Cloud computing has become an established paradigm for running services on external infrastructure that dynamically allocates virtually unlimited capacity. This paradigm creates a new scenario for the deployment of applications and information technology (IT) services. In this model, complete applications and machine infrastructure are offered to users, who are charged only for the resources they consume. Thus, cloud resources are offered through service abstractions classified into three main categories: Software as a Service (SaaS), Platform as a Service (PaaS), and Infrastructure as a Service (IaaS). In IaaS, computing resources are offered as virtual machines to the end user. The aim to offer such unlimited resources necessitates distributing these virtual machines through multiple data centers. This distribution makes harder to fulfill a number of requirements including security, reliability, availability, and accounting. Accounting refers to how resources are recorded, accounted for, and charged. Even for a single cloud provider this task is hard, and it becomes more difficult for a federation of cloud computing, or federated cloud, in which a cloud provider dynamically outsources resources to other providers in response to demand variation. Thus, a cluster of clouds shares heterogeneous resources, requiring greater effort to manage and accurately account for the distributed resources. Some earlier research has addressed the development of platforms for federated clouds but without considering the accounting requirement. This dissertation presents a framework for charging IaaS with a focus on federated cloud. In order to gather information about this topic area and to generate guidelines for building the framework, we applied a systematic mapping study. This dissertation also presents an initial validation of the tool through a case study, showing evidence that the requirements generated through the mapping study were fulfilled by the framework and presenting indications of its feasibility in a real cloud service scenario / Submitted by João Arthur Martins (joao.arthur@ufpe.br) on 2015-03-10T18:37:17Z No. of bitstreams: 2 Dissertação Francisco Airton da Silva.pdf: 5605679 bytes, checksum: 61aa2b6df102174ff2e190ab47678cbf (MD5) license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) / Made available in DSpace on 2015-03-11T17:45:06Z (GMT). No. of bitstreams: 2 Dissertação Francisco Airton da Silva.pdf: 5605679 bytes, checksum: 61aa2b6df102174ff2e190ab47678cbf (MD5) license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) Previous issue date: 2013-02-27 / A Computação na Nuvem se tornou um paradigma consumado para executar serviços em infraestruturas externas, onde de uma forma virtual a capacidade praticamente ilimitada pode ser alocada dinamicamente. Este paradigma estabelece um novo cenário para a implantação de aplicações e serviços de TI. Neste modelo, desde aplicações completas até infraestrutura de máquinas são ofertadas a usuários, que são cobrados apenas pelo uso dos recursos consumidos. Desta forma, os bens de consumo da nuvem são ofertados através de abstrações de serviços, onde atualmente são citadas três principais categorias: Software como Serviço (SaaS), Plataforma como Serviço (PaaS) e Infraestrutura como Serviço (IaaS). No caso do IaaS são oferecidos recursos computacionais na forma de Máquinas Virtuais para o cliente final. Para atingir o aspecto ilimitado de recursos é necessário distribuir estas Máquinas Virtuais por vários Data Centers. Esta distribuição dificulta atender uma série de requisitos como Segurança, Confiabilidade, Disponibilidade e a Tarifação pelos recursos consumidos. A Tarifação refere-se a como os recursos são registrados, contabilizados e cobrados. Mesmo no caso de um único provedor esta tarefa não é fácil e existe um contexto em que esta dificuldade se torna ainda maior, conhecida como Federação de Computação na Nuvem ou também chamadas de Nuvens Federadas. Nuvens Federadas ocorrem quando um provedor de Computação na Nuvem terceiriza recursos dinamicamente para outros provedores em resposta à variação da demanda. Desta forma ocorre um aglomerado de nuvens, porém seus recursos são heterogêneos, acarretando num maior esforço para gerenciar os recursos distribuídos e por consequência para a Tarifação. Neste contexto foram identificadas pesquisas nesta área sobre plataformas para Nuvens Federadas, que não abordam o requisito de Tarifação. Esta dissertação apresenta um framework voltado à tarifação de Infraestrutura como Serviço com foco em Nuvens Federadas. Objetivando colher informações sobre esta área e consequentemente gerar insumos para fundamentar as decisões na construção do framework, foi aplicado um Estudo de Mapeamento Sistemático. Esta dissertação também apresenta uma validação inicial da ferramenta, através de um estudo experimental, mostrando indícios de que os requisitos gerados pelo Mapeamento Sistemático foram atendidos, bem como será viável a aplicação da solução por provedores de serviços de nuvem em um cenário real.

