We study the supply chain with two manufacturers producing competing products and selling them through a common retailer. The two manufacturers must decide on the wholesale price and the level of service they plan to provide to the consumer. Each firm are assumed to optimize only its own profit (uncoordinated). The consumer demand depends on two factors: (1) retail price, and (2) service level provided by the manufacturer. We extend the study on this basic model in three directions. First, we explore the role of bargaining power in supply chain strategic interactions. We derive and compare equilibrium solutions for the supply chain under three different scenarios (e.g., Manufacturer Stackelberg, Retailer Stackelberg, and Vertical Nash). Second, we extend the framework to study multi-period model. In this model, demand also depends on the past period retail prices and service levels, as well as current prices and service levels. Game-theoretic approaches and dynamic system and control theory are used as tools to model the problem. Finally, we examine a single period problem with stochastic demand. When demand is uncertain, the retailer faces a newsvendor-type problem. In our model, the newsvendor must manage two competing products against a price-dependent demand. We derive an expression for the newsvendor's optimal retail prices. Next, we provide an algorithm to search for the equilibrium wholesale price and service level, given that the manufacturers know the retailer's reaction function. Some numerical examples are provided.

E-learning is increasingly adopted to support face-to-face classroom-based learning or implemented as a complete standalone learning system. Its inherent adaptable nature and ability to provide learning anywhere, everywhere and anytime makes it a versatile tool for access to basic, professional and higher education. This research proposes and develops an adaptable e-learning approach, focusing on the learner's requirement specification and negotiation of course with multiple providers to improve online learning. This addresses issues of inflexible learning model, narrow coverage of subject domains in existing systems and ineffective use of educational resources, using design research methodology (DRM). The proposed Intelligent Learning approach provides learning support by applying collaborative and deliberative capabilities of software agents to e-learning systems. Designated learning support agents negotiate with providers on behalf of the learner for courses, matching specified requirements. This is achieved through agent negotiation strategies, devising dynamic learning plans (DPLAN) and online learning contract (or EContract) between the system and a range of providers, to harness the changing needs of the learner, hence, providing an Adaptive Agent Learner Plan (ADALP) approach. It develops and applies a 'Basic Requirements Learning' model, addressing specific learning objectives, supported by a two way evaluation process that enforces learning flexibility, empowering learners and accommodating a wide spectrum of learning needs. Unlike traditional Intelligent Tutoring System (ITS), learning objectives are not fixed and are constituted dynamically from learner specifications. The ADALP approach provides multiple provider support options, generating learner feedback for goal oriented, but flexible learning. This deviates from the traditional 'top-down' approach, where instructors and designers create fixed models of different categories of learners and their needs. The prototype of multi-agent system (MAS) demonstrates contributions of the approach, applying Multi-issue-Negotiation and Contracting Courses with Multiple Providers; devising dynamic personalised learning plans and learning commitment (or e-contracts) between learners and providers. It implements designated agents which generate tasks and sub-tasks corresponding to the learners' goals and objectives; 'biding' for learning and tutoring resources from multiple providers to deliver on the derived tasks. Personalised learning plan aligned with online learning contract is generated for each learner based on the specified requirements and learning goals, as a result. It is argued that the ADALP approach empowers learners and improves on similar approaches, in comparison to existing adaptive learning systems.

The major objectives of this thesis involve the development of both dynamic and interactive applications aimed at complementing traditional engineering and science coursework, laboratory exercises, research, and providing users with easy access by publishing the applications on Wolframs Demonstration website. A number of applications have been carefully designed to meet cognitive demands as well as provide easy-to-use interactivity. Recent technology introduced by Wolfram Mathematica called CDF (Computable Document Format) provides a resource that gives ideas a communication pipeline in which technical content can be presented in an interactive format. This new and exciting technology has the potential to help students enhance depth and quality of understanding as well as provide teachers and researchers with methods to convey concepts at all levels. Our approach in helping students and researchers with teaching and understanding traditionally difficult concepts in science and engineering relies on the ability to use dynamic, interactive learning modules anywhere at any time. The strategy for developing these applications resulted in some excellent outcomes. A variety of different subjects were explored, which included; numerical integration, Green's functions and Duhamel's methods, chaotic maps, one-dimensional diffusion using numerical methods, and two-dimensional wave mechanics using analytical methods. The wide range of topics and fields of study give CDF technology a powerful edge in connecting with all types of learners through interactive learning. / Master of Science

