O raciocínio realizado em agentes BDI envolve essencialmente manipular três estruturas de dados representando suas crenças, desejos e intenções. Crenças de agentes BDI tradicionais não representam incerteza, e podem ser expressas como um conjunto fechado de literais ground. As restrições que indicam se um dado desejo é viável e pode ser adotado como uma intenção em agentes BDI tradicionais podem ser representadas como expressões lógicas sobre crenças. Dado que Redes Bayesianas permitem que representem-se informações com incerteza probabilisticamente, agentes BDI bayesianos as empregam para suportar incerteza em suas crenças. Em agentes BDI bayesianos, crenças representadas em Redes Bayesianas referem-se a estados de variáveis de eventos, possuindo probabilidades dinâmicas individuais que referem-se à incerteza. Os processos the constituem o raciocínio neste modelo de agente requerem mudanças a fim de acomodar esta diferença. Dentre estes processos, este trabalho concentra-se especificamente na seleção de desejos. Uma estratégia prévia para seleção de desejos é baseada em aplicar um limiar a probabilidades de crenças. Entretanto, tal abordagem impede que um agente selecione desejos condicionados em crenças cujas probabilidades estejam abaixo de um certo limiar, mesmo que tais desejos pudessem ser atingidos caso fossem selecionados. Para lidar com esta limitação, desenvolvemos três abordagens alternativas para seleção de desejos sob incerteza: Ranking Probabilístico, Loteria Viciada e Seleção Multidesejos Aleatória com Viés. Probability Ranking seleciona um desejo usando uma lista de desejos ordenados em ordem decrescente de probabilidade de pré-condição. Loteria Viciada seleciona um desejo usando um valor numérico aleatório e intervalos numéricos – associados a desejos – proporcionais às probabilidades de suas pré-condições. Seleção Multidesejos Aleatória com Viés seleciona múltiplos desejos usando valores numéricos aleatórios e considerando as probabilidades de suas pré-condições. Apresentamos exemplos, incluindo o agente Vigia, assim como experimentos envolvendo este, para mostrar como essas abordagens permitem que um agente às vezes selecione desejos cujas crenças pré-condições possuem probabilidades muito baixas. / The reasoning performed in BDI agents essentially involves manipulating three data structures representing their beliefs, desires and intentions. Traditional BDI agents’ beliefs do not represent uncertainty, and may be expressed as a closed set of ground literals. The constraints that indicate whether a given desire is viable and passive to be adopted as an intention in traditional BDI agents may be represented as logical expressions over beliefs. Given that Bayesian Networks allow one to represent uncertain information probabilistically, Bayesian BDI agents employ Bayesian Networks to support uncertainty in their beliefs. In Bayesian BDI agents, beliefs represented in Bayesian Networks refer to states of event variables, holding individual dynamic probabilities that account for the uncertainty. The processes that constitute reasoning in this agent model require changes in order to accomodate this difference. Among these processes, this work is specifically concerned with desire selection. A previous strategy for desire selection is based on applying a threshold on belief probabilities. However, such an approach precludes an agent from selecting desires conditioned on beliefs with probabilities below a certain threshold, even if those desires could be achieved if they were selected. To address this limitation, we develop three alternative approaches to desire selection under uncertainty: Probability Ranking, Biased Lottery and Multi-Desire Biased Random Selection. Probability Ranking selects a desire using a list of desires sorted in decreasing order of precondition probability. Biased Lottery selects a desire using one random numeric value and desire-associated numeric intervals proportional to the probabilities of the desires’ preconditions. Multi-Desire Biased Random Selection selects multiple desires using random numeric values and considering the probabilities of their preconditions. We present examples, including theWatchman agent, as well as experiments involving the latter, to show how these approaches allow an agent to sometimes select desires whose belief preconditions have very low probabilities.

