Dissertations / Theses on the topic 'Multi-domain'

Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Electrical Engineering and Computer Science, 2004.
This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections.
Includes bibliographical references (p. 121-128).
by Christine J. Alvarado.
Ph.D.

Internet propose aujourd’hui aux utilisateurs de services en ligne de commenter, d’éditer et de partager leurs points de vue sur différents sujets de discussion. Ce type de contenu est maintenant devenu la ressource principale pour les analyses d’opinions sur Internet. Néanmoins, à cause des abréviations, du bruit, des fautes d’orthographe et toutes autres sortes de problèmes, les outils de traitements automatiques des langues, y compris les reconnaisseurs d’entités nommées et les étiqueteurs automatiques morphosyntaxiques, ont des performances plus faibles que sur les textes bien-formés (Ritter et al., 2011).Cette thèse a pour objet la reconnaissance d’entités nommées sur les contenus générés par les utilisateurs sur Internet. Nous avons établi un corpus d’évaluation avec des textes multi-sources et multi-domaines. Ensuite, nous avons développé un modèle de champs conditionnels aléatoires, entrainé sur un corpus annoté provenant des contenus générés par les utilisateurs.Dans le but d’améliorer les résultats de la reconnaissance d’entités nommées, nous avons d’abord développé un étiqueteur morpho-syntaxique sur les contenus générés par les utilisateurs et nous avons utilisé les étiquettesprédites comme un attribut du modèle des champs conditionnels aléatoire. Enfin, pour transformer les contenus générés par les utilisateurs en textes bien-formés, nous avons développé un modèle de normalisation lexicale basé sur des réseaux de neurones pour proposer une forme correcte pour les mots non-standard
The increasing mass of User-Generated Content (UGC) on the Internet means that people are now willing to comment, edit or share their opinions on different topics. This content is now the main ressource for sentiment analysis on the Internet. Due to abbreviations, noise, spelling errors and all other problems with UGC, traditional Natural Language Processing (NLP) tools, including Named Entity Recognizers and part-of-speech (POS) taggers, perform poorly when compared to their usual results on canonical text (Ritter et al., 2011).This thesis deals with Named Entity Recognition (NER) on some User-Generated Content (UGC). We have created an evaluation dataset including multi-domain and multi-sources texts. We then developed a Conditional Random Fields (CRFs) model trained on User-Generated Content (UGC).In order to improve NER results in this context, we first developed a POStagger on UGC and used the predicted POS tags as a feature in the CRFs model. To turn UGC into canonical text, we also developed a normalization model using neural networks to propose a correct form for Non-Standard Words (NSW) in the UGC
各种社交网络应用使得互联网用户对各种话题的实时评价，编辑和分享成为可能。这类用户生成的文本内容(User Generated content)已成为社交网络上意见分析的主要目标和来源。但是，此类文本内容中包含的缩写，噪声（不规则词），拼写错误以及其他各种问题导致包括命名实体识别，词性标注在内的传统的自然语言处理工具的性能，相比良好组成的文本降低了许多【参见Ritter 2011】。本论文的主要目标是针对社交网络上用户生成文本内容的命名实体识别。我们首先建立了一个包含多来源，多领域文本的有标注的语料库作为标准评价语料库。然后，我们开发了一个由社交网络用户生成文本训练的基于条件随机场(Conditional Random Fields)的序列标注模型。基于改善这个命名实体识别模型的目的，我们又开发了另一个同样由社交网络用户生成内容训练的词性标注模型，并使用此模型预测的词性作为命名实体识别的条件随机场模型的特征。最后，为了将用户生成文本内容转换成相对标准的良好文本内容，我们开发了一个基于神经网络的词汇标准化模型，用以改正用户生成文本内容中的不标准字，并使用模型提供的改正形式作为命名实体识别的条件随机场模型的特征，借以改善原模型的性能。

Mestrado em Engenharia de Computadores e Telemática
This dissertation proposes and presents a technologic framework to perform record linkage. The proposed approach begins by defining a sequence of tasks necessary for record linkage. Then, several methods to split these tasks in small work units are discussed. The solution architecture is based on various executors that carry out the work units in a parallel manner, thereby making the process quicker. Finally, the benefits of using this approach in different contexts are also presented.
Esta dissertação propõe e apresenta uma ferramenta tecnológica que permite realizar mapeamento de registos. A abordagem proposta começa por definir uma sequência de tarefas necessárias para efetuar o mapeamento de registos. São depois discutidos métodos de separar estas tarefas em unidades de trabalho reduzidas. A solução baseia-se numa arquitetura composta por vários executores que levam a cabo essas unidades de trabalho de uma forma paralela, objectivando-se um processo mais rápido. As vantagens da utilização desta abordagem em diferentes contextos são também estudadas.

Ce travail a été réalisé dans un contexte industriel dans l’entreprise European Aeronautics Defense and Space Company (EADS). EADS recherche donc constamment de nouveaux moyens d’analyser la sécurité du système homme-machine dans sa globalité, c'est-à-dire l’avion, les pilotes et les procédures, comme un tout. Ces analyses de sécurité sont réalisées tout au long du cycle de conception, depuis la conception préliminaire à la conception détaillée et jusqu'à la phase de certification. EADS essaye de réaliser ces analyses beaucoup plus tôt dans le cycle de conception, lorsque seulement des modèles sont disponibles. Une question en suspend est alors comment assurer la collaboration entre les experts de plusieurs domaines (cockpit, procédures, psychologie cognitive) pour qu’ils puissent construite un artefact commun (un modèle) sur lequel ils pourraient s’appuyer pour réaliser les analyses de la sécurité du système homme-machine dans sa globalité. Cette thèse identifie et adresse trois questions. Premièrement, les langages de modélisation dédiés aux domaines doivent être sémantiquement alignés pour que l’artefact commun puisse être exprimé de manière cohérente. Deuxièmement, des notations visuelles spécifiques aux domaines doivent être produites pour le même artefact commun. Troisièmement, l’activité de modélisation des experts doit être supportée au mieux et pour cela les notations visuelles des langages de modélisation doivent être le plus proche possible des pratiques courantes dans les domaines respectifs. Cette thèse propose l’Infrastructure xOWL comme une solution intégrée à ces trois questions
This work has been realized in an industrial context at the European Aeronautics Defense and Space Company (EADS). EADS is researching new ways to assess the safety of the overall human-machine system, i. E. , the aircraft, pilots and operating procedures as a whole. These safety assessments are conducted throughout the design cycle of the product, from the preliminary design to the detailed design and up to the certification phase. EADS is trying to perform these thorough safety assessments much earlier in the development cycle, when only models are available, thus, phasing-in a model-driven approach of the problem. An issue is then how to enable the collaboration of experts from multiple domains (cockpit, procedures, and cognitive psychology) for them to build a common model artifact that can be leveraged in the safety assessment of the overallhuman-machine system. This work considers that experts in each domain must be provided a domain-specific modeling environment, giving them access to a common model artifact, but through a domain-specific notation. This thesis identifies and considers three issues in this regard. First, the domain-specific modeling languages need to be semantically aligned so that the common model artifact can be consistently expressed. Second, multiple domain-specific visual notations need to be produced for the same underlying common model artifact. Third, domain experts modeling activities need to be supported at best and thus the provided domain-specific notations need to be as close as possible to the existing practices in the domains. This thesis then proposes the xOWL Infrastructure as an integrated solution to the three issues

Traceability is an important concern in projects that span dierent engineering domains. Traceability can also be mandated, exploited and managed across the engineering lifecycle, and may involve defining connections between heterogeneous models. As a result, traceability can be considered to be multi-domain. This thesis introduces the concept and challenges of multi-domain traceability and explains how it can be used to support typical traceability scenarios. It proposes a model-based approach to develop a traceability solution which eectively operates across multiple engineering domains. The approach introduced a collection of tasks and structures which address the identified challenges for a traceability solution in multi-domain projects. The proposed approach demonstrates that modelling principles and MDE techniques cab help to address current challenges and consequently improve the eectiveness of a multi-domain traceability solution. A prototype of the required tooling to support the approach is implemented with EMF and atop Epsilon; it consists of an implementation of the proposed structures (models) and model management operations to support traceability. Moreover, the approach is illustrated in the context of two safety-critical projects where multi-domain traceability is required to underpin certification arguments.