Submitted by Marilene Donadel (marilene.donadel@unioeste.br) on 2017-10-31T00:10:09Z No. of bitstreams: 1 Tiago_R_Fischer_2017.pdf: 2003875 bytes, checksum: c1bdcdd4bc45df8951aa64fa2fbe233d (MD5) / Made available in DSpace on 2017-10-31T00:10:09Z (GMT). No. of bitstreams: 1 Tiago_R_Fischer_2017.pdf: 2003875 bytes, checksum: c1bdcdd4bc45df8951aa64fa2fbe233d (MD5) Previous issue date: 2017-08-30 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / Collective actions are present in several organizational structures, among them credit union. This structure has been showing continuous growth rates in Brazil, and has increased its participation in the National Financial System. Credit union has played an important role in the economic and social spheres, emphasizing the role of agent that promotes regional development. Due to this situation, it should be pointed out that according to the literature, if, on the one hand, the growth of credit cooperatives has favored economic results, on the other hand it has generated challenges for the social sphere, due to the greater difficulty in organizing and motivating the credit union governance and the sustainability of this structure. As of 2011, Sicredi Alliance PR/SP, divided its cooperative base into 38 nuclei, so that the large group of cooperatives was divided into smaller groups. Such a procedure is known by Collective Action Theory as the creation of federative groups, maintaining the economic potential of a large group, without, however, losing the characteristics and benefits of small groups. In this sense, the objective of this research is to analyze the implications of group size on the collective performance of Sicredi Aliança PR/SP. The methodology used was the descriptive research, a method that allows the description of the characteristics of a given population or phenomenon, or the establishment of relations between the analyzed variables. The primary data were collected through the application of forms to the cooperative and observation of the assemblies; Already the secondary ones were collected in management reports and minutes of the assemblies. Data analysis was performed using Excel software and SPSS Statistic 24.0, through descriptive statistics and cross-tables, with quantitative-qualitative approach. It was possible to identify the positive influence of federative groups on the collective performance of the credit union under study. Specifically, it was observed that leftovers and the number of credit union increased significantly after the implementation of the new governance model. On the social side, it was also possible to verify a greater participation in assemblies and also the beginning of a credit union education program, which proved to be fundamental for the sustainability of the collective action under study. However, it has been identified that groups are still large and that reducing the size of the group without the accompaniment of cooperative education is not enough. / As ações coletivas estão presentes em diversas estruturas organizacionais, dentre as quais destaca-se o cooperativismo de crédito. Essa estrutura vem apresentando taxas de crescimento contínuas no Brasil, e tem aumentado sua participação no SFN – Sistema Financeiro Nacional. O cooperativismo de crédito, tem desempenhado relevante papel nas esferas econômica e social, destacando-se pelo papel de agente fomentador do desenvolvimento regional. Decorrente dessa conjuntura, cabe destacar que de acordo com a literatura, se por um lado, o crescimento das cooperativas de crédito tem favorecido os resultados econômicos, por outro tem gerado desafios para a esfera social, devido a maior dificuldade de organização e de motivar a participação dos cooperados, consequentemente, prejudicando a governança cooperativa e a sustentabilidade dessa estrutura. A partir de 2011, a Sicredi Aliança PR/SP, dividiu sua base de cooperados em 38 núcleos, de forma que o grupo grande de cooperados foi dividido em grupos menores. Tal procedimento é conhecido pela Teoria da Ação Coletiva como a criação de grupos federativos, mantendo o potencial econômico de um grande grupo, sem, contudo, perder as características e os benefícios dos grupos pequenos. Nesse sentido, o objetivo desta pesquisa foi analisar as implicações do tamanho do grupo no desempenho coletivo da Sicredi Aliança PR/SP. A metodologia utilizada foi a pesquisa descritiva, método que permite a descrição das características de uma determinada população ou fenômeno, ou ainda, o estabelecimento de relações entre as variáveis analisadas. Os dados primários foram coletados através da aplicação de formulários aos cooperados e observação das assembleias; já os secundários foram coletados em relatórios gerenciais e atas das assembleias. A análise dos dados foi feita com utilização dos softwares Excel e o SPSS Statistic 24.0, através da estatística descritiva e tabelas cruzadas, com abordagem quantitativa-qualitativa. Foi possível identificar a influência positiva dos grupos federativos no desempenho coletivo da cooperativa em estudo. Especificamente observou-se que as sobras e o número de cooperados aumentaram de forma significativa após a implantação do novo modelo de governança. Já no aspecto social, também foi possível verificar uma maior participação em assembleias e ainda o início de um programa de educação cooperativa, que se mostrou fundamental para a sustentabilidade da ação coletiva em estudo. Contudo, foi identificado que os grupos ainda continuam grandes e que a redução do tamanho do grupo sem o acompanhamento da educação cooperativa não é suficiente.