Le besoin de compétitivité des entreprises, dans un contexte économique mondialisé, repose sur l'amélioration de la qualité des produits et la réduction des coûts et du temps de mise sur le marché. Pour atteindre ces objectifs, la simulation numérique est couramment utilisée pour la conception de produits complexes et mobilise des expertises diverses. Les Plans d'Expériences Numériques (PEN) sont de plus en plus utilisés pour simuler les variabilités des propriétés et de l’environnement du produit. Un processus de PEN apporte des méthodes de planification et d'analyse d'un ensemble de simulations, pour mieux maîtriser les performances du produit. La problématique traitée repose sur deux points. D'une part, la définition d'un processus de PEN repose sur de nombreux choix et l'utilisation de méthodes complexes, nécessitant une expertise avancée. Cette définition est d'autant plus complexe que le modèle de simulation est complexe et coûteux à exécuter. D'autre part, l'utilisation de PEN conduit à une production de grands volumes de données en multipliant les simulations. Ces travaux portent sur l'obtention rapide de la configuration optimale du processus de PEN pour raccourcir la préparation et l’exécution d’un PEN. Ces travaux se sont orientés vers la réutilisation des connaissances en entreprise pour un système à base de connaissances, composé d'une ontologie spécifique, pour capitaliser et partager les connaissances, et d'un moteur d'inférences, basé sur les réseaux bayésiens, pour proposer aux concepteurs des configurations efficaces et innovantes. Cette proposition est illustrée par une application sur un produit industriel issue du secteur automobile. / In order to improve industrial competitiveness, product design relies more and more on numerical tools, such as numerical simulation, to develop better and cheaper products faster. Numerical Design of Experiments (NDOE) are more and more used to include variabilities during simulation processes, to design more robust, reliable and optimized product earlier in the product development process. Nevertheless, a NDOE process may be too expensive to be applied to a complex product, because of the high computational cost of the model and the high number of required experiments. Several methods exist to decrease this computational cost, but they required expert knowledge to be efficiently applied. In addition to that, NDoE process produces a large amount of data which must be managed. The aim of this research is to propose a solution to define, as fast as possible, an efficient NDoE process, which produce as much useful information as possible with a minimal number of simulations, for complex products. The objective is to shorten both process definition and execution steps. A knowledge-based system is proposed, based on a specific ontology and a bayesian network, to capitalise, share and reuse knowledge and data to predict the best NDoE process definition regarding to a new product. This system is validated on a product from automotive industry.