Simulators are important tools for analyzing and evaluating different design options for wireless sensor networks (sensornets) and hence, have been intensively studied in the past decades. However, existing simulators only support evaluations of protocols and software aspects of sensornet design. They cannot accurately capture the significant impacts of various hardware designs on sensornet performance. As a result, the performance/energy benefits of customized hardware designs are difficult to be evaluated in sensornet research. To fill in this technical void, in this thesis, we describe the design and implementation of SUNSHINE, a scalable hardware-software cross-domain simulator for sensornet applications. SUNSHINE is the first sensornet simulator that effectively supports joint evaluation and design of sensor hardware and software performance in a networked context. SUNSHINE captures the performance of network protocols, software and hardware up to cycle-level accuracy through its seamless integration of three existing sensornet simulators: a network simulator TOSSIM, an instruction-set simulator SimulAVR and a hardware simulator GEZEL. SUNSHINE solves challenging design problems, including data exchanges and time synchronizations across different simulation domains and simulation accuracy levels. SUNSHINE also provides hardware specification scheme for simulating flexible and customized hardware designs. Several experiments are given to illustrate SUNSHINEâ s cross-domain simulation capability, demonstrating that SUNSHINE is an efficient tool for software-hardware codesign in sensornet research.
Master of Science

Telecom operators networks' management and control remains partitioned by technology, equipment supplier and networking layer. In some segments, the network operations are highly costly due to the need of the individual, and even manual, configuration of the network equipment by highly specialized personnel. In multi-vendor networks, expensive and never ending integration processes between Network Management Systems (NMSs) and the rest of systems (OSSs, BSSs) is a common situation, due to lack of adoption of standard interfaces in the management systems of the different equipment suppliers. Moreover, the increasing impact of the new traffic flows introduced by the deployment of massive Data Centers (DCs) is also imposing new challenges that traditional networking is not ready to overcome. The Fifth Generation of Mobile Technology (5G) is also introducing stringent network requirements such as the need of connecting to the network billions of new devices in IoT paradigm, new ultra-low latency applications (i.e., remote surgery) and vehicular communications. All these new services, together with enhanced broadband network access, are supposed to be delivered over the same network infrastructure. In this PhD Thesis, an holistic view of Network and Cloud Computing resources, based on the recent innovations introduced by Software Defined Networking (SDN), is proposed as the solution for designing an end-to-end multi-layer, multi-technology and multi-domain cloud and transport network management architecture, capable to offer end-to-end services from the DC networks to customers access networks and the virtualization of network resources, allowing new ways of slicing the network resources for the forthcoming 5G deployments. The first contribution of this PhD Thesis deals with the design and validation of SDN based network orchestration architectures capable to improve the current solutions for the management and control of multi-layer, multi-domain backbone transport networks. These problems have been assessed and progressively solved by different control and management architectures which has been designed and evaluated in real evaluation environments. One of the major findings of this work has been the need of developed a common information model for transport network's management, capable to describe the resources and services of multilayer networks. In this line, the Control Orchestration Protocol (COP) has been proposed as a first contriution towards an standard management interface based on the main principles driven by SDN. Furthermore, this PhD Thesis introduces a novel architecture capable to coordinate the management of IT computing resources together with inter- and intra-DC networks. The provisioning and migration of virtual machines together with the dynamic reconfiguration of the network has been successfully demonstrated in a feasible timescale. Moreover, a resource optimization engine is introduced in the architecture to introduce optimization algorithms capable to solve allocation problems such the optimal deployment of Virtual Machine Graphs over different DCs locations minimizing the inter-DC network resources allocation. A baseline blocking probability results over different network loads are also presented. The third major contribution is the result of the previous two. With a converged cloud and network infrastructure controlled and operated jointly, the holistic view of the network allows the on-demand provisioning of network slices consisting of dedicated network and cloud resources over a distributed DC infrastructure interconnected by an optical transport network. The last chapters of this thesis discuss the management and orchestration of 5G slices based over the control and management components designed in the previous chapters. The design of one of the first network slicing architectures and the deployment of a 5G network slice in a real Testbed, is one of the major contributions of this PhD Thesis.
La gestión y el control de las redes de los operadores de red (Telcos), todavía hoy, está segmentado por tecnología, por proveedor de equipamiento y por capa de red. En algunos segmentos (por ejemplo en IP) la operación de la red es tremendamente costosa, ya que en muchos casos aún se requiere con guración individual, e incluso manual, de los equipos por parte de personal altamente especializado. En redes con múltiples proveedores, los procesos de integración entre los sistemas de gestión de red (NMS) y el resto de sistemas (p. ej., OSS/BSS) son habitualmente largos y extremadamente costosos debido a la falta de adopción de interfaces estándar por parte de los diferentes proveedores de red. Además, el impacto creciente en las redes de transporte de los nuevos flujos de tráfico introducidos por el despliegue masivo de Data Centers (DC), introduce nuevos desafíos que las arquitecturas de gestión y control de las redes tradicionales no están preparadas para afrontar. La quinta generación de tecnología móvil (5G) introduce nuevos requisitos de red, como la necesidad de conectar a la red billones de dispositivos nuevos (Internet de las cosas - IoT), aplicaciones de ultra baja latencia (p. ej., cirugía a distancia) y las comunicaciones vehiculares. Todos estos servicios, junto con un acceso mejorado a la red de banda ancha, deberán ser proporcionados a través de la misma infraestructura de red. Esta tesis doctoral propone una visión holística de los recursos de red y cloud, basada en los principios introducidos por Software Defined Networking (SDN), como la solución para el diseño de una arquitectura de gestión extremo a extremo (E2E) para escenarios de red multi-capa y multi-dominio, capaz de ofrecer servicios de E2E, desde las redes intra-DC hasta las redes de acceso, y ofrecer ademas virtualización de los recursos de la red, permitiendo nuevas formas de segmentación en las redes de transporte y la infrastructura de cloud, para los próximos despliegues de 5G. La primera contribución de esta tesis consiste en la validación de arquitecturas de orquestración de red, basadas en SDN, para la gestión y control de redes de transporte troncales multi-dominio y multi-capa. Estos problemas (gestion de redes multi-capa y multi-dominio), han sido evaluados de manera incremental, mediante el diseño y la evaluación experimental, en entornos de pruebas reales, de diferentes arquitecturas de control y gestión. Uno de los principales hallazgos de este trabajo ha sido la necesidad de un modelo de información común para las interfaces de gestión entre entidades de control SDN. En esta línea, el Protocolo de Control Orchestration (COP) ha sido propuesto como interfaz de gestión de red estándar para redes SDN de transporte multi-capa. Además, en esta tesis presentamos una arquitectura capaz de coordinar la gestión de los recursos IT y red. La provisión y la migración de máquinas virtuales junto con la reconfiguración dinámica de la red, han sido demostradas con éxito en una escala de tiempo factible. Además, la arquitectura incorpora una plataforma para la ejecución de algoritmos de optimización de recursos capaces de resolver diferentes problemas de asignación, como el despliegue óptimo de Grafos de Máquinas Virtuales (VMG) en diferentes DCs que minimizan la asignación de recursos de red. Esta tesis propone una solución para este problema, que ha sido evaluada en terminos de probabilidad de bloqueo para diferentes cargas de red. La tercera contribución es el resultado de las dos anteriores. La arquitectura integrada de red y cloud presentada permite la creación bajo demanda de "network slices", que consisten en sub-conjuntos de recursos de red y cloud dedicados para diferentes clientes sobre una infraestructura común. El diseño de una de las primeras arquitecturas de "network slicing" y el despliegue de un "slice" de red 5G totalmente operativo en un Testbed real, es una de las principales contribuciones de esta tesis.
La gestió i el control de les xarxes dels operadors de telecomunicacions (Telcos), encara avui, està segmentat per tecnologia, per proveïdors d’equipament i per capes de xarxa. En alguns segments (Per exemple en IP) l’operació de la xarxa és tremendament costosa, ja que en molts casos encara es requereix de configuració individual, i fins i tot manual, dels equips per part de personal altament especialitzat. En xarxes amb múltiples proveïdors, els processos d’integració entre els Sistemes de gestió de xarxa (NMS) i la resta de sistemes (per exemple, Sistemes de suport d’operacions - OSS i Sistemes de suport de negocis - BSS) són habitualment interminables i extremadament costosos a causa de la falta d’adopció d’interfícies estàndard per part dels diferents proveïdors de xarxa. A més, l’impacte creixent en les xarxes de transport dels nous fluxos de trànsit introduïts pel desplegament massius de Data Centers (DC), introdueix nous desafiaments que les arquitectures de gestió i control de les xarxes tradicionals que no estan llestes per afrontar. Per acabar de descriure el context, la cinquena generació de tecnologia mòbil (5G) també presenta nous requisits de xarxa altament exigents, com la necessitat de connectar a la xarxa milers de milions de dispositius nous, dins el context de l’Internet de les coses (IOT), o les noves aplicacions d’ultra baixa latència (com ara la cirurgia a distància) i les comunicacions vehiculars. Se suposa que tots aquests nous serveis, juntament amb l’accés millorat a la xarxa de banda ampla, es lliuraran a través de la mateixa infraestructura de xarxa. Aquesta tesi doctoral proposa una visió holística dels recursos de xarxa i cloud, basada en els principis introduïts per Software Defined Networking (SDN), com la solució per al disseny de una arquitectura de gestió extrem a extrem per a escenaris de xarxa multi-capa, multi-domini i consistents en múltiples tecnologies de transport. Aquesta arquitectura de gestió i control de xarxes transport i recursos IT, ha de ser capaç d’oferir serveis d’extrem a extrem, des de les xarxes intra-DC fins a les xarxes d’accés dels clients i oferir a més virtualització dels recursos de la xarxa, obrint la porta a noves formes de segmentació a les xarxes de transport i la infrastructura de cloud, pels propers desplegaments de 5G. La primera contribució d’aquesta tesi doctoral consisteix en la validació de diferents arquitectures d’orquestració de xarxa basades en SDN capaces de millorar les solucions existents per a la gestió i control de xarxes de transport troncals multi-domini i multicapa. Aquests problemes (gestió de xarxes multicapa i multi-domini), han estat avaluats de manera incremental, mitjançant el disseny i l’avaluació experimental, en entorns de proves reals, de diferents arquitectures de control i gestió. Un dels principals troballes d’aquest treball ha estat la necessitat de dissenyar un model d’informació comú per a les interfícies de gestió de xarxes, capaç de descriure els recursos i serveis de la xarxes transport multicapa. En aquesta línia, el Protocol de Control Orchestration (COP, en les seves sigles en anglès) ha estat proposat en aquesta Tesi, com una primera contribució cap a una interfície de gestió de xarxa estàndard basada en els principis bàsics de SDN. A més, en aquesta tesi presentem una arquitectura innovadora capaç de coordinar la gestió de els recursos IT juntament amb les xarxes inter i intra-DC. L’aprovisionament i la migració de màquines virtuals juntament amb la reconfiguració dinàmica de la xarxa, ha estat demostrat amb èxit en una escala de temps factible. A més, l’arquitectura incorpora una plataforma per a l’execució d’algorismes d’optimització de recursos, capaços de resoldre diferents problemes d’assignació, com el desplegament òptim de Grafs de Màquines Virtuals (VMG) en diferents ubicacions de DC que minimitzen la assignació de recursos de xarxa entre DC. També es presenta una solució bàsica per a aquest problema, així com els resultats de probabilitat de bloqueig per a diferents càrregues de xarxa. La tercera contribució principal és el resultat dels dos anteriors. Amb una infraestructura de xarxa i cloud convergent, controlada i operada de manera conjunta, la visió holística de la xarxa permet l’aprovisionament sota demanda de "network slices" que consisteixen en subconjunts de recursos d’xarxa i cloud, dedicats per a diferents clients, sobre una infraestructura de Data Centers distribuïda i interconnectada per una xarxa de transport òptica. Els últims capítols d’aquesta tesi tracten sobre la gestió i organització de "network slices" per a xarxes 5G en funció dels components de control i administració dissenyats i desenvolupats en els capítols anteriors. El disseny d’una de les primeres arquitectures de "network slicing" i el desplegament d’un "slice" de xarxa 5G totalment operatiu en un Testbed real, és una de les principals contribucions d’aquesta tesi.