Due to the recent phenomenon of globalization, the mobility of people has increased significantly. In a world where national and cultural borders are getting more blurred and undefinable, people from different parts of the world encounter individuals with different cultural backgrounds. Due to these different worldviews, perceptions and experiences, misunderstandings may arise when people engage in cross-cultural communication. This is true for recreational as well as professional cross-cultural communication. An individual who successfully interacts and communicates with people from other cultures possesses what has come to be known as cultural intelligence. In order to get a deeper understanding for how people perceive cross-cultural communication in the professional field, this study investigates how managers in the forestry based industry in Västerbotten have experienced communication with their international business partners. The investigation was conducted through personal interviews and a self-completion questionnaire was also distributed to make an attempt to measure the respondents’ cultural intelligence. The results point in the direction that cross-cultural communication is vital for organizations that operate on the global market. Both verbal communication and non-verbal communication do affect the collaboration, but there is no general answer to how. Every situation has to be dealt with in a unique way. This implies that the process of cultural learning and understanding is ongoing and dynamic.

Neste trabalho investigamos o aprendizado supervisionado on-line, com ênfase nas habilidades de generalização, de redes neurais feedforward. O estudo de algoritmos de aprendizagem ótimos, no sentido da generalização, é estendido para duas diferentes classes de arquiteturas: a máquina paridade com estrutura de árvore e K unidades escondidas, e o perceptron reversed wedge, uma máquina de uma camada com função de transferência não monotônica. O papel do ruído é de fundamental importância na teoria de aprendizagem. Neste trabalho estudamos os processos com ruído que podem ser parametrizados por uma única quantidade, o nível de ruído. No caso da máquina paridade analisamos o aprendizado na presença de ruído multiplicativo (na saída). O algoritmo ótimo é muito superior aos algoritmos de aprendizagem até então apresentados, como o algoritmo de mínima ação (LAA), como podemos ver, por exemplo, através do comportamento do erro de generalização que decai após a apresentação de p exemplos, com l/p ao invés de l/\'p POT. 1/3\' como no caso do LAA. Além deste fato, observa-se que não existe um nível de ruído crítico a partir do qual a rede não é capaz de generalizar, como ocorre no LAA. Além do ruído multiplicativo, no caso do perceptron reversed wedge consideramos também o ruído aditivo. Analisamos a função de modulação fornecida pelo algoritmo ótimo e as curvas de aprendizagem. A aprendizagem ótima requer o uso de parâmetros que usualmente não estão disponíveis. Neste caso estudamos a influência da utilização de uma estimativa do nível de ruído sobre as curvas de aprendizado. Estes resultados são apresentados na forma do que chamamos de diagrama de robustez, no espaço de nível de ruído real versus nível de ruído estimado. As linhas de transição deste diagrama definem regiões com comportamentos dinâmicos diferentes. Entre as propriedades mais interessantes encontradas, destacamos a universalidade do diagrama de robustez para ruído multiplicativo, uma vez que é exatamente o mesmo para a máquina paridade e comitê com estrutura de árvore, e para o perceptron reversed-wedge. Entretanto, esta universalidade não se estende para o caso de ruído aditivo, uma vez que, neste caso, os diagramas dependem da arquitetura em questão. / In this work online supervised learning is investigated with emphasis on the generalization abilities of feedforward neural networks. The study of optimal learning algorithms, in the sense of generalization, is extended to two different classes of architectures; the tree parity machine (PM) with K hidden units and the reverse wedge perceptron (RWP), a single layer machine with a non monotonic transfer function. The role of noise is of fundamental importance in learning theory, and we study noise processes which can be parametrized by a single quantity, the noise level. For the PM we analize learning in the presence of multiplicative or output noise. The optimal algorithm is far superior than previous learning algorithms, such as the Least Action Algorithm (LAA), since for example, the generalization error\'s decay is proportional to l /p instead of l/\'p POT. 1/3\' for the LAA, after p examples have been used for training. Furthermore there is no critical noise level, beyond which no generalization ability is attainable, as is the case for the LAA. For the RW perceptron in addition to multiplicative noise we also consider additive noise. The optimal algorithm modulation function and the learning curves are analized. Optimal learning requires using certain usually unavailable parameters. In this case, we study the influence that misevaluation of the noise levels has on the learning curves. The results are presented in terms of what we have called Robustness Phase Diagrams (RPD), in a space of real noise level against assumed noise level. The RPD boundary lines separate between different dynamical behaviours. Among the most interesting properties, we have found the universality of the RPD for multiplicative noise, since it is exactly the same for the PM, RWP and the tree committee machine. However this universality does not hold for the additive noise case, since RPD\'s are shown to be architecture dependent.