Main concern of this thesis is to help design a secure and reliable network system which keeps growing in complexity due to the interfaces with multiple logging sub-systems and to ensure the safety of the network environment for everyone involved. The parties somewhat involved in network systems are always in need of developing new solutions to security problems and striving to have a secure access into a network so as to fulfil their job in safe computing environments. Implementation and use of SSO (Single Sign-On) offering secure and reliable network in complex systems has been specifically defined for the overall security aspects of enterprises. The information to be used within and out of organization was structured layer by layer according to the organizational needs to define the sub-systems. The users in the enterprise were defined according to their role based profiles. Structuring the information layer by layer was shown to improve the level of security by providing multiple authentication mechanisms. Before implementing SSO system necessary requirements are identified. Thereafter, user identity management and different authentication mechanisms were defined together with the network protocols and standards to insure a safe exchange of information within and outside the organization. A marketing research was conducted in line of the SSO solutions. Threat and risk analysis was conducted according to ISO/IEC 27003:2010 standard. The degree of threat and risk were evaluated by considering their consequences and possibilities. These evaluations were processed by risk treatments. MoDAF (Ministry of Defence Architecture Framework) used to show what kind of resources, applications and the other system related information are needed and exchanged in the network. In essence some suggestions were made concerning the ideas of implementing SSO solutions presented in the discussion and analysis chapter.

Methods: We address these issues by transforming the TCGA data into the Semantic Web standard Resource Description Format (RDF), link it to relevant datasets in the Linked Open Data (LOD) cloud and further propose an efficient data distribution strategy to host the resulting 20.4 billion triples data via several SPARQL endpoints. Having the TCGA data distributed across multiple SPARQL endpoints, we enable biomedical scientists to query and retrieve information from these SPARQL endpoints by proposing a TCGA tailored federated SPARQL query processing engine named TopFed. Results: We compare TopFed with a well established federation engine FedX in terms of source selection and query execution time by using 10 different federated SPARQL queries with varying requirements. Our evaluation results show that TopFed selects on average less than half of the sources (with 100% recall) with query execution time equal to one third to that of FedX. Conclusion: With TopFed, we aim to offer biomedical scientists a single-point-of-access through which distributed TCGA data can be accessed in unison. We believe the proposed system can greatly help researchers in the biomedical domain to carry out their research effectively with TCGA as the amount and diversity of data exceeds the ability of local resources to handle its retrieval and parsing.