With the advent of 5G, telecommunications are undergoing unprecedented changes. Network architectures are evolving to support new business models and services. The transition to 5G encompasses many technological aspects and design processes. The new generation system focuses on bringing an overall performance improvement in terms of latency and throughput, as well as on exibility and scalability of the infrastructure. Network Slicing plays a key role in the aforementioned transition, allowing to build different logical networks on top of the same common infrastructure. Moreover, the evolution of the core network is a key aspect as well for the integration of new functionalities and to enable full 5G capabilities. In this regard, the core network needs to move from the current complex, cost-inefficient infrastructure to a more dynamic architecture. This can be achieved by leveraging virtualization technologies supporting paradigms including Network Function Virtualization (NFV), Software Defined Networking (SDN) and Cloud Computing, which are considered key enablers for 5G. This thesis focuses on the orchestration of a virtualized mobile core network, starting from the modelling of MANO descriptors and moving to the deployment in a multi-domain cloud scenario. In particular, the objective of the activity is to highlight methodologies and problems relevant to that deployment, including the configuration of all the involved elements and available software solutions. For the implementation, Open Source MANO (OSM) is used as management and orchestration platform, and OpenStack as cloud platform and Virtualized Infrastructure Manager (VIM). Moreover, the NextEPC software suite is utilized to implement the core network elements whereas eNodeB and UEs are simulated with the L2 nFAPI software provided by OpenAirInterface.

Optical networks have been widely adopted as the transport network in research industry for years. To better utilize the high speed benefit of optical transport networks, control architecture have become the focus of research in recent years. In this thesis, I have developed a control plane architecture integrated with a set of algorithms, schemes and protocols. Based on the two-level inter-domain and intra-domain hierarchical control architecture, I proposed an inter-domain routing algorithm, ring-based topology aggregation scheme and modified RSVP-TE signalling protocol. This thesis aims to provide an efficient control plane architecture in terms of inter-domain routing, intra-domain routing, topology aggregation and signalling.

Orientadora : Profª. Drª. Aurora Pozo
Co-orientador : Prof. Dr. Richard Aderbal Gonçalves
Dissertação (mestrado) - Universidade Federal do Paraná, Setor de Ciências Exatas, Programa de Pós-Graduação em Informática. Defesa: Curitiba, 26/02/2016
Inclui referências : f. 64-70
Resumo: Muitos problemas de otimização do mundo real são complexos e possuem muitas variáveis e restrições. Por esta causa, o uso de meta-heurísticas tornou-se a principal maneira de resolver problemas com essas características. Uma das principais desvantagens do uso de meta-heurísticas e que são geralmente desenvolvidas utilizando características do domínio fazendo com que sejam atreladas a ele dificultando sua utilização em outros problemas. Em buscas de algoritmos mais adaptáveis o conceito de hiper-heurísticas surgiu. Hiper- heurísticas são métodos de busca que visam solucionar problemas de otimização selecionando ou gerando heurísticas. Hiper-heurísticas de seleção escolhem uma boa heurística para ser aplicada a partir de um conjunto de heurísticas. O método de seleção e a principal peca de uma hiper-heurística de seleção tendo impacto fundamental em sua performance. Apesar de existirem vários trabalhos sobre hiper-heurísticas de seleção, ainda não existe consenso sobre como uma boa estratégia de seleção deve ser definida. Em busca de uma estratégia de seleção, algoritmos inspirados nos conceitos do problema Multi-Armed Bandit (MAB) serão estudados. Estes algoritmos foram aplicados ao contexto da Seleção Adaptativa de Operadores obtendo resultados promissores. Entretanto, ainda existem poucas abordagens para o contexto de hiper-heurísticas. Nesta dissertação propomos uma hiper-heurística que utiliza algoritmos MAB como sua estratégia de seleção. A abordagem proposta e desenvolvida utilizando o framework HyFlex, que foi proposto para facilitar a implementação e comparação de novas Hiper- heurísticas. Os parâmetros foram configurados através de um estudo empírico, e a melhor configuração encontrada foi comparada com os 10 primeiros colocados da competição CHeSC 2011. Os resultados obtidos foram bons e comparáveis com os das melhores abordagens da literatura. O algoritmo proposto alcançou a quarta colocação. Apesar dos bons resultados, os experimentos demonstram que a abordagem proposta sofre grande influencia dos parâmetros. Trabalhos futuros irão investigar formas de amenizar esta influência.
Abstract: Many real word optimization problems are very complex with many variables and constraints, and cannot be solved by exact methods in a reasonable computational time. As an alternative, meta-heuristics emerged as an efficient way to solve this type of problems even though they cannot ensure optimal values. The main issue of meta-heuristics is that they are built using domain-specific knowledge, therefore they require a great effort to be used in a new domain. In order to solve this problem, the concept of Hyper-heuristics were proposed. Hyper-heuristics are search methods that aim to solve optimization problems by selecting or generating heuristics. Selection hyper-heuristics choose from a pool of heuristics a good one to be applied at the current stage of the optimization process. The selection mechanism is the main part of a selection hyper-heuristic and has a great impact on its performance. Although there are several works focused on selection hyperheuristics, there is no unanimity about which is the best way to define a selection strategy. In this dissertation, a deterministic selection strategy based on the concepts of the MultiArmed Bandit (MAB) problem is proposed to cross-domain optimization. Multi-armed bandit approaches define a selection function with two components, the first is based on the performance of an operator and the second based on the number of times that the operator was used. These approaches had showed a promising performance over the Adaptive Operator Selection context. However, there are few works on literature that aim the hyper-heuristic context, as proposed here. The proposed approach is integrated into the HyFlex framework, that was developed to facilitate the implementation and comparison of hyper-heuristics. An empirical parameter configuration was performed and the best setup was compared to the top ten CHeSC 2011 algorithms using the same methodology adopted during the competition. The results obtained were good comparable to those attained by the literature. Moreover, it was concluded that the behavior of MAB selection is heavily affected by its parameters. As this is not a desirable behavior to hyper-heuristics, future research will investigate ways to better deal with the parameter setting.