Neste trabalho investigamos o aprendizado supervisionado on-line, com ênfase nas habilidades de generalização, de redes neurais feedforward. O estudo de algoritmos de aprendizagem ótimos, no sentido da generalização, é estendido para duas diferentes classes de arquiteturas: a máquina paridade com estrutura de árvore e K unidades escondidas, e o perceptron reversed wedge, uma máquina de uma camada com função de transferência não monotônica. O papel do ruído é de fundamental importância na teoria de aprendizagem. Neste trabalho estudamos os processos com ruído que podem ser parametrizados por uma única quantidade, o nível de ruído. No caso da máquina paridade analisamos o aprendizado na presença de ruído multiplicativo (na saída). O algoritmo ótimo é muito superior aos algoritmos de aprendizagem até então apresentados, como o algoritmo de mínima ação (LAA), como podemos ver, por exemplo, através do comportamento do erro de generalização que decai após a apresentação de p exemplos, com l/p ao invés de l/\'p POT. 1/3\' como no caso do LAA. Além deste fato, observa-se que não existe um nível de ruído crítico a partir do qual a rede não é capaz de generalizar, como ocorre no LAA. Além do ruído multiplicativo, no caso do perceptron reversed wedge consideramos também o ruído aditivo. Analisamos a função de modulação fornecida pelo algoritmo ótimo e as curvas de aprendizagem. A aprendizagem ótima requer o uso de parâmetros que usualmente não estão disponíveis. Neste caso estudamos a influência da utilização de uma estimativa do nível de ruído sobre as curvas de aprendizado. Estes resultados são apresentados na forma do que chamamos de diagrama de robustez, no espaço de nível de ruído real versus nível de ruído estimado. As linhas de transição deste diagrama definem regiões com comportamentos dinâmicos diferentes. Entre as propriedades mais interessantes encontradas, destacamos a universalidade do diagrama de robustez para ruído multiplicativo, uma vez que é exatamente o mesmo para a máquina paridade e comitê com estrutura de árvore, e para o perceptron reversed-wedge. Entretanto, esta universalidade não se estende para o caso de ruído aditivo, uma vez que, neste caso, os diagramas dependem da arquitetura em questão. / In this work online supervised learning is investigated with emphasis on the generalization abilities of feedforward neural networks. The study of optimal learning algorithms, in the sense of generalization, is extended to two different classes of architectures; the tree parity machine (PM) with K hidden units and the reverse wedge perceptron (RWP), a single layer machine with a non monotonic transfer function. The role of noise is of fundamental importance in learning theory, and we study noise processes which can be parametrized by a single quantity, the noise level. For the PM we analize learning in the presence of multiplicative or output noise. The optimal algorithm is far superior than previous learning algorithms, such as the Least Action Algorithm (LAA), since for example, the generalization error\'s decay is proportional to l /p instead of l/\'p POT. 1/3\' for the LAA, after p examples have been used for training. Furthermore there is no critical noise level, beyond which no generalization ability is attainable, as is the case for the LAA. For the RW perceptron in addition to multiplicative noise we also consider additive noise. The optimal algorithm modulation function and the learning curves are analized. Optimal learning requires using certain usually unavailable parameters. In this case, we study the influence that misevaluation of the noise levels has on the learning curves. The results are presented in terms of what we have called Robustness Phase Diagrams (RPD), in a space of real noise level against assumed noise level. The RPD boundary lines separate between different dynamical behaviours. Among the most interesting properties, we have found the universality of the RPD for multiplicative noise, since it is exactly the same for the PM, RWP and the tree committee machine. However this universality does not hold for the additive noise case, since RPD\'s are shown to be architecture dependent.