Product information search has become one of the most important application areas of the Web. Especially considering pricey technical products, consumers tend to carry out intensive research activities previous to the actual acquisition for creating an all-embracing view on the product of interest. Federated search backed by ontology-based product information representation shows great promise for easing this research process. The topic of this thesis is to develop a comprehensive technique for locating, extracting, and integrating information of arbitrary technical products in a widely unsupervised manner. The resulting homogeneous information sets allow a potential consumer to effectively compare technical products based on an appropriate federated product information system.:1. Introduction 1.1. Online Product Information Research 1.1.1. Current Online Product Information Research 1.1.2. Aspired Online Product Information Research 1.2. Federated Shopping Portals 1.3. Research Questions 1.4. Approach and Theses 1.4.1. Approach 1.4.2. Theses 1.4.3. Requirements 1.5. Goals and Non-Goals 1.5.1. Goals 1.5.2. Non-Goals 1.6. Contributions 1.7. Structure 2. Federated Information Systems 2.1. Information Access 2.1.1. Document Retrieval 2.1.2. Federated Search 2.1.3. Federated Ranking 2.2. Information Extraction 2.2.1. Information Extraction from Structured Sources 2.2.2. Information Extraction from Unstructured Sources 2.2.3. Information Extraction from Semi-structured Sources 2.3. Information Integration 2.3.1. Ontologies 2.3.2. Ontology Matching 2.4. Information Presentation 2.5. Product Information 2.5.1. Product Information Source Characteristics 2.5.2. Product Information Source Types 2.5.3. Product Information Integration Types 2.5.4. Product Information Types 2.6. Conclusions 3. A Federated Product Information System 3.1. Finding Basic Product Information 3.2. Enriching Product Information 3.3. Administrating Product Information 3.4. Displaying Product Information 3.5. Conclusions 4. Product Information Extraction from the Web 4.1. Vendor Product Information Search 4.1.1. Vendor Product Information Ranking 4.1.2. Vendor Product Information Extraction 4.2. Producer Product Information Search 4.2.1. Producer Product Document Retrieval 4.2.2. Producer Product Information Extraction 4.3. Third-Party Product Information Search 4.4. Conclusions 5. Product Information Integration for the Web 5.1. Product Representation 5.1.1. Domain Product Ontology 5.1.2. Application Product Ontology 5.1.3. Product Ontology Management 5.2. Product Categorization 5.3. Product Specifications Matching 5.3.1. General Procedure 5.3.2. Elementary Matchers 5.3.3. Evolutionary Matcher 5.3.4. Naïve Bayes Matcher 5.3.5. Result Selection 5.4. Product Specifications Normalization 5.4.1. Product Specifications Atomization 5.4.2. Product Specifications Value Normalization 5.5. Product Comparison 5.6. Conclusions 6. Evaluation 6.1. Implementation 6.1.1. Offers Service 6.1.2. Products Service 6.1.3. Snippets Service 6.1.4. Fedseeko 6.1.5. Fedseeko Browser Plugin 6.1.6. Fedseeko Mobile 6.1.7. Lessons Learned 6.2. Evaluation 6.2.1. Evaluation Measures 6.2.2. Gold Standard 6.2.3. Product Document Retrieval 6.2.4. Product Specifications Extraction 6.2.5. Product Specifications Matching 6.2.6. Comparison with Competitors 6.3. Conclusions 7. Conclusions and Future Work 7.1. Summary 7.2. Conclusions 7.3. Future Work A. Pseudo Code and Extraction Properties A.1. Pseudo Code A.2. Extraction Algorithm Properties A.2.1. Clustering Properties A.2.2. Purging Properties A.2.3. Dropping Properties B. Fedseeko Screenshots B.1. Offer Search B.2. Product Comparison / Die Produktinformationssuche hat sich zu einem der bedeutendsten Themen im Web entwickelt. Speziell im Bereich kostenintensiver technischer Produkte führen potenzielle Konsumenten vor dem eigentlichen Kauf des Produkts langwierige Recherchen durch um einen umfassenden Überblick für das Produkt von Interesse zu erlangen. Die föderierte Suche in Kombination mit ontologiebasierter Produktinformationsrepräsentation stellt eine mögliche Lösung dieser Problemstellung dar. Diese Dissertation stellt Techniken vor, die das automatische Lokalisieren, Extrahieren und Integrieren von Informationen für beliebige technische Produkte ermöglichen. Die