in software or hardware are fundamental in the context of this work. A selection of the most usual in systems that demand high reliability was detailed with the goal of emphasizing its potential application in the case study of a railway control-interlocking simulator. / Este trabalho trata da implementação de um simulador de controle de tráfego ferroviário cujos requisitos funcionais possam ser adaptáveis, isto é, modificados, complementados ou mesmo redefinidos, através de uma abordagem baseada em máquinas de estado finitas. Dependendo de seus requisitos, um controlador de tráfego ferroviário pode assumir características simples ou críticas em termos de segurança ou tolerância a falhas. A modelagem baseada na abordagem de máquinas de estado provê as condições para a adaptabilidade de alguns requisitos funcionais. Esta característica é melhor enfatizada quando usamos ferramentas automatizadas de definição de máquinas de estado, que permitem a fácil integração com o restante da aplicação. É apresentada uma discussão detalhada sobre as considerações de segurança que devem ser seguidas em um projeto de missão e segurança crítica, onde é dada ênfase nas questões relacionadas à segurança do software. Uma metodologia e ferramenta (SpecTRM-RL) que a implementa, derivada de pesquisas recentes em software de segurança é mostrada. Estas considerações desempenham um papel fundamental no projeto e análise de requisitos em sistemas de controle metroferroviários. Por serem estes classificados como de missão e segurança críticas e principalmente por estarem migrando gradativamente de dispositivos com lógica de controle baseada em relês para controladores microprocessados onde no todo ou em parte os requisitos de segurança são implementados por software. Também de fundamental importância no contexto deste trabalho são as técnicas de tolerância a falha tanto em hardware como em software. Uma seleção das mais usuais em sistemas que demandam elevada confiabilidade foi detalhada com objetivo de enfatizar a aplicabilidade no estudo de caso de um simulador de intertravamento ferroviário.

The rapid increase in clothing consumption over the last decades has meant a significant burden on the environment. Some examples of this are the high consumption of natural resources and land, increasing amounts of carbon emissions, excessive use of toxic chemicals, polluting air, land and water and increasing amounts of waste. The average clothing consumption in Europe is far higher than the world average, especially in Nordic countries including Sweden. Adopting and encouraging more circular economies is a part of the European commission‘s strategy to achieve sustainable consumption of resources and protection for environment. The purpose of this Research is to study the possibilities of improving the practice of second hand clothes consumption in Sweden and make it more effective towards the goal of sustainable consumption and circular economies. In order to achieve this goal, it is necessary to understand the challenges and find abilities for improvement in the value chain of used clothes in Sweden, with focus on the reverse logistics system of used clothes (collection and sorting processes) as well as the market of second hand clothes. In order to achieve the purpose of the study, the conceptual framework has been formulated upon studying relevant literature and results of previous studies. Value chains of three charity organizations have been studied and empirical data have been collected through semi-structured interviews with participants from these three organizations. The empirical findings have been interpreted in relation to the conceptual framework and the results of previous studies, in order to answer the research questions. The results of the study showed potential for improvement and development in the reverse logistics system (collections and sorting processes) of used clothes, as well as in the Swedish market of second hand clothes. However, benefiting of this potential requires collaboration between all stakeholders in the value chain of used clothes, including individuals (either in a position of donors or consumers), interested businesses (nonprofit or commercial organizations), government, municipalities, politicians and finally the private sector in clothing industry.

Breast cancer is the single most prevalent form of cancer in the UK today, accounting for around 16% of all diagnoses, and around 31% of diagnoses in women. The survival rates are good, however the prognosis is heavily dependent on the stage to which the cancer has progressed at diagnosis. In order to help accurately determine this stage, the sentinel lymph node of patients undergoing tumour resection surgery is removed and examined cytologically for the presence of cancerous cells. This examination of the lymph node is currently the rate-limiting step in the operation as a whole. There is evidence in the literature to suggest that cancerous tissue has a measurably different infrared spectrum from healthy tissue owing to chemical and morphological differences in the tissue. There is further evidence to suggest that in the visible and near infrared region, the spectra of healthy lymph node tissue is different from that of cancerous tissue. This thesis details a project, performed in collaboration with a surgical team at St Mary's Hospital, Newport, Isle of Wight, to analyse spectral images taken in the visible and near infrared, of biopsied lymph node tissue. In the course of the project, an unsupervised classificaton technique, based on an extension to the well establised 'spectral angle', was developed to analyse the spectral images. Psoriasis affects 2-3% of the UK population causing itchy and/or painful plaques on the skin. One of the main treatments for psoriasis is UV phototherapy, exposure to which is a risk factor for burning and the development of cancers. This thesis details an investigation into the possibility of developing a targeted UV phototherapy system based on spectral imaging to delineate the plaques and a proposed new UV laser for treatment.