Thesis (Ph. D.)--Massachusetts Institute of Technology, Engineering Systems Division, 2010.
Cataloged from PDF version of thesis.
Includes bibliographical references (p. 379-385).
Processes in manufacturing, services, and healthcare are complex socio-technical systems composed of intricately sequenced activities supported by elements drawn from multiple domains. While many of these processes offer high performance, their complexity can make their design, improvement, troubleshooting, and change difficult due to the many possible and unforeseen interactions between elements. This thesis develops a design methodology and multi-domain network model for complex process design, change management, process improvement, and troubleshooting. As part of the methodology a feasibility analysis method based on solving the minimum cost flow problem for a network of process alternatives is presented to identify feasible processes subject to stakeholder requirements and constraints including performance, flexibility, modularity, and other system properties. A model based on Multi-Domain Matrix (MDM) concepts is developed specifically for process analysis called the Multi-Domain Process Matrix model (MDPM) to enumerate and analyze the interactions between process elements such that process performance under change and troubleshooting scenarios can be improved. The graph theory basis of the MDPM model enables its analysis using a proposed set of metrics derived from communications, social, and process network literature. As a demonstration of the use of the methodology, a complex DNA sequencing based surveillance process for Methicillin resistant Staphylococcus aureus (MRSA) in the US healthcare system is designed and a prototype implemented. Rapid advances in DNA-based technologies have greatly expanded the range of processes available to the clinical microbiology laboratory, however, integrating these new processes into a comprehensive surveillance system presents significant challenges. Many of these new technologies are still in early stages of development, require multidisciplinary teams to support them, and must undergo significant optimization presenting significant barriers to their rapid adoption despite the pressing need to understand and control antibiotic resistance. Data from the prototype MRSA surveillance process show significant variation at the DNA level between patient cases, providing evidence for the urgent need for a DNA sequencing based microbial surveillance process as part of clinical microbiology efforts in the US healthcare system. However, results of applying the process design methodology and MDPM model analysis indicate significant work remains to reduce complexity, further improve key technology elements, gain acceptance, develop key organizational infrastructure, and Page 3 of 385 redesign the process to efficiently absorb the rapid technology change expected in DNA sequencing. The MDPM model is used to develop a roadmap of specific multi-domain projects addressing these issues to accelerate the deployment of a national DNA sequencing based surveillance system.
by Robert Nicol.
Ph.D.

Multi-functional RNA binding proteins regulate and coordinate the many steps of RNA metabolism. Accurate functioning of these processes is vital in cells and misregulation has been linked to many human diseases. RNA binding proteins contain multiple RNA binding domains. The ability to perform multiple functions depends on the recognition of a diverse range of targets and domains are used combinatorially to achieve this. In this thesis I ask how the sequence specificity of low affinity RNA-binding domains and the interplay between said domains plays a role in RNA target selectivity. Within this question I focus on three proteins; TUT4, a uridyl transferase involved in the regulation of both non-coding RNAs and histone mRNA; FMRP, a translational repressor whose loss in cells is the cause of Fragile X Syndrome; and RBM10 a regulator of alternative splicing and miRNA biogenesis. I found that through the use of separate RNA binding domains both TUT4 and RBM10 are able to exert flexibility in target recognition; TUT4 by using two CCHC-type zinc fingers, working independently to recognise short RNA stretches; and RBM10 by using different subsets of domains to recognise either specific high affinity splice site sequences or pre-miRNAs. In FMRP the determination of the sequence specificity of KH1 allowed us to isolate its contribution to target selection. In a secondary objective, looking at methodologies used in RNA-protein interaction, SIA was improved to make it both less laborious and to reduce the sample requirements, and with FMRP a novel mutational strategy was used in combination with SIA to determine the sequence specificity of this low affinity domain. In summary these data extend our understanding of the RNA binding mechanisms of the three systems studied and introduces improved or novel methodologies to the future study of protein-RNA interactions.

Thesis (M. Phil.)--Hong Kong University of Science and Technology, 2004.
Includes bibliographical references (leaves 82-89). Also available in electronic version. Access restricted to campus users.

Thesis: S.M., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2019
Cataloged from PDF version of thesis.
Includes bibliographical references (pages 35-37).
We propose a mixture-of-experts approach for unsupervised domain adaptation from multiple sources. The key idea is to explicitly capture the relationship between a target example and different source domains. This relationship, expressed by a point-to-set metric, determines how to combine predictors trained on various domains. The metric is learned in an unsupervised fashion using meta-training. Experimental results on sentiment analysis and part-of-speech tagging demonstrate that our approach consistently outperforms multiple baselines and can robustly handle negative transfer.
by Darsh J. Shah.
S.M.
S.M. Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science

With a triple-bottom-line lens on sustainability, this study examines the effects of culture on companies' economic, social, and environmental sustainability practices. Drawing on institutional theory and project GLOBE, we delineate cultural practices dimensions that consistently predict sustainability practices related to each of the three domains. Based on a sample of 1924 companies in 36 countries and nine cultural clusters, we find that future orientation, gender egalitarianism, uncertainty avoidance, and power distance practices positively, and performance orientation practices negatively, predict corporate sustainability practices. Further, our findings suggest that these effects might vary according to the country vis-à-vis cluster level of analysis.

Unsupervised image-to-image translation (UNIT) is a challenging task in the image manipulation field, where input images in a visual domain are mapped into another domain with desired visual patterns (also called styles). An ideal direction in this field is to build a model that can map an input image in a domain to multiple target domains and generate diverse outputs in each target domain, which is termed as multi-domain and multi-modal unsupervised image-to-image translation (MMUIT). Recent studies have shown remarkable results in UNIT but they suffer from four main limitations: (1) State-of-the-art UNIT methods are either built from several two-domain mappings that are required to be learned independently or they generate low-diversity results, a phenomenon also known as model collapse. (2) Most of the manipulation is with the assistance of visual maps or digital labels without exploring natural languages, which could be more scalable and flexible in practice. (3) In an MMUIT system, the style latent space is usually disentangled between every two image domains. While interpolations within domains are smooth, interpolations between two different domains often result in unrealistic images with artifacts when interpolating between two randomly sampled style representations from two different domains. Improving the smoothness of the style latent space can lead to gradual interpolations between any two style latent representations even between any two domains. (4) It is expensive to train MMUIT models from scratch at high resolution. Interpreting the latent space of pre-trained unconditional GANs can achieve pretty good image translations, especially high-quality synthesized images (e.g., 1024x1024 resolution). However, few works explore building an MMUIT system with such pre-trained GANs. In this thesis, we focus on these vital issues and propose several techniques for building better MMUIT systems. First, we base on the content-style disentangled framework and propose to fit the style latent space with Gaussian Mixture Models (GMMs). It allows a well-trained network using a shared disentangled style latent space to model multi-domain translations. Meanwhile, we can randomly sample different style representations from a Gaussian component or use a reference image for style transfer. Second, we show how the GMM-modeled latent style space can be combined with a language model (e.g., a simple LSTM network) to manipulate multiple styles by using textual commands. Then, we not only propose easy-to-use constraints to improve the smoothness of the style latent space in MMUIT models, but also design a novel metric to quantitatively evaluate the smoothness of the style latent space. Finally, we build a new model to use pretrained unconditional GANs to do MMUIT tasks.