Dans un contexte de développement durable, les enjeux de la dernière phase du cycle de vie d'un système, la phase de retrait de service, se sont accrus ces dernières années. Les systèmes en fin de vie doivent être déconstruits afin d'être revalorisés pour répondre aux différentes exigences environnementales. Cette responsabilité incombe au concepteur qui doit définir le sous-système support de la phase de retrait de service : le système de déconstruction. Sa principale fonction est la réalisation de l'activité de déconstruction dans l'objectif de favoriser en aval le recyclage de la matière des constituants du système en fin de vie et/ou leur recyclage fonctionnel. Les stratégies de déconstruction doivent répondre à l'ensemble des problèmes de décision posés lors de la phase de retrait de service d'un système. Il s'agit notamment de sélectionner les constituants valorisables suivant des critères techniques, économiques et environnementaux puis de définir et optimiser le système de déconstruction permettant l'obtention de ces produits. La solution obtenue définie ce que nous avons appelé une trajectoire de déconstruction. Nos travaux portent sur la modélisation et l'optimisation de ces trajectoires. Nos développements s'articulent en quatre phases. Etat de l'art et démarche de définition d'une trajectoire. Dans cette phase, une structure de démarche de définition de trajectoires de déconstruction est proposée puis instrumentée. Les modèles généralement utilisés dans ce cadre sont de type déterministe et ne permettent pas de prendre en compte et de gérer les incertitudes inhérentes au processus de déconstruction (état dégradé du système en fin de vie et de ses constituants, demandes en produits issus de la déconstruction, dates de fin de vie des systèmes, …). Pour déterminer une solution robuste de déconstruction d'un système en fin de vie, l'aide à la décision proposée doit intégrer des incertitudes de nature diverse tout en facilitant leur gestion et leurs mises à jour. Incertitudes en déconstruction. Sur la base de ce constat, l'ensemble d'incertitudes couramment mises en jeu dans l'optimisation des trajectoires est identifié et caractérisé. Les méthodes probabilistes apparaissent comme des approches privilégiées pour intégrer ces incertitudes dans une démarche d'aide la décision. Les réseaux bayésiens et leur extension aux diagrammes d'influence sont proposés pour répondre à différents problèmes de décision posés lors de la définition d'une trajectoire de déconstruction. Ils servent de support au développement d'un outil d'aide à la décision. Modélisation de trajectoires de déconstruction : principes et approche statique d'optimisation. Après avoir présenté ses principes de modélisation, l'outil est développé dans une approche de détermination d'une trajectoire de déconstruction d'un système en fin de vie donné. La trajectoire obtenue fixe la profondeur de déconstruction, les options de revalorisation, les séquences et les modes de déconstruction suivant des critères économiques et environnementaux tout en permettant de gérer différents types d'incertitude. L'utilisation de critères économiques est ici privilégiée. Un exemple d'application sur un système aéronautique est développé pour illustrer les principes de modélisation. Approche dynamique pour l'optimisation d'une trajectoire de déconstruction. Le champ d'application de l'outil d'aide à la décision est étendu en intégrant une dimension temporelle à la modélisation du problème à l'aide des réseaux bayésiens dynamiques. Les trajectoires de déconstruction peuvent ainsi être établies sur des horizons couvrant les arrivées de plusieurs systèmes en fin de vie en présence d'incertitudes. Le modèle permet de déterminer des politiques de déconstruction pour chaque opération identifiée dans la trajectoire en fonction de différents paramètres liés à la gestion des demandes et des arrivées ou encore au processus d'obtention de ces produits. Le décideur peut ainsi adapter l'outil à différents contextes de détermination de trajectoire de déconstruction de systèmes en fin de vie. / In a sustainable development context, stakes of the last stage of system life cycle, the end-of-life stage, increase these last years. End-of-life systems have to be demanufactured in order to be valued and answer environmental requirements. The aim of disassembly strategies is to bring solutions to the whole decision problem put during the end-of-life stage of systems. In particular, decision maker have to select valuable products in function of technical, economical and environmental criteria and, then, design and optimise disassembly support system allowing generating these products. The solution determines what we call a disassembly trajectory and ours works deal with modelling and optimization of these trajectories. Definition steps of disassembly trajectories are proposed, structured and instrumented. Models that are generally used in this frame are determinist and do not allow taking into account and managing uncertainties that are inherent to disassembly process (degradation of products, demand for valuable product, systems end-of-life dates, ...). In order to determine a robust disassembly solution, decision aid has to integrate uncertainties from various origins while facilitating their management and their update. On the basis this observation, all the uncertainties involved in disassembly trajectory optimization are identified and characterized. Basing on Bayesian networks, the proposed tool is developed through a “static” approach of disassembly trajectory. Indeed, the obtained trajectory gives the disassembly level of the end-of-life system, recycling options, sequences and disassembly modes in function of economical criteria while allowing managing uncertainties. An application example on an aeronautical system is developed to illustrate the modelling method. The application field of the model is extended to take into account time dimension (dynamic approach) by using dynamic Bayesian networks. Trajectories can be defined on horizons that integrate several arrivals of end-of-life systems. Decision makers can so adapt the model to various contexts

L'objet de cette thèse est l'utilisation de données transcriptomiques actuelles dans le but d'en inférer un réseau de régulation génique. Ces données sont souvent complexes, et en particulier des données d'interventions peuvent être présente. L'utilisation de la théorie de la causalité permet d'utiliser ces interventions afin d'obtenir des réseaux causaux acycliques. Je questionne la notion d'acyclicité, puis en m'appuyant sur cette théorie, je propose plusieurs algorithmes et/ou améliorations à des techniques actuelles permettant d'utiliser ce type de données particulières. / The purpose of this thesis is the use of current transcriptomic data in order to infer a gene regulatory network. These data are often complex, and in particular intervention data may be present. The use of causality theory makes it possible to use these interventions to obtain acyclic causal networks. I question the notion of acyclicity, then based on this theory, I propose several algorithms and / or improvements to current techniques to use this type of data.