There is a requirement to assess the safety of chemicals to both 'man' and the environment. Traditionally this was determined through the use of animal testing. However, there is an increased need to develop alternatives to animal testing for the determination of toxicity due to ethical and legislative reasons. One approach to replacing the use of animals is the application of computational methods. These include Quantitative Structure-Activity Relationships ((Q)SARs), which are the formalisation of the relationship of the effects (e.g. toxicity) for a series of chemicals and their physico-chemical and structural properties. Most QSARs for toxicity require knowledge of a chemicals hydrophobicity. Traditionally hydrophobicity has been characterised by the logarithm of the octanol/water partition coefficient (log P). Current experimental and predictive methods are limited in terms of applicability for compounds with extreme log P values, compounds ionised under the conditions of analysis and surface active agents. An alternative technique to assess hydrophobicity is Immobilised Artificial Membrane High Performance Liquid Chromatography (IAM-HPLC). The IAM stationary phase was developed initially to mimic biological membranes more realistically than octanol/water partitioning. This study has collated published literature values for the IAM retention index (kIAM), including details of the experimental procedure, into a database. The database includes 1910 values for 647 compounds. The effect of variability of experimental procedure on reported values was investigated. Key experimental parameters were identified that ensure comparable log kIAM values. An IAM-HPLC method was optimised; the HPLC method covers a range of hydrophobicities (log P of -1.35 to 6.03) and includes both unionised and ionised compounds under the conditions of analysis. Additionally the method has been demonstrated to be robust across system of analysis, column and stationary phase batch. The assessment of robustness increases confidence in the log kIAM (pH 7.4) values for 66 aliphatic and aromatic compounds determined as part of this work. Methods to predict log klAM (pH 7.4) were investigated. Both a fragment and correction factor method, based on theoretical structural features, and a 'classical' descriptor based QSAR approach, was applied to both the experimental log kIAM (pH 7.4) values determined in this work and comparable values collated from the literature. QSARs have been developed using log klAM as a descriptor to predict the ability of a chemical to cross the skin barrier and to predict various acute aquatic toxicity endpoints, using published skin absorption and ecotoxicity data respectively.

Uppsatsen jämför (a) Restaurang- och livsmedelsprogrammet med inriktningen bageri och konditori (RL-BAG) och (b) Livsmedelsprogrammet mot bageri (LP-BAG) i relation till varandra och fortsatta akademiska studier. Det är motsvarande program från Gy11 respektive Gy2000.Frågeställningen fokuserar på skillnaden mellan RL-BAG och LP-BAG gällande behörigheten och tillgång till akademiska utbildningar vid två lärosäten.Syftet med uppsatsen är att ta reda på hur skillnaden ser ut för de aktuella gymnasieprogrammen. Att det finns en skillnad är redan konstaterad i bl.a. utformningen av den nya gymnasieskolan. En aspekt att se på uppsatsen är att den har en utvärderande funktion av den senaste gymnasieförordningens yrkesprogram (Gy11).Metoderna som används i undersökningen är i huvudsak kvantitativa (i skildringen/resultatet). Den mäter det bivariata förhållandet mellan kombinationer av behörighetsgivande gymnasiala kurser och programtillgången vid de aktuella lärosätena. I metoden används bl.a. SWOT för att definiera skillnaderna.Resultatet visar att det finns en skillnad i urvalet. Den grundläggande behörigheten för Gy11 är betydligt mer fördelaktig än den motsvarande nivån för Gy2000. Även om kärnämnena räknas med i LP-BAG. Däremot har LP-BAG tillgång till fler särskilt behörighetsgivande kurser som i olika kombinationer ger tillgång till fler utbildningar.