A key challenge facing designers creating innovative products is concept generation. Conceptual design is more effective when the design space is broadened by using an integrated design of product and material concepts approach. Conceptual design can also be accelerated by including problem solving and solution triggering tools in its structure. In this approach, structured analogy is used to transfer underlying principles from a solution suitable in one domain (i.e., product or mechanical domain) to an analogous solution in another domain (i.e., material domain). The nature of design analogy does not require as full of an exploration of the target domain as would otherwise be necessary; affording the possibility of a more rapid development. The addition of problem solving and solution triggering tools to a design method also decreases the design time and/or improves the quality of the final solution. This approach is formulated through a combination of the Theory of Inventive Problem Solving (TRIZ) proposed by Altshuller, and the systematic approach of Pahl and Beitz, for products that are jointly considered at the product and material level. These types of problems are ones where customer performance requirements are fulfilled through both the designed product and the designed material. The systematic approach of Pahl and Beitz is used as the base method through which TRIZ is used as a means of transferring abstract information about the design problem between the domains with the aim of accelerating conceptual design. This also allows for multi-domain design tools such as Su-Field-Model integration with design repositories for the transfer of information at different levels of abstraction; expanding the design space and effectively directing the designer. The explanation of this approach is presented through a simple example of a spring design improvement and validated through concept generation of a reactive material containment system.

Discrete Multitone is adopted in Digital Subscriber Line to offer high-speed data communication on the UTP channel. To combat the channel distortion, DMT system employed Time Domain Equalizer and Cyclic Prefix. In this thesis, we analyze different existing time domain equalizer design methods and their performances are compared based on our simulation results. Modification of the current method is proposed and significant performance improvement is obtained. Multi-rate equalization is studied theoretically, which enables us to achieve zero ISI channels that can be used not only in DMT system but also in all other communication systems. Our simulation results show that the new design method is superior and practical.

Advances in medical imaging have greatly aided in providing accurate and fast medical diagnosis, followed by recent deep learning developments enabling the efficient and cost-effective analysis of medical images. Among different image processing tasks, medical segmentation is one of the most crucial aspects because it provides the class, location, size, and shape of the subject of interest, which is invaluable and essential for diagnostics. Nevertheless, acquiring annotations for training data usually requires expensive manpower and specialised expertise, making supervised training difficult. To overcome these problems, unsupervised domain adaptation (UDA) has been adopted to bridge knowledge between different domains. Despite the appearance dissimilarities of different modalities such as MRI and CT, researchers have concluded that structural features of the same anatomy are universal across modalities, which unfolded the study of multi-modality image segmentation with UDA methods. The traditional UDA research tackled the domain shift problem by minimising the distance of the source and target distributions in latent spaces with the help of advanced mathematics. However, with the recent development of the generative adversarial network (GAN), the adversarial UDA methods have shown outstanding performance by producing synthetic images to mitigate the domain gap in training a segmentation network for the target domain. Most existing studies focus on modifying the network architecture, but few investigate the generative adversarial training strategy. Inspired by the recent success of state-of-the-art data augmentation techniques in classification tasks, we designed a novel mix-up strategy to assist GAN training for the better synthesis of structural details, consequently leading to better segmentation results. In this thesis, we propose SynthMix, an add-on module with a natural yet effective training policy that can promote synthetic quality without altering the network architecture. SynthMix is a mix-up synthesis scheme designed for integration with the adversarial logic of GAN networks. Traditional GAN approaches judge an image as a whole which could be easily dominated by discriminative features, resulting in little improvement of delicate structures in synthesis. In contrast, SynthMix uses the data augmentation technique to reinforce detail transformation at local regions. Specifically, it coherently mixes up aligned images of real and synthetic samples at local regions to stimulate the generation of fine-grained features examined by an associated inspector for domain-specific details. We evaluated our method on two segmentation benchmarks among three publicly available datasets. Our method showed a significant performance gain compared with existing state-of-the-art approaches.

La tendance à la numérisation de nombreux secteurs industriels tels que l’automobile, l’agriculture, les transports, la gestion urbaine, etc. révèle la nécessité de nouveaux usages des services de communication point-à-multipoint, tels que la fourniture massive de mises à jour logicielles et livraison fiable de messages d’alerte à la population, etc. D’un autre côté, la mise en logiciel des réseaux de nouvelle génération, avec notamment l’adoption croissante des réseaux définis par logiciel (SDN), apporte au réseau la flexibilité et les capacités de programmation permettant de prendre en charge des services de distribution point-à-multipoint de manière rentable. Cette thèse contribue au problème général de la fourniture de services de communication point-à-multipoint avec des exigences de qualité de service (QoS) dans un réseau SDN multi-domaines. Il considère également que certains des domaines sont des réseaux multi-sauts sans-fil. Tout d’abord, une technique d’agrégation de topologie de domaine basée sur une arborescence de Steiner, combinée à un algorithme heuristique d’allocation de ressources, sont proposés pour prendre en charge des services point-à-multipoint couvrant plusieurs domaines. Ensuite, un service de découverte de topologie générique est proposé pour les réseaux multi-sauts sans-fil basés sur le SDN afin de permettre au contrôleur SDN de créer et de gérer une vue complète du réseau avec divers attributs de nœud et de liaison sans-fil. Le contrôleur peut alors exposer des vues personnalisées aux applications de contrôle du réseau, telles que, par exemple, l’application en charge de la fourniture de services point-à-multipoint sur un réseau multi-sauts sans-fil basé sur le paradigme SDN. Un algorithme basé sur la programmation linéaire en nombres entiers et un algorithme génétique sont également proposés pour l’allocation de liens virtuels point-à-multipoint sur un réseau sans-fil multi-radio, multi-canaux et multi-sauts basé sur SDN. Enfin, pour traiter le cas des services dynamiques point-à-multipoint, nous proposons un schéma de réallocation de ressources qui répond aux exigences changeantes tout en réduisant les interuptions de service
The movement towards the digitalization of many industry sectors such as automotive, agriculture, transportation, city management, etc. is revealing the need for novel usages of point-to-multipoint network delivery services, such as massive delivery of software updates to objects, secure and reliable delivery of alert messages to population, etc. On another side, the softwarization of next generation networks, with amongst, the increasing adoption of Software Defined Networks (SDN) is bringing to the network the flexibility and programming capabilities that enable the support of point-to-multipoint delivery services in an efficient and cost-effective way. This PhD work contributes to the general problem of providing point-tomultipoint delivery services with Quality of Service (QoS) requirements in a multidomain SDN network. It also considers that some of the domains are wireless multi hop networks. First, a Steiner tree based network domain topology aggregation combined with a resource allocation heuristic algorithm is proposed to support point-to-multipoint delivery services that span multiple domains. Then, a generic topology discovery service is proposed for SDN based wireless multi-hop networks to let the SDN controller build and maintain a comprehensive view of the network with various node and wireless link attributes. From there, customized views can be exposed by the controller to network control applications, as, for instance, the application in charge of provisioning point-to-multipoint services on a SDN based wireless multi-hop network. An Integer linear programming based algorithm and a genetic algorithm are also proposed for the embedding of point-to-multipoint services on a SDN based multi-radio, multi-channel and multi-hop wireless network. Last, to address the case of dynamic point-to-multipoint services, we propose a resource reallocation scheme that meets the changing requirements while reducing service disruption

Power converter has played a very important role in modern electric power systems. The control of power converters is necessary to achieve high performance. In this study, a dc-dc buck converter is studied. The parameters of a notional proportional-integral controller are to be selected. Genetic algorithms (GAs), which have been widely used to solve multi-objective optimization problems, is used in order to locate appropriate controller design. The control metrics are specified as phase margin in frequency domain and voltage error in time-domain. GAs presented the optimal tradeoffs between these two objectives. Three candidate control designs are studied in simulation and experimentally. There is some agreement between the experimental results and the simulation results, but there are also some discrepancies due to model error. Overall, the use of multi-domain, multi-objective-optimization-based approach has proven feasible.