Artificial agents are usually designed to achieve specific goals. An agent's competency can be defined as its ability to accomplish its goals under different conditions. This thesis restricts attention to a specific type of goal, namely reaching a desired state without exceeding a tolerance threshold of undesirable events in a first-order Markov process. For such goals, the state-dependent competency for an agent can be defined as the probability of reaching the desired state without exceeding the threshold and within a time limit given an initial state. The thesis further defines total competency as the set of state-dependent competency relationships over all possible initial states. The thesis uses a Monte Carlo approach to establish a baseline for estimating state-dependent competency. The Monte Carlo approach (a) uses trajectories sampled from an agent behaving in the environment, and then (b) uses nonlinear regression over the trajectory samples to estimate the competency curve. The thesis further presents an equation demonstrating recurrent relations for total competency and an algorithm based on that equation for computing total competency whose worst case computation time grows quadratically with the size of the state space. Simple maze-based Markov chains show that the Monte Carlo approach to estimating the competency agrees with the results computed by the proposed algorithm. Lastly, the thesis explores a special case where there are multiple sequential atomic goals that make up a complex goal. The thesis models a set of sequential goals as a Bayesian network and presents an equation based on the chain rule for deriving the competency for the complex goal from the competency for atomic goals. Experiments for the canonical taxi problem with sequential goals show the correctness of the Bayesian network-based decomposition approach.

This master's thesis deals with possible applications of Bayesian networks. The theoretical part is mainly of mathematical nature. At first, we focus on general probability theory and later we move on to the theory of Bayesian networks and discuss approaches to inference and to model learning while providing explanations of pros and cons of these techniques. The practical part focuses on applications that demand learning a Bayesian network, both in terms of network parameters as well as structure. These applications include general benchmarks, usage of Bayesian networks for knowledge discovery regarding the causes of criminality and exploration of the possibility of using a Bayesian network as a spam filter.

In this study, a method is developed for modelling and classifying behaviour of maritime vessels in a piracy situation. Prior knowledge is used to construct a probabilistic graphical model of maritime vessel behaviour. This model is a novel variant of a dynamic Bayesian network (DBN), that extends the switching linear dynamic system (SLDS) to accommodate contextual information. A generative model and a classifier model are developed. The purpose of the generative model is to generate simulated data by modelling the behaviour of fishing vessels, transport vessels and pirate vessels in a maritime piracy situation. The vessels move, interact and perform various activities on a predefined map. A novel methodology for evaluating and optimising the generative model is proposed. This methodology can easily be adapted to other applications. The model is evaluated by comparing simulation results with 2011 pirate attack reports. The classifier model classifies maritime vessels into predefined categories according to their behaviour. The classification is performed by inferring the class of a vessel as a fishing, transport or pirate vessel class. The classification method is evaluated by classifying the data generated by the generative model and comparing it to the true classes of the simulated vessels. / Thesis (PhD)--University of Pretoria, 2014. / tm2015 / Electrical, Electronic and Computer Engineering / PhD / Unrestricted