Dans ce travail, nous proposons un système auto-organisé composé d'agents-robots contrôlés par une architecture de subsomption et des règles locales probabilistes de prises et de dépôts. Ces agents-robots sont capables, grâce au développement de leurs capacités cognitives de se créer une carte cognitive, d'apprendre plusieurs lieux buts et de planifier le retour vers ces buts. Bien que formellement l'algorithme ne permette pas à chaque agent de "mélanger ni de fusionner ou d'optimiser" plusieurs objectifs, nous montrerons que le système global est capable de réaliser une optimisation multi-objectifs. Particulièrement, la fusion de l'apprentissage local avec l'accumulation de décisions individuelles permet de faire émerger (i) des structures dans l'environnement et (ii) des dynamiques tel que le comportement de spécialisation ou les comportements que nous pouvons considérer comme étant "égoïstes" ou "altruistes". Nous montrons qu'un mécanisme d'imitation simple contribue à l'amélioration des performance de notre SMAC et de notre SMRC, à savoir l'optimisation de la durée pour découvrir des différentes ressources, le temps moyen de planification, le niveau global de satisfaction des agents et enfin le temps moyen de convergence vers une solution stable. Particulièrement, l'ajout d'une capacité d'imitation améliore la construction des cartes cognitives pour chaque agent et stimule le partage implicite des informations dans un environnement a priori inconnu. En effet, les découvertes individuelles peuvent avoir un effet au plan social et donc inclure l'apprentissage de nouveaux comportements au niveau individuel. Pour finir, nous comparons les propriétés émergentes de notre SMAC à un modèle mathématique basé sur la programmation linéaire (PL). Cette évaluation montre les bonnes performances de notre SMAC qui permet d'avoir des solutions proches des solutions de la PL pour un coût de calcul réduit. Dans une dernière série d'expériences, nous étudions notre système d'agrégation dans un environnement réel. Nous mettons en place un SMRC, composé par des robots qui sont capables d'effectuer les opérations de prise, de dépôt et de maintien. Nous montrons via les premiers tests d'agrégation que les résultats sont prometteurs. / In this work, we propose a self-organized system composed by agents-robots, controlled by a subsumption architecture with probabilistic local rules of deposits and taking. These agents-robots are able, thanks to the development of their cognitive abilities to create a cognitive map, to learn various goals' locations and to plan the return to these goals. Although formally the algorithm does not allow each agent to « mix or merge or optimize » several objectives, we show that the overall system is able to perform a multi-objective optimization. Specifically, the fusion of local learning with the accumulation of the individual decisions allows to emerge (i) structures in the environment and (ii) several dynamics such as specialization behavior or those that we can consider as « selfish » or « altruistic ». We show that the imitation strategy contributes to the improvement of the performance of our SMAC and our SMRC, namely the optimization of time to explore the various resources, the average planning time, the overall satisfaction level of agents and finally the the average time of convergence to a stable solution. Specifically, the addition of an imitation ability improves the construction of cognitive maps for each agent and stimulates the implicit sharing of informations in an unknown environment. Indeed, individual discoveries can affect the social level and therefore include learning new behaviors at the individual level. Finally, we compare the emergent properties of our SMAC with a mathematical model based on linear programming (LP). This evaluation shows the good performance of our SMAC which allows to obtain solutions close to the solution of the PL for a low cost of computation. In a final series of experiments, we study our aggregation system in a real environment. We set up a SMRC, composed by robots that are able to perform taking operations, deposits operations and refueling operations. We show through the first tests of aggregation that the results are promising.

這篇論文研究跨國投資與監理寬容下保險公司之破產問題，同時論文的相關內容簡述於論文第一章中。第二章研究考慮匯率可預測下對跨國投資人資產配置的影響，結果顯示匯率可預測性能有效的提升投資人期末財富。第三章考慮監理寬容下保險公司的破產問題，在美國破產保護法第11章的架構下，保險人與被保險人之權利義務關係，可利用巴黎式選擇權描述，同時建構保證給付指標來衡量不同監理干預準則，數值結果顯示過於寬鬆的監理準則將導致被保險人的財務損失。第四章探討監理寬容下保險安定基金保險費率問題，依照美國破產保護法第11章的架構，安定基金保費可簡化成2個巴黎式選擇權，結果顯示，當前台灣保險單定基金費率有偏低的情形，建議主管機關訂定安定基金費率時需更加謹慎小心。 / This thesis focuses on the international portfolio selection and the bankruptcy cost of the insurance company under regulatory forbearance. The main theme of this thesis is outlined in chapter 1, which also serves as an introduction to the three papers (appearing here as Chapter 2, Chapter 3 and Chapter 4) collected in this thesis. In the theme of the international portfolio selection, Chapter 2 investigates the investment behaviors when learning effect is considered. According to the exchange rate predictability, the investor updates his information and adjusts his portfolio allocation. Finally, the numerical results show that the learning mechanism significantly improves the terminal wealth. In the theme of the regulatory forbearance, Chapter 3 provides an illustration of the impact on the ruin cost due to regulatory forbearance. The concept of the U.S. Chapter 11 bankruptcy code is employed to determine regulatory forbearance. Throughout the framework of Parisian option, a quantitative index of regulatory forbearance called Guarantee Benefit Index (GBI) is developed. The GBI is used to evaluate the different supervisory intervention criteria i.e., relative and absolute intervention criteria. Finally, numerical analysis is performed to illustrate the influence of different financial factors and the intervention criteria. Another important issue in bankruptcy problem is discussed in Chapter 4, i.e., the cost of insurance guaranty fund. It is important to determine the cost of bankruptcy when the insolvent insurance company is took over by the government. Under the U.S. Chapter 11 bankruptcy code, the cost of guaranty fund can be determined through Parisian options. Results show that the current premium rates of Taiwan insurance guarantee fund are far from risk sensitive. Hence the results suggest the government should more prudent to face the bankruptcy problem in insurance industry.