Fifth Generation (5G) introduces technologies that expedite the adoption of mobile networks, such as densely connected devices, ultra-fast data rate, low latency and more. With those visions in 5G and 6G in the next step, the need for a higher transmission rate and lower latency is more demanding, possibly breaking Moore’s law. With Artificial Intelligence (AI) techniques becoming mature in the past decade, optimizing resource allocation in the network has become a highly demanding problem for Mobile Network Operators (MNOs) to provide better Quality of Service (QoS) with less cost. This thesis proposes a Reinforcement Learning (RL) solution on bandwidth allocation for network slicing integration in disaggregated Open Radio Access Network (O-RAN) architecture. O-RAN redefines traditional Radio Access Network (RAN) elements into smaller components with detailed functional specifications. The concept of open modularization leads to greater potential for managing resources of different network slices. In 5G mobile networks, there are three major types of network slices, Enhanced Mobile Broadband (eMBB), Ultra-Reliable Low Latency Communications (URLLC), and Massive Machine Type Communications (mMTC). Each network slice has different features in the 5G network; therefore, the resources can be relocated depending on different needs. The virtualization of O-RAN divides the RAN into smaller function groups. This helps the network slices to divide the shared resources further down. Compared to traditional sequential signal processing, allocating dedicated resources for each network slice can improve the performance individually. In addition, shared resources can be customized statically based on the feature requirement of each slice. To further enhance the bandwidth utilization on the disaggregated O-RAN, a RL algorithm is proposed in this thesis on midhaul bandwidth allocation shared between Centralized Unit (CU) and Distributed Unit (DU). A Python-based simulator has been implemented considering several types of mobile User Equipment (UE)s for this thesis. The simulator is later integrated with the proposed Q-learning model. The RL model finds the optimization on bandwidth allocation in midhaul between Edge Open Cloud (O-Cloud)s (DUs) and Regional O-Cloud (CU). The results show up to 50% improvement in the throughput of the targeted slice, fairness to other slices, and overall bandwidth utilization on the O-Clouds. In addition, the UE QoS has a significant improvement in terms of transmission time.

This work evaluated the influence of the composition on the physico-chemical characteristics, stability and photostability of nanostructures containing ubiquinone. Nanocapsules (NC) and nanoemulsions (NE) were prepared by interfacial deposition of preformed polymers and spontaneous emulsification, respectively. The formulations were characterized and compared in relation to the influence of the oil (argan or linseed oil) and its amount in the formulation (3% or 1.5%), the type of polymer (poli(ε-caprolactone), PCL or Eudragit® RL100, EUD), drug presence, besides the comparison between both nanostructures. To assay ubiquinone, the analytical method was validated and was considered linear, specific, precise and accurate. The formulations had adequate physico-chemical characteristics, with drug contents close to the theoretical value (1mg/mL), encapsulation efficiencies close to 100% and polydispersity index lower than 0.2 for formulations with 1.5% of oil. The reduction of oil concentration caused a decrease in the average diameter and polidispersity index of PCL NC of argan oil and EUD NC and NE of linseed oil. Furthermore, the addition of ubiquinone was able to modify the zeta potential of these formulations. Regarding the type of structure (NE or NC), the pH was influenced. EUD NC presented values lower than NE and PCL NC, regardless of the oil used. In addition, EUD NC with linseed oil had a higher zeta potential in module in relation to the NE and PCL NC. Concerning the type of oil used, particle diameter and polydispersity index were lower for EUD NC of linseed oil in relation to EUD NC of argan oil. Moreover, all formulations were able to photoprotect ubiquinone in comparison with free drug, occurring influence of the oil and the polymer. Regarding the stability, the formulations showed a reduced level of drug over 60 days, while the type of oil influenced this parameter. Nanostructures of argan oil showed higher levels. NC presented higher drug contents in relation to NE. The increase in particle size was only significant for NC of argan oil and PCL at 15 days, which showed a greater diameter than those of PCL NC of linseed oil and EUD NC of argan oil. An increase in zeta potential for NE was detected at 60 days, while EUD NC of argan oil showed a decrease in this parameter. The zeta potential was higher in module for NC of EUD in relation to the respective NE. / Este trabalho avaliou a influência da composição sobre as características físico-químicas, estabilidade e fotoestabilidade de nanoestruturas contendo ubiquinona. As nanocápsulas (NC) e nanoemulsões (NE) foram preparadas por deposição interfacial de polímero pré-formado e emulsificação espontânea, respectivamente. As formulações foram caracterizadas e comparadas quanto à influência do tipo de óleo (óleo de argan ou de linhaça) e de sua quantidade (3% ou 1,5%), quanto ao polímero poli(-caprolactona), PCL ou Eudragit® RL 100, EUD, quanto à presença do fármaco, além da comparação entre ambas as nanoestruturas. Para a quantificação da ubiquinona, o método foi validado, apresentando-se