resultierenden homogenen Produktinformationen erlauben einem potenziellen Konsumenten, zugehörige Produkte effektiv über ein föderiertes Produktinformationssystem zu vergleichen.:1. Introduction 1.1. Online Product Information Research 1.1.1. Current Online Product Information Research 1.1.2. Aspired Online Product Information Research 1.2. Federated Shopping Portals 1.3. Research Questions 1.4. Approach and Theses 1.4.1. Approach 1.4.2. Theses 1.4.3. Requirements 1.5. Goals and Non-Goals 1.5.1. Goals 1.5.2. Non-Goals 1.6. Contributions 1.7. Structure 2. Federated Information Systems 2.1. Information Access 2.1.1. Document Retrieval 2.1.2. Federated Search 2.1.3. Federated Ranking 2.2. Information Extraction 2.2.1. Information Extraction from Structured Sources 2.2.2. Information Extraction from Unstructured Sources 2.2.3. Information Extraction from Semi-structured Sources 2.3. Information Integration 2.3.1. Ontologies 2.3.2. Ontology Matching 2.4. Information Presentation 2.5. Product Information 2.5.1. Product Information Source Characteristics 2.5.2. Product Information Source Types 2.5.3. Product Information Integration Types 2.5.4. Product Information Types 2.6. Conclusions 3. A Federated Product Information System 3.1. Finding Basic Product Information 3.2. Enriching Product Information 3.3. Administrating Product Information 3.4. Displaying Product Information 3.5. Conclusions 4. Product Information Extraction from the Web 4.1. Vendor Product Information Search 4.1.1. Vendor Product Information Ranking 4.1.2. Vendor Product Information Extraction 4.2. Producer Product Information Search 4.2.1. Producer Product Document Retrieval 4.2.2. Producer Product Information Extraction 4.3. Third-Party Product Information Search 4.4. Conclusions 5. Product Information Integration for the Web 5.1. Product Representation 5.1.1. Domain Product Ontology 5.1.2. Application Product Ontology 5.1.3. Product Ontology Management 5.2. Product Categorization 5.3. Product Specifications Matching 5.3.1. General Procedure 5.3.2. Elementary Matchers 5.3.3. Evolutionary Matcher 5.3.4. Naïve Bayes Matcher 5.3.5. Result Selection 5.4. Product Specifications Normalization 5.4.1. Product Specifications Atomization 5.4.2. Product Specifications Value Normalization 5.5. Product Comparison 5.6. Conclusions 6. Evaluation 6.1. Implementation 6.1.1. Offers Service 6.1.2. Products Service 6.1.3. Snippets Service 6.1.4. Fedseeko 6.1.5. Fedseeko Browser Plugin 6.1.6. Fedseeko Mobile 6.1.7. Lessons Learned 6.2. Evaluation 6.2.1. Evaluation Measures 6.2.2. Gold Standard 6.2.3. Product Document Retrieval 6.2.4. Product Specifications Extraction 6.2.5. Product Specifications Matching 6.2.6. Comparison with Competitors 6.3. Conclusions 7. Conclusions and Future Work 7.1. Summary 7.2. Conclusions 7.3. Future Work A. Pseudo Code and Extraction Properties A.1. Pseudo Code A.2. Extraction Algorithm Properties A.2.1. Clustering Properties A.2.2. Purging Properties A.2.3. Dropping Properties B. Fedseeko Screenshots B.1. Offer Search B.2. Product Comparison

Den operativa miljön har förändrats vilket leder till att Försvarsmaktens ledningssystem behöver utvecklas. Sveriges nuvarande ledningssystem omhändertar inte de nya utmaningarna och Försvarsmaktens uppgifter blir allt mer komplexa. Sveriges militärstrategiska koncept baseras på att vinna tillsammans och undvika att förlora ensamma. Ordet tillsammans driver utvecklingen av internationellt sammabete och gemenskap. Under 2016 beslutade Sverige att ansluta sig till Natos utveckling av Federated Mission Networking (FMN). FMN syftar till förbättrat informationsutbyte mellan Nato, Nato-länderna och icke-Natoenheter. FMN är ett ramverk som omfattar alla ingående delar i ett ledningssystem. Konceptet bygger på principerna smidighet, flexibilitet och skalbarhet. Syftet med studien är att belysa ett svenskt införande av FMN-konceptet och undersöka hur detta kan bidra till ökad förmåga att omhänderta komplexiteten i ledning av gemensamma operationer. Studiens slutsats är att FMN-konceptet bidrar till ledningssystemets förmåga att skapa ordning genom fastställda rutiner och metoder. Det underlättar informationsdelning och ökar möjligheten till samordning och samverkan. Konceptet bidrar till en ökad interoperabilitet inom alla ingående delar i ledningssystemet.