The planning and implementation of effective water resources management strategies need an assessment of multiple (physical, environmental, and socio-economic) issues, and often requires new research in which knowledge of diverse disciplines are combined in a unified methodological and operational framework. Such integrative research to link different knowledge domains faces several practical challenges. The complexities are further compounded by multiple actors frequently with conflicting interests and multiple uncertainties about the consequences of potential management decisions. This thesis aims to overcome some of these challenges, and to demonstrate how new modeling approaches can provide successful integrative water resources research. It focuses on the development of new integrated modeling approaches which allow integration of not only physical processes but also socio-economic and environmental issues and uncertainties inherent in water resources systems. To achieve this goal, two new approaches are developed in this thesis. At first, a Bayesian network (BN)-based decision support tool is developed to conceptualize hydrological and socio-economic interaction for supporting management decisions of coupled groundwater-agricultural systems. The method demonstrates the value of combining different commonly used integrated modeling approaches. Coupled component models are applied to simulate the nonlinearity and feedbacks of strongly interacting groundwater-agricultural hydrosystems. Afterwards, a BN is used to integrate the coupled component model results with empirical knowledge and stakeholder inputs. In the second part of this thesis, a fuzzy-stochastic multiple criteria decision analysis tool is developed to systematically quantify both probabilistic and fuzzy uncertainties associated with complex hydrosystems management. It integrates physical process-based models, fuzzy logic, expert involvement and stochastic simulation within a general framework. Subsequently, the proposed new approaches are applied to a water-scarce coastal arid region water management problem in northern Oman, where saltwater intrusion into a coastal aquifer due to excessive groundwater extraction for irrigated agriculture has affected the aquifer sustainability, endangering associated socio-economic conditions as well as traditional social structures. The results show the effectiveness of the proposed methods. The first method can aid in the impact assessment of alternative management interventions on sustainability of aquifer systems while accounting for economic (agriculture) and societal interests (employment in agricultural sector) in the study area. Results from the second method have provided key decision alternatives which can serve as a platform for negotiation and further exploration. In addition, this approach suits to systematically quantify both probabilistic and fuzzy uncertainties associated with the decision problem. The new approaches can be applied to address the complexities and uncertainties inherent in water resource systems to support management decisions, while serving as a platform for stakeholder participation.

Treatment decision-making in head and neck oncology is gaining complexity by the increasing evidence pointing towards more individualized and selective treatment options. Therefore, decision making in multidisciplinary teams is becoming the key point in the clinical pathways. Clinical decision-support systems based on Bayesian networks can support complex decision-making processes by providing mathematically correct and transparent advises. In the last three decades, different clinical applications of Bayesian networks have been proposed. Because appropriate data for model learning and testing is often unobtainable, expert modeling is required. To decrease the modeling and validation effort, networks usually represent small or highly simplified decision structures. However, especially systems for supporting multidisciplinary treatment decisions may only gain a user’s confidence if the systems’ results are comprehensive and comprehensible. Challenges in developing such systems relate to knowledge engineering, model validation, system interaction, clinical implementation and standardization. These challenges are well-known, however, they are not or only partially addressed by the developers. The thesis presented a methodology for the development of Bayesian network-based clinical treatment decision support systems. For this purpose, a concept introduced interactions between actors and systems. The proposed concept emphasizes model development with an exemplary use case of model interaction. A graph model design was presented that allows integrating all relevant variables of multidisciplinary treatment decisions. At the current stage, we developed TreLynCa: A graph model representing the treatment decisions of laryngeal cancer. From TreLynCa, a subnetwork that represents the TNM staging is completed by the required probabilistic parameters, and finally validated. The model validation required the development of a validation cycle in combination with existing data- and expert-based validation methods. Furthermore, modeling methods were developed that enable domain experts to model autonomously without Bayesian network expertise. Specifically, a novel graph modeling method was developed, and an existing method for modeling probabilistic parameters was extended. Both methods transform Bayesian network modeling tasks into a natural language form and provide a regulated modeling environment. A method for graph modeling is based on the presented graph model design with a regulated and restricted modeling procedure. This modeling procedure is supposed to enable collaborative modeling of compatible models. The method is currently under development. A method for probabilistic modeling is extended to reduce the modeling effort to a linear time. The method has been implemented as a web tool and was tested and evaluated in two studies. Finally, for clinical application of the TNM model, requirements were collected and constructed in a visual framework. In collaboration with visual scientists, the framework has been implemented and evaluated.

It is rarely possible to use an optimal classifier. Often the classifier used for a specific problem is an approximation of the optimal classifier. Methods are presented for evaluating the performance of an approximation in the model class of Bayesian Networks. Specifically for the approximation of class conditional independence a bound for the performance is sharpened. The class conditional independence approximation is connected to the minimum description length principle (MDL), which is connected to Jeffreys’ prior through commonly used assumptions. One algorithm for unsupervised classification is presented and compared against other unsupervised classifiers on three data sets. / <p>Report code: LiU-TEK-LIC 2006:11.</p>