linear, específico, preciso e exato. As formulações apresentaram características físico-químicas adequadas, com teores próximos ao valor teórico de 1mg/mL, além de eficiências de encapsulamento próximas de 100% e de índices de polidispersão inferiores a 0,2 para as formulações com 1,5% de óleo. A redução na concentração de óleo causou diminuição do diâmetro médio e no índice de polidispersão das NC de óleo de argan e PCL e das NE e NC de EUD e óleo de linhaça. Além disso, a adição de ubiquinona foi capaz de alterar o potencial zeta destas formulações. Quanto ao tipo de estrutura, o pH sofreu influência, onde a NC de EUD apresentou valores inferiores à NE e NC de PCL, independente do óleo utilizado. Além disso, a NC de EUD e óleo de linhaça apresentou potencial zeta superior em módulo em relação à NE e NC de PCL. Quanto ao tipo de óleo empregado, o diâmetro de partícula e o índice de polidispersão foram menores para as NC de EUD e óleo de linhaça em relação às NC de EUD e óleo de argan. Além disso, todas as formulações foram capazes de fotoproteger a ubiquinona em comparação ao fármaco livre, havendo influência do óleo e do polímero. Quanto à estabilidade, as formulações apresentaram redução no teor de fármaco ao longo de 60 dias, sendo que as nanoestruturas de óleo de argan apresentaram teores maiores. As NC apresentaram teores maiores em relação às NE. O aumento no tamanho de partícula só foi significativo para a NC de óleo de argan e PCL aos 15 dias, onde apresentou diâmetro maior em relação às NC de PCL e óleo de linhaça e de EUD e óleo de argan. Um aumento no potencial zeta foi detectado para as NE aos 60 dias, enquanto que a NC de EUD e óleo de argan apresentou decréscimo neste parâmetro. O potencial zeta ainda apresentou-se maior em módulo para as NC de EUD em relação às respectivas NE.

Die Umweltveränderungen durch den Menschen sind auf unserer Erde allgegenwärtig. Entwaldung und Waldschädigung beeinflussen das System Erde entscheidend, denn Wälder bieten wichtige Ökosystemleistungen und sind Kernelement der Debatte um den Klimawandel, speziell hinsichtlich der globalen Kohlenstoffbilanz. Veränderungen der Waldbedeckung zu quantifizieren ist daher von herausragendem wissenschaftlichen Interesse. Ziel dieser Arbeit ist es, Waldbedeckungsveränderungen in den Westlichen Karpaten grenzübergreifend zu bestimmen, sowie Dynamiken der Biomasse von Nadelwäldern und deren Auswirkungen auf die oberirdische Kohlenstoffspeicherung abzuleiten. Die Karpatenwälder zeichnen sich durch ein hohes Maß an Biodiversität, einen großen Holzvorrat und als wichtiger Kohlenstoffspeicher für Europa aus. Jedoch sind diese Wälder auch geprägt von einer bewegten Geschichte der Landnutzung, hoher Luftverschmutzung und einer andauernden Waldabnahme. Mittels Methoden der Fernerkundung wurden Veränderungen in der Waldbedeckung für die Jahre 1985 bis 2010 abgeleitet. Die Ergebnisse zeigen, dass insbesondere das frühere Forstmanagement sowie die starke Luftverschmutzung zu Zeiten des Kommunismus gemeinsam die erhebliche Schädigung von Nadelwäldern bedingen. Fichtendominierte Bestände offenbaren dabei eine geringere Widerstandsfähigkeit gegenüber biotischen sowie abiotischen Belastungen, z.B. Schädlingen und Extremwettersituationen. Seit 2005 verwandelten sich die Nadelwälder infolge eines weit verbreiteten Biomasseverlustes von einer Netto-Kohlenstoffsenke in eine Netto-Kohlenstoffquelle. Die Analysen betonen den Einfluss bestimmter Standortfaktoren wie Waldtyp, vorherrschende Baumart, topographische Gegebenheiten, Brennpunkte der Umweltverschmutzung, Mikroklima und deren Interaktion auf die Waldabnahme. Die Arbeit legt eine komplexe sozio-ökologische Geschichte dar und erbringt Schätzungen über die Veränderung des oberirdischen Kohlenstoffvorrates der Wälder der Westlichen Karpaten. / Human-induced environmental change is evident across the globe. Deforestation and forest degradation are among the most critical impacts of humanity on the Earth system, as forests provide crucial ecosystem services, and are a key element in the global climate change discussion, specifically considering the global carbon balance. Therefore, monitoring and quantifying forest changes are of prime scientific interest. The main goals of this thesis were to monitor forest change across country borders in the Western Carpathians, and to assess coniferous forest biomass dynamics and their impact on aboveground forest carbon storage. Generally, Carpathian forests provide outstanding biodiversity levels, high growing stocks, and an important European carbon sink. However, the Western Carpathian forests are exceptional, with a turbulent land-use history, high airborne pollution loads, and ongoing forest decline. Forest change between 1985 and 2010 was quantified using remote sensing techniques. Results show that the synergistic effect of unsustainable forest management in the past and high pollution levels during communist times significantly damaged coniferous forests. Spruce-dominated stands exhibit lower resistance against biotic and abiotic impacts, and are more susceptible to pests and extreme weather events. Widespread biomass loss since 2005 has converted coniferous forests from a net carbon sink into a net carbon source. Cross-border analysis emphasized the role of site characteristics such as forest type, predominant species, topographic conditions, pollution hotspots, microclimate, and their interactions for forest decline. Summarizing, this thesis tells a complex socio-ecological story and provides estimates of aboveground carbon stock changes in Western Carpathian forests.