Collaboration between multiple organizations creates new opportunities for businesses. With such collaborations becoming a reality, it is necessary to have an access control policy integration approach to form a global policy consistent with the partner organizations. Research on policy integration has led to the proposal of several frameworks to uniformly express policies and to integrate such policies. But most of these frameworks are complex and compromise the privacy of the constituent domains by sharing all the components of an access control policy including access control lists. In this thesis, a unique policy integration technique is described to merge Role-Based Access Control (RBAC) policies of multiple-security domains in a heterogeneous environment. The proposed mechanism uses user credentials associated with roles as the main criteria in mapping inter-domain roles. Integration of the proposed policy greatly minimizes the administration overhead while efficiently merging the policies in a heterogeneous environment. Then, an approach to extend the community-based authorization framework to include the proposed integration tool is presented. A practical implementation is provided that enables collaboration among autonomous domains. Keywords. Policy Integration, Role-Based Access Control (RBAC), Community Authorization Service (CAS)

In the paper, the parallelization of multi-grid methods for solving second-order elliptic boundary value problems in two-dimensional domains is discussed. The parallelization strategy is based on a non-overlapping domain decomposition data structure such that the algorithm is well-suited for an implementation on a parallel machine with MIMD architecture. For getting an algorithm with a good paral- lel performance it is necessary to have as few communication as possible between the processors. In our implementation, communication is only needed within the smoothing procedures and the coarse-grid solver. The interpolation and restriction procedures can be performed without any communication. New variants of smoothers of Gauss-Seidel type having the same communication cost as Jacobi smoothers are presented. For solving the coarse-grid systems iterative methods are proposed that are applied to the corresponding Schur complement system. Three numerical examples, namely a Poisson equation, a magnetic field problem, and a plane linear elasticity problem, demonstrate the efficiency of the parallel multi- grid algorithm.

Thesis (S.M.)--Massachusetts Institute of Technology, Dept. of Electrical Engineering and Computer Science, 2004.
Includes bibliographical references (p. 65-66).
In recent years, mixed-signal designs have become more pervasive, due to their efficient use of area and power. Unfortunately, with sensitive analog and fast digital circuits sharing a common, non-ideal substrate, such designs carry the additional design burden of electromagnetic coupling between contacts. This thesis presents a method that quickly extracts the electroquasistatic coupling resistances between contacts on a planar, rectangular, two-layer lossy substrate, using an FFT-accelerated multi-domain surface integral formulation. The multi-domain surface integral formulation allows for multi-layered substrates, without meshing the volume. This method has the advantages of easy meshing, simple implementation, and FFT-accelerated iterative methods. Also, a three-dimensional variant of this method allows for more complex substrate geometries than some other surface integral techniques, such as multilayered Green's functions; this three-dimensional problem and its solution are presented in parallel with the planar substrate problem and solution. Results from a C++ implementation are presented for the planar problem.
by Anne M. Vithayathil.
S.M.

Thesis (Ph.D.)--Massachusetts Institute of Technology, Dept. of Electrical Engineering and Computer Science, 2001.
Includes bibliographical references (p. 201-208).
In developing telephone-based conversational systems, we foresee future systems capable of supporting multiple domains and flexible vocabulary. Users can pursue several topics of interest within a single telephone call, and the system is able to switch transparently among domains within a single dialog. This system is able to detect the presence of any out-of-vocabulary (OOV) words, and automatically hypothesizes each of their pronunciation, spelling and meaning. These can be confirmed with the user and the new words are subsequently incorporated into the recognizer lexicon for future use. This thesis will describe our work towards realizing such a vision, using a multi-stage architecture. Our work is focused on organizing the application of linguistic constraints in order to accommodate multiple domain topics and dynamic vocabulary at the spoken input. The philosophy is to exclusively apply below word-level linguistic knowledge at the initial stage. Such knowledge is domain-independent and general to all of the English language. Hence, this is broad enough to support any unknown words that may appear at the input, as well as input from several topic domains. At the same time, the initial pass narrows the search space for the next stage, where domain-specific knowledge that resides at the word-level or above is applied. In the second stage, we envision several parallel recognizers, each with higher order language models tailored specifically to its domain. A final decision algorithm selects a final hypothesis from the set of parallel recognizers.
(cont.) Part of our contribution is the development of a novel first stage which attempts to maximize linguistic constraints, using only below word-level information. The goals are to prevent sequences of unknown words from being pruned away prematurely while maintaining performance on in-vocabulary items, as well as reducing the search space for later stages. Our solution coordinates the application of various subword level knowledge sources. The recognizer lexicon is implemented with an inventory of linguistically motivated units called morphs, which are syllables augmented with spelling and word position. This first stage is designed to output a phonetic network so that we are not committed to the initial hypotheses. This adds robustness, as later stages can propose words directly from phones. To maximize performance on the first stage, much of our focus has centered on the integration of a set of hierarchical sublexical models into this first pass. To do this, we utilize the ANGIE framework which supports a trainable context-free grammar, and is designed to acquire subword-level and phonological information statistically. Its models can generalize knowledge about word structure, learned from in-vocabulary data, to previously unseen words. We explore methods for collapsing the ANGIE models into a finite-state transducer (FST) representation which enables these complex models to be efficiently integrated into recognition. The ANGIE-FST needs to encapsulate the hierarchical knowledge of ANGIE and replicate ANGIE's ability to support previously unobserved phonetic sequences ...
by Grace Chung.
Ph.D.

Dans cette thèse, nous présentons un intergiciel modulaire pour le calcul distribué dans les plateformes multi-domaine Grid/Cloud, qui permet le traitement de ces questions d’une façon extérieure aux applications. L’idée principale derrière cet intergiciel c’est d’offrir une infrastructure modulaire qui peut être composée hiérarchiquement selon la topologie des ressources et dynamiquement en fonction des ressources disponibles. Cet intergiciel fonctionne comme une colle entre les processus des applications en cours d’exécution dans des différents domaines administratifs, à travers les mécanismes tels que la communication point à point et collective en prenant en compte la topologie des réseaux. Cet intergiciel est basé sur GCM (Grid component model) et l’intergiciel ProActive, que nous l’avons amélioré avec des fonctionnalités telles que : des sémantiques de communication gathercast (Mx1) et multicast (Mx1) génériques, des interfaces gather-multicast (MxN) avec le support à la création des raccourcis pour la communication MxN directe entre les composants, le déploiement automatisé, le tunneling et le redirectionnement des communications. Tout au long de cette thèse, nous motivons notre travail en mettant en perspective deux plateformes multi-domaines hautement communicantes, que nous présentons comme des cas d’utilisation de notre intergiciel : une plateforme HPC, qui permet le couplage des applications basées sur la méthode de décomposition de domaine et une approche de programmation SPMD similaire à MP1, dans des environnement hétérogènes (Runtime DiscoGrid) et une fédération des Enterprise service Buses (ESB) à l’échelle de l’Internet qui permet à des ESBs indépendants d’être fédérées selon les relations de partenariat entre les fournisseurs-consommateurs de services. Les résultats expérimentaux obtenus dans le cadre des deux cas d’utilisation montrent que l’approche proposée est prometteuse, non seulement en termes d’approche de programmation, mais aussi en termes de performance
In this thesis, we introduce a modular middleware for multi-domain Grid and Cloud computing that allows the treatment of issues related to deployment, resources access and communication in heterogeneous networks externally to applications. The main idea behind this middleware is to offer a modular infrastructure that can be composed hierarchically, according to resources topology, and dynamically, according to the availability resources. This middleware works as a glue between application processes running in different domains, featuring mechanisms like topology-aware point-to-point and collective communication. Our middleware grounds from the GCM (the Grid Component Model) and the ProActive Grid middleware, that we improved with features, such as : generic gathergast (Mx1) and multigast (Mx1) communication semantics, gateher-multigast (MxN) component interfaces, MxN shortcuts, automated deployment and communication tunnelling and forwarding. All along this thesis, we motivate our work by putting in perspective two highly communicating multi-domain frameworks, which we present as us-cases of our middleware : an HPC runtime, which allow the coupling of domain-decomposition applications in heterogeneous environments through an MPI-like SPMD programming (the DicoGrid Runtime) and an Internet wide federation of Distributed Enterprise Service Buses, which allows independent distributed service buses to be federated according to partnership relations among service providers. Experimental results obtained in the context of both use-cases show that the proposed approach is promising, not only in terms of programming approach but also in terms of performance