While it is difficult for base stations to estimate the millimeter wave (mmWave) channels and find the optimal mmWave beam for user equipments (UEs) quickly, the sub-6 GHz channels which are usually easier to obtain and more robust to blockages could be used to reduce the time before initial access and enhance the reliability of mmWave communication. Considering that the channel information is collected by a massive number of radio base stations and would be sensitive to privacy and security, Federated Learning (FL) is a match for this use case. In practice, the channel vectors are usually subject to Non-Independently Distributed (non-IID) distributions due to the greatly varying wireless communication environments between different radio base stations and their UEs. To achieve satisfying performance for all radio base stations instead of only the majority of them, a useful solution is designing personalized methods for each radio base station. In this thesis, we implement two personalized FL methods including 1) Finetuning FL Model on Private Dataset of Each Client and 2) Adaptive Expert Models for FL to predict the optimal mmWave beamforming vector directly from the non-IID sub-6 GHz channel vectors generated from DeepMIMO. According to our experimental results, Finetuning FL Model on Private Dataset of Each Client achieves higher average mmWave downlink spectral efficiency than the global FL. Besides, in terms of the average Top-1 and Top-3 classification accuracies, its performance improvement over the global FL model even exceeds the improvement of the global FL over the pure local models. / Även om det är svårt för en basstation att uppskatta en kanal för millimetervåg (mmWave) och snabbt hitta den bästa mmWave-strålen för en användarutrustning (UE), kan den dra fördel av kanaler under 6 GHz, som i allmänhet är mer lättillgängliga och mer motståndskraftig mot blockering, för att minska tid för första besök och förbättra tillförlitligheten hos mmWave-kommunikation. Med tanke på att kanalinformation samlas in av ett stort antal radiobasstationer och är känslig för integritet och säkerhet är federated learning (FL) väl lämpat för detta användningsfall. I praktiken, eftersom den trådlösa kommunikationsmiljön varierar mycket mellan olika radiobasstationer och deras UE, följer kanalvektorer vanligtvis en icke-oberoende distribution (icke-IID). För att uppnå tillfredsställande prestanda för alla radiobasstationer, inte bara de flesta radiobasstationer, är en användbar lösning att utforma ett individuellt tillvägagångssätt för varje radiobasstation. I detta dokument implementerar vi två personliga FL-metoder, inklusive 1) finjustering av FL-modellen på varje klients privata datauppsättning och 2) en adaptiv expertmodell av FL för att direkt generera icke-IID sub-6 GHz kanalvektorer förutsäga optimal mmWave beamforming vektorer. Enligt våra experimentella resultat uppnår finjustering av FL-modellen på varje klients privata datauppsättning högre genomsnittlig mmWave-nedlänksspektral effektivitet än global FL. Dessutom överträffar dess prestandaförbättring jämfört med den globala FL-modellen till och med den för den globala FL jämfört med den rent lokala modellen vad gäller genomsnittlig klassificeringsnoggrannhet i topp-1 och topp-3.

Dans l’apprentissage fédéré, un modèle global est appris en agrégeant les mises à jour du modèle calculées à partir d’un ensemble de nœuds clients, un défi clé dans ce domaine est l’hétérogénéité des données entre les clients qui dégrade les performances du modèle. Les algorithmes d’apprentissage fédéré standard effectuent plusieurs étapes de gradient avant de synchroniser le modèle, ce qui peut amener les clients à minimiser exagérément leur propre objectif local et à s’écarter de la solution globale. Nous démontrons que dans un tel contexte, les modèles de clients individuels subissent un oubli catastrophique par rapport aux données d’autres clients et nous proposons une approche simple mais efficace qui modifie l’objectif d’entropie croisée sur une base par client en repondérant le softmax de les logits avant de calculer la perte. Cette approche protège les classes en dehors de l’ensemble d’étiquettes d’un client d’un changement de représentation brutal. Grâce à une évaluation empirique approfondie, nous démontrons que notre approche peut atténuer ce problème, en apportant une amélioration continue aux algorithmes d’apprentissage fédéré standard. Cette approche est particulièrement avantageux dans les contextes d’apprentissage fédéré difficiles les plus étroitement alignés sur les scénarios du monde réel où l’hétérogénéité des données est élevée et la participation des clients à chaque cycle est faible. Nous étudions également les effets de l’utilisation de la normalisation par lots et de la normalisation de groupe avec notre méthode et constatons que la normalisation par lots, qui était auparavant considérée comme préjudiciable à l’apprentissage fédéré, fonctionne exceptionnellement bien avec notre softmax repondéré, remettant en question certaines hypothèses antérieures sur la normalisation dans un système fédéré / In Federated Learning, a global model is learned by aggregating model updates computed from a set of client nodes, a key challenge in this domain is data heterogeneity across clients which degrades model performance. Standard federated learning algorithms perform multiple gradient steps before synchronizing the model which can lead to clients overly minimizing their own local objective