Submitted by Victor de Abreu Campos null (victorde.ac@gmail.com) on 2017-09-27T02:41:28Z No. of bitstreams: 1 dissertacao.pdf: 5473435 bytes, checksum: 1e76ecc15a4499dc141983740cc79e5a (MD5) / Approved for entry into archive by Monique Sasaki (sayumi_sasaki@hotmail.com) on 2017-09-28T13:43:21Z (GMT) No. of bitstreams: 1 campos_va_me_sjrp.pdf: 5473435 bytes, checksum: 1e76ecc15a4499dc141983740cc79e5a (MD5) / Made available in DSpace on 2017-09-28T13:43:21Z (GMT). No. of bitstreams: 1 campos_va_me_sjrp.pdf: 5473435 bytes, checksum: 1e76ecc15a4499dc141983740cc79e5a (MD5) Previous issue date: 2017-08-31 / Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP) / A quantidade vertiginosa de conteúdo multimídia acumulada diariamente tem demandado o desenvolvimento de abordagens eficazes de recuperação. Nesse contexto, ferramentas de reconhecimento de locutor capazes de identificar automaticamente um indivíduo pela sua voz são de grande relevância. Este trabalho apresenta uma nova abordagem de reconhecimento de locutor modelado como um cenário de recuperação e usando algoritmos de aprendizado não supervisionado recentes. A abordagem proposta considera Coeficientes Cepstrais de Frequência Mel (MFCCs) e Coeficientes de Predição Linear Perceptual (PLPs) como características de locutor, em combinação com múltiplas abordagens de modelagem probabilística, especificamente Quantização Vetorial, Modelos por Mistura de Gaussianas e i-vectors, para calcular distâncias entre gravações de áudio. Em seguida, métodos de aprendizado não supervisionado baseados em ranqueamento são utilizados para aperfeiçoar a eficácia dos resultados de recuperação e, com a aplicação de um classificador de K-Vizinhos Mais Próximos, toma-se uma decisão quanto a identidade do locutor. Experimentos foram conduzidos considerando três conjuntos de dados públicos de diferentes cenários e carregando ruídos de diversas origens. Resultados da avaliação experimental demonstram que a abordagem proposta pode atingir resultados de eficácia altos. Adicionalmente, ganhos de eficácia relativos de até +318% foram obtidos pelo procedimento de aprendizado não supervisionado na tarefa de recuperação de locutor e ganhos de acurácia relativos de até +7,05% na tarefa de identificação entre gravações de domínios diferentes. / The huge amount of multimedia content accumulated daily has demanded the development of effective retrieval approaches. In this context, speaker recognition tools capable of automatically identifying a person through their voice are of great relevance. This work presents a novel speaker recognition approach modelled as a retrieval scenario and using recent unsupervised learning methods. The proposed approach considers Mel-Frequency Cepstral Coefficients (MFCCs) and Perceptual Linear Prediction Coefficients (PLPs) as features along with multiple modelling approaches, namely Vector Quantization, Gaussian Mixture Models and i-vector to compute distances among audio objects. Next, rank-based unsupervised learning methods are used for improving the effectiveness of retrieval results and, based on a K-Nearest Neighbors classifier, an identity decision is taken. Several experiments were conducted considering three public datasets from different scenarios, carrying noise from various sources. Experimental results demonstrate that the proposed approach can achieve very high effectiveness results. In addition, effectiveness gains up to +318% were obtained by the unsupervised learning procedure in a speaker retrieval task. Also, accuracy gains up to +7,05% were obtained by the unsupervised learning procedure in a speaker identification task considering recordings from different domains. / FAPESP: 2015/07934-4