Memoria para optar al título de Ingeniero Civil Mecánico
Con el uso de las redes neuronal profundas ganando terreno en el área de PHM, los sensores disminuyendo progresivamente su precio y mejores algoritmos, la falta de datos se ha vuelto un problema principal para los modelos enfocados en datos. Los datos etiquetados y aplicables a escenarios específicos son, en el mejor de los casos, escasos. El objetivo de este trabajo es desarrollar un método para diagnosticas el estado de un rodamiento en situaciones con datos limitados. Hoy en día la mayoría de las técnicas se enfocan en mejorar la precisión del diagnóstico y en estimar la vida útil remanente en componentes bien documentados. En el presente, los métodos actuales son ineficiente en escenarios con datos limitados. Se desarrolló un método en el cual las señales vibratorias son usadas para crear escalogramas y espectrogramas, los cuales a su vez se usan para entrenar redes neuronales generativas y de clasificación, en función de diagnosticar un set de datos parcial o totalmente desconocido, en base a uno conocido. Los resultados se comparan con un método más sencillo en el cual la red para clasificación es entrenada con el set de datos conocidos y usada directamente para diagnosticar el set de datos desconocido. El Case Western Reserve University Bearing Dataset y el Machine Failure Prevention Technology Bearing Dataset fueron usados como datos de entrada. Ambos sets se usaron como conocidos tanto como desconocidos. Para la clasificación una red neuronal convolucional (CNN por sus siglas en inglés) fue diseñada. Una red adversaria generativa (GAN por sus siglas en inglés) fue usada como red generativa. Esta red fue basada en una introducida en el paper StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation. Los resultados fueron favorables para la red CNN mientras que fueron -en general- desfavorables para la red GAN. El análisis de resultados sugiere que la función de costo es inapropiada para el problema propuesto. Las conclusiones dictaminan que la traducción imagen-a-imagen basada en la función ciclo no funciona correctamente en señal vibratorias para diagnóstico de rodamientos. With the use of deep neural networks gaining notoriety on the prognostics & health management field, sensors getting progressively cheaper and improved algorithms, the lack of data has become a major issue for data-driven models. Data which is labelled and applicable for specific scenarios is scarce at best. The purpose of this works is to develop a method to diagnose the health state of a bearing on limited data situations. Now a days most techniques focus on improving accuracy for diagnosis and estimating remaining useful life on well documented components. As it stands, current methods are ineffective on limited data scenarios. A method was developed were in vibration signals are used to create scalograms and spectrograms, which in turn are used to train generative and classification neural networks with the goal of diagnosing a partially or totally unknown dataset based on a fully labelled one. Results were compared to a simpler method in which a classification network is trained on the labelled dataset to diagnose the unknown dataset. As inputs the Case Western Reserve University Bearing Dataset (CWR) and the Society for Machine Failure Prevention Technology Bearing Dataset. Both datasets are used as labelled and unknown. For classification a Convolutional Neural Network (CNN) is designed. A Generative Adversarial Network (GAN) is used as generative model. The generative model is based of a previous paper called StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation. Results were favourable for the CNN network whilst generally negative for the GAN network. Result analysis suggests that the cost function is unsuitable for the proposed problem. Conclusions state that cycle based image-to-image translation does not work correctly on vibration signals for bearing diagnosis.

As distributed collaborative applications and architectures are adopting policy based management for tasks such as access control, network security and data privacy, the management and consolidation of a large number of policies is becoming a crucial component of such policy based systems. In large-scale distributed collaborative applications like web services, there is the need of analyzing policy interactions and integrating policies. In this thesis, we propose and implement EXAM-S, a comprehensive environment for policy analysis and management, which can be used to perform a variety of functions such as policy property analyses, policy similarity analysis, policy integration etc. As part of this environment, we have proposed and implemented new techniques for the analysis of policies that rely on a deep study of state of the art techniques. Moreover, we propose an approach for solving heterogeneity problems that usually arise when considering the analysis of policies belonging to different domains. Our work focuses on analysis of access control policies written in the dialect of XACML (Extensible Access Control Markup Language). We consider XACML policies because XACML is a rich language which can represent many policies of interest to real world applications and is gaining widespread adoption in the industry.

As distributed collaborative applications and architectures are adopting policy based management for tasks such as access control, network security and data privacy, the management and consolidation of a large number of policies is becoming a crucial component of such policy based systems. In large-scale distributed collaborative applications like web services, there is the need of analyzing policy interactions and integrating policies. In this thesis, we propose and implement EXAM-S, a comprehensive environment for policy analysis and management, which can be used to perform a variety of functions such as policy property analyses, policy similarity analysis, policy integration etc. As part of this environment, we have proposed and implemented new techniques for the analysis of policies that rely on a deep study of state of the art techniques. Moreover, we propose an approach for solving heterogeneity problems that usually arise when considering the analysis of policies belonging to different domains. Our work focuses on analysis of access control policies written in the dialect of XACML (Extensible Access Control Markup Language). We consider XACML policies because XACML is a rich language which can represent many policies of interest to real world applications and is gaining widespread adoption in the industry.

Abstract This thesis focuses on advanced multi-antenna receiver and transmission techniques to improve the utilization efficiencies of radio resources in broadband single carrier communications. Special focus is devoted to the development of computationally efficient frequency domain (FD) turbo equalization techniques for single and multiuser MIMO frequency selective channels. Another special emphasis is given to transmission power optimization for single user MIMO communications, which takes into account the convergence properties of the iterative equalizer. A new iterative FD soft cancellation (SC) and minimum mean square error (MMSE) filtering based joint-over-antenna (JA) multiuser MIMO signal detection technique for multiuser MIMO uplink transmission in frequency-selective channels is proposed. The proposed FD multiuser MIMO detection technique requires significantly lower computational complexity than its time-domain counterpart. Furthermore, significant performance gains can be achieved with the proposed JA turbo receiver compared to an antenna-by-antenna (AA) turbo receiver when the total number of transmitter antennas and users is larger than the number of receiver antennas, as well as in the presence of spatial correlation. The impact of existing linear precoding techniques, e.g, maximum information rate (MaxRate) and minimum sum mean square error (MinSumMSE), on the performance of frequency domain turbo equalization is investigated by utilizing extrinsic information transfer (EXIT) chart analysis. A novel transmission power minimization framework based on an EXIT analysis of single carrier MIMO transmission with iterative FD SC-MMSE equalization is then proposed. The proposed optimization framework explicitly takes into account the convergence properties of the iterative equalizer. The proposed convergence constrained power allocation (CCPA) technique decouples the spatial interference between streams using singular value decomposition (SVD), and minimizes the transmission power while achieving the mutual information target for each stream after iterations at the receiver side. The transmission power allocation can be formulated as a convex optimization problem. A special case having only two mutual information constraints is considered, for which the Lagrange dual function is derived and its dual problem is solved. Inspired by the Lagrange duality, two CCPA based heuristic schemes are developed. The numerical results demonstrate that the proposed CCPA schemes outperform the existing power allocation schemes
Tiivistelmä Tässä työssä tutkitaan edistyksellisten moniantennivastaanotto- ja lähetysmenetelmien käyttöä radioresurssien tehokkuuden parantamiseen laajakaistaisessa yhden kantoaallon kommunikaatiossa. Työssä keskitytään erityisesti laskennallisesti tehokkaiden taajuustasossa suoritettavien iteratiivisten kanavakorjaintekniikoiden kehittämiseen yhden ja usean käyttäjän multiple-input multiple-output (MIMO) -kommunikaatiossa taajuusselektiivisen radiokanavan yli. Toinen tutkimuksen painopiste on lähetystehon optimointi yhden käyttäjän MIMO-kommunikaatiossa, jossa iteratiivisen kanavakorjaimen konvergenssiominaisuudet otetaan huomioon. Työssä ehdotetaan uudenlaista iteratiivista taajuustasossa suoritettavaa soft-cancellation (SC) ja minimum mean square error (MMSE) -suodatukseen pohjautuvaa joint-over-antenna (JA) monen käyttäjän ilmaisumenetelmää nousevan siirtokanavan tiedonsiirtoon taajuusselektiivisessa radiokanavassa. Ehdotettu tajuustasossa suoritettava usean käyttäjän MIMO-lähetyksen ilmaisumenetelmä vaatii selvästi vähemmän laskentatehoa verrattuna aikatason menetelmään. Tämän lisäksi ehdotetulla menetelmällä voidaan saavuttaa merkittävää suorituskykyhyötyä verrattuna antenna-by-antenna (AA) -pohjaiseen iteratiiviseen vastaanottimeen kun lähetysantennien ja käyttäjien kokonaislukumäärä on suurempi kuin vastaanotinantennien. Suorituskykyhyöty pätee myös tilakorrelaation tapauksessa. Työssä tutkitaan lisäksi olemassa olevien lineaaristen esikoodaustekniikoiden, esim. maximum information rate (MaxRate) and minimum sum mean square error (MinSumMSE), vaikutusta taajuustasossa suoritettavaan iteratiivisen kanavakorjaimen konvergenssiominaisuuksiin xtrinsic information transfer (EXIT) -analyysin avulla. Työssä ehdotetaan uudenlaista EXIT-analyysi-pohjaista lähetystehon minimointimenetelmää yhden kantoaallon MIMO-lähetykseen käyttäen iteratiivista taajuustason SC-MMSE-kanavakorjainta. Menetelmä ottaa huomioon iteratiivisen kanavakorjaimen konvergenssiominaisuudet. Ehdotettu convergence constrained power allocation (CCPA) -menetelmä erottaa tilatason häiriön lähetteiden välillä hyödyntäen singular value decomposition (SVD) -tekniikkaa ja minimoi lähetystehon ja saavuttaa samalla keskinäisinformaatiotavoitteet jokaiselle lähetteelle iteraatioiden jälkeen vastaanottimessa. Lähetystehon minimointiongelma voidaan muotoilla konveksiksi optimointiongelmaksi. Kahden keskinäisinformaatiorajoitteen erityistapaukselle johdetaan Lagrangen duaalifunktio ja ratkaistaan sen duaalifunktio. Työssä kehitetään lisäksi kaksi CCPA-pohjaista heuristista menetelmää. Numeeriset tulokset osoittavat ehdotettujen CCPA-pohjaisten menetelmien suoriutuvan paremmin verrattuna olemassa oleviin menetelmiin