and diverging from the global solution. We demonstrate that in such a setting, individual client models experience a catastrophic forgetting with respect to data from other clients and we propose a simple yet efficient approach that modifies the cross-entropy objective on a per-client basis by re-weighting the softmax of the logits prior to computing the loss. This approach shields classes outside a client’s label set from abrupt representation change. Through extensive empirical evaluation, we demonstrate our approach can alleviate this problem, providing consistent improvement to standard federated learning algorithms. It is particularly beneficial under the challenging federated learning settings most closely aligned with real world scenarios where data heterogeneity is high and client participation in each round is low. We also investigate the effects of using batch normalization and group normalization with our method and find that batch normalization which has previously been considered detrimental to federated learning performs particularly well with our re-weighted softmax, calling into question some prior assumptions about normalization in a federated setting

Machine Learning (ML) is the current trend that most industries opt for to improve their business and operations. ML has also been adopted in the financial markets, where well-funded financial institutions employ the latest ML algorithms to gain an advantage on the market. The darker side of ML is the potential emergence of complex algorithmic trading schemes that are abusive and manipulative. Because of this, it is inevitable that ML will be applied to financial market surveillance in order to detect these abusive and manipulative trading strategies. Ideally, an accurate ML detection model would be developed with data from many financial institutions or trading venues. However, such ML models require vast quantities of data, which poses a problem in market surveillance where data is sensitive or limited. Data sharing between companies or countries is typically accompanied by legal and privacy concerns. By training ML models on distributed datasets, Federated Learning (FL) overcomes these issues by eliminating the need to centralise sensitive data. This thesis aimed to address these ML related issues in market surveillance by implementing and evaluating a FL model. FL enables a group of independent data-holding clients with the same intention to build a shared ML model collaboratively without compromising private data. In this work, a ML model is initially deployed in a centralised data setting and trained to detect the manipulative trading scheme known as spoofing. The LSTM-Autoencoder was the model chosen method for this task. The same model is also implemented in a federated setting but with decentralised data, using the FL framework FEDn. Another FL framework, Flower, is also employed to evaluate the performance of FEDn. Experiments were conducted comparing the FL models to the conventional centralised learning model, as well as comparing the two frameworks to each other. The results showed that under certain circumstances, the FL models performed better than the centralised model in detecting spoofing. FEDn was equivalent to Flower in terms of detection performance. In addition, the results indicated that Flower was marginally faster than FEDn. It is assumed that variations in the experimental setup and stochasticity account for the performance disparity.

Federated learning poses a statistical challenge when training on highly heterogeneous sequence data. For example, time-series telecom data collected over long intervals regularly shows mixed fluctuations and patterns. These distinct distributions are an inconvenience when a node not only plans to contribute to the creation of the global model but also plans to apply it on its local dataset. In this scenario, adopting a one-fits-all approach might be inadequate, even when using state-of-the-art machine learning techniques for time series forecasting, such as Long Short-Term Memory (LSTM) networks, which have proven to be able to capture many idiosyncrasies and generalise to new patterns. In this work, we show that by clustering the clients using these patterns and selectively aggregating their updates in different global models can improve local performance with minimal overhead, as we demonstrate through experiments using realworld time series datasets and a basic LSTM model. / Federated Learning utgör en statistisk utmaning vid träning med starkt heterogen sekvensdata. Till exempel så uppvisar tidsseriedata inom telekomdomänen blandade variationer och mönster över längre tidsintervall. Dessa distinkta fördelningar utgör en utmaning när en nod inte bara ska bidra till skapandet av en global modell utan även ämnar applicera denna modell på sin lokala datamängd. Att i detta scenario införa en global modell som ska passa alla kan visa sig vara otillräckligt, även om vi använder oss av de mest framgångsrika modellerna inom maskininlärning för tidsserieprognoser, Long Short-Term Memory (LSTM) nätverk, vilka visat sig kunna fånga komplexa mönster och generalisera väl till nya mönster. I detta arbete visar vi att genom att klustra klienterna med hjälp av dessa mönster och selektivt aggregera deras uppdateringar i olika globala modeller kan vi uppnå förbättringar av den lokal prestandan med minimala kostnader, vilket vi demonstrerar genom experiment med riktigt tidsseriedata och en grundläggande LSTM-modell.