Fire dynamics is a complex process involving multi-mode heat transfer, reacting fluid flow, and the reaction of combustible materials. High-fidelity predictions of fire behavior using computational fluid dynamics (CFD) models come at a significant computational cost where simulation times are often measured in hours, days, or even weeks. A new simulation method is to use a machine learning approach which uses artificial neural networks (ANNs) to represent underlying connections between data to make predictions of new inputs. The field of image analysis has seen significant advancements in ANN performance by using feature based layers in the network architecture. Inspired by these advancements, a generalized procedure to design ANNs to make spatially resolved predictions in multi-physics applications is presented and applied to different fire applications. A deep convolutional inverse graphics network (DCIGN) was developed to predict the two-dimensional spatially resolved spread of a wildland fire. The network uses an image stack corresponding to the spatially resolved landscape, weather, and current fire perimeter (which can be obtained from measurements) to predict the fire perimeter six hours in the future. A transpose convolutional neural network (TCNN) was developed to predict the spatially resolved thermal flow field in a compartment fire from coarse zone fire model predictions. The network uses thirty-five parameters describing the geometry of the room and the ventilation conditions to predict the full-field temperature and velocity throughout the room. The data for use in training and testing both networks was generated using high-fidelity CFD fire simulations. Overall, the ANN predictions in each network agree with simulation predictions for validation scenarios. The computational time to evaluate the ANNs is 10,000x faster than the high-fidelity fire simulations. This work represents a first step in developing super real-time full-field fire predictions for different applications.
Ph. D.
The National Fire Protection Agency estimates the total cost of fire in the United States at $300 billion annually. In 2017 alone, there were 3,400 civilian fire fatalities, 14,670 civilian fire injuries, and an estimated $23 billion direct property loss in the United States. Large scale fires in the wildland urban interface (WUI) and in large buildings still represent a significant hazard to life, property, and the environment. Researchers and fire safety engineers often use computer simulations to predict the behavior of a fire to assist in reducing the hazard of fire. Unfortunately, typical simulations of fire scenarios may take hours, days, or even weeks to run which limits their use to small areas or sections of buildings. A new method is to use a machine learning approach which uses artificial neural networks (ANNs) to represent underlying connections between data to make new predictions of fire behavior. Inspired by advancements in the field of image processing, this research developed a procedure to use machine learning to make rapid high resolution predictions of fire behavior. An ANN was developed to predict the perimeter of a wildland fire six hours in the future based on a set of images corresponding to the landscape, weather, and current fire perimeter, all of which can be obtained directly from measurements (US Geological Survey, Automated Surface Observation System, and satellites). In addition, an ANN was developed to predict high-resolution temperature and velocity fields within a floor of a building based on predictions from a coarse model. The data for use in training and testing these networks was generated using high-resolution fire simulations. Overall, the network predictions agree well with simulation predictions for new scenarios. In addition, the time to run the model is 10,000x faster than the typical simulations. The work presented herein represents a first step in developing high resolution computer simulations for different fire scenarios that run very quickly.

This dissertation presents the theory, design, implementation and successful deployment of a cognitive engine decision algorithm by which a cognitive radio-equipped mobile robot may adapt its motion and radio parameters through multi-objective optimization. This provides a proof-of-concept prototype cognitive system that is aware of its envirionment, its userâ •s needs, and the rules governing its operation. It is to take intelligent action based on this awareness to optimize its performance across both the mobility and radio domains while learning from experience and responding intelligently to ongoing environmental mission changes. The prototype combines the key features of cognitive radios and autonomous vehicles into a single package whose behavior integrates the essential features of both.

The use case for this research is a scenario where a small unmanned aerial vehicle (UAV) is traversing a nominally cyclic or repeating flight path (an â •orbitâ •) seeking to observe targets and where possible avoid hostile agents. As the UAV traverses the path, it experiences varying RF effects, including multipath propagation and terrain shadowing. The goal is to provide the capability for the UAV to learn the flight path with respect both to motion and RF characteristics and modify radio parameters and flight characteristics proactively to optimize performance. Using sensor fusion techniques to develop situaitonal awareness, the UAV should be able to adapt its motion or communication based on knolwedge of (but not limited to) physical location, radio performance, and channel conditions. Using sensor information from RF and mobility domains, the UAV uses the mission objectives and its knowledge of the world to decide on a course of action. The UAV develops and executes a multi-domain action; action that crosses domains, such as changing RF power and increasing its speed.

This research is based on a simple observation, namely that cognitive radios and autonomous vehicles perform similar tasks, albeit in different domains. Both analyze their environment, make and execute a decision, evaluate the result (learn from experience), and repeat as required. This observation led directly to the creation of a single intelligent agent combining cognitive radio and autonomous vehicle intelligence with the ability to leverage flexibility in the radio frequency (RF) and motion domains. Using a single intelligent agent to optimize decision making across both mobility and radio domains is unified multi-domain decision making (UMDDM).
Ph. D.

In this thesis we investigate methods for deploying machine translation (MT) in real-world application scenarios related to the use of MT in computer assisted translation (CAT), where human translators post-edit MT outputs. In particular, we investigate (in chronological order) MT adaptation under two working conditions: single-domain and multi-domain. In the former, we assume that MT receives requests by a single user working on a single domain, while in the latter we assume the MT system to receive requests i) from multiple users working on different domains, ii) with no predefined order, and iii) without domain information. In the single-domain case, we first focus on word alignment, a core component of online adaptive phrase-based MT (PBMT) that is crucial for extracting features from a post-edited segment. In particular, we concentrate on improving word alignment in presence of out-of-vocabulary words observed in the source sentences or introduced by the post-editor. In the multi-domain scenario we turned our focus to the neural MT (NMT) paradigm. In particular, we introduce a scalable solution that adapts on-the-fly a generic NMT model to each incoming translation request. It relies on a procedure that locally fine-tunes the model to each input sentence using samples retrieved from a pool of parallel data. Our instance-based adaptation uses a more general formulation of the log-likelihood approach to control the contribution of relevant and irrelevant words during model update. Finally, we test our approach on a simulated continuous learning setting, where the system receives user feedback under form of post-editing.