A framework for optimization of reinsurance strategy is proposed for an insurance company with several lines of business (LoB), maximizing the Economic Value of purchasing reinsurance. The economic value is defined as the sum of the average ceded loss, the deducted risk premium, and the reduction in the cost of capital. The framework relies on simulated large claims per LoB rather than specific distributions, which gives more degrees of freedom to the insurance company.  Three models are presented, two non non-linear optimization models and a benchmark model. One non-linear optimization model is on individual LoB level and the other one is on company level with additional constraints using space bounded black box algorithms. The benchmark model is a Brute Force method using quantile discretization of potential retention levels, that helps to visualize the optimization surface.  The best results are obtained by a two-stage optimization using a mixture of global and local optimization algorithms. The economic value is maximized by 30% and reinsurance premium is halved if the optimization is made at the company level, by putting more emphasis on reduction in the cost of capital and less to average ceded loss. The results indicate an over-fitting when using VaR as the risk measure, impacting reduction in the cost of capital. As an alternative, Average VaR is recommended being numerically more robust.

La Modélisation spatio-temporelle permet la prédiction d’une variable régionalisée à des sites non observés du domaine d’étude, basée sur l’observation de cette variable en quelques sites du domaine à différents temps t donnés. Dans cette thèse, l’approche que nous avons proposé consiste à coupler des modèles numériques et statistiques. En effet en privilégiant l’approche bayésienne nous avons combiné les différentes sources d’information : l’information spatiale apportée par les observations, l’information temporelle apportée par la boîte noire ainsi que l’information a priori connue du phénomène. Ce qui permet une meilleure prédiction et une bonne quantification de l’incertitude sur la prédiction. Nous avons aussi proposé un nouveau critère d’optimalité de plans d’expérience incorporant d’une part le contrôle de l’incertitude en chaque point du domaine et d’autre part la valeur espérée du phénomène. / Spatio-temporal modeling allows to make the prediction of a regionalized variable at unobserved points of a given field, based on the observations of this variable at some points of field at different times. In this thesis, we proposed a approach which combine numerical and statistical models. Indeed by using the Bayesian methods we combined the different sources of information : spatial information provided by the observations, temporal information provided by the black-box and the prior information on the phenomenon of interest. This approach allowed us to have a good prediction of the variable of interest and a good quantification of incertitude on this prediction. We also proposed a new method to construct experimental design by establishing a optimality criterion based on the uncertainty and the expected value of the phenomenon.

Hypothesis testing has long been a formal and standardized process. Hypothesis generation, on the other hand, remains largely informal. This thesis assess whether eXplainable AI (XAI) can aid in the standardization of hypothesis generation through its utilization as a hypothesis generating tool for medical research. We produce XAI heat maps for a Convolutional Neural Network (CNN) trained to classify Microsatellite Instability (MSI) in colon and gastric cancer with four different XAI methods: Guided Backpropagation, VarGrad, Grad-CAM and Sobol Attribution. We then compare these heat maps with pathology annotations in order to look for differences to turn into new hypotheses. Our CNN successfully generates non-random XAI heat maps whilst achieving a validation accuracy of 85% and a validation AUC of 93% – as compared to others who achieve a AUC of 87%. Our results conclude that Guided Backpropagation and VarGrad are better at explaining high-level image features whereas Grad-CAM and Sobol Attribution are better at explaining low-level ones. This makes the two groups of XAI methods good complements to each other. Images of Microsatellite Insta- bility (MSI) with high differentiation are more difficult to analyse regardless of which XAI is used, probably due to exhibiting less regularity. Regardless of this drawback, our assessment is that XAI can be used as a useful hypotheses generating tool for research in medicine. Our results indicate that our CNN utilizes the same features as our basic pathology annotations when classifying MSI – with some additional features of basic pathology missing – features which we successfully are able to generate new hypotheses with.

[pt] Atuadores baseados em materiais piezelétricos apresentam características ideais para aplicações como transmissão acústica e micromanipulação. No entanto, não-linearidades inerentes a estes atuadores, como histerese e fluência, aumentam o desafio de controla-los. Além disso, a crescente necessidade de atuadores mais precisos e rápidos aliada a frequentes mudanças nas condições ambientais e operacionais agravam ainda mais o problema. Modelagens analíticas são específicas ao sistema ao qual foram feitas, o que significa que elas não são facilmente escalonáveis e eficientes para todos os tipos de sistemas. Adicionalmente, com o aumento da complexidade, os fenômenos que regem a física do sistema não são totalmente conhecidos, tornando difícil o desenvolvimento destes modelos. Este trabalho investiga esses desafios do ponto de vista da metodologia de identificação de sistemas e modelos baseados em dados para atuadores piezelétricos. A abordagem de modelagem caixa preta foi testada com dados experimentais adquiridos em um ambiente de laboratório para os estudos de caso de micromanipulação e transmissão acústica. Sinais de uso geral foram empregados como entrada de excitação do sistema de modo a acelerar a aquisição e estimação dos parâmetros. Parte dos modelos desenvolvidos foram validados com um conjunto de dados separado. Em ambos os casos foi necessário pré-processamento para otimização da quantidade de dados. Os modelos testados incluem a Média Móvel AutoRegressiva com entradas eXógenas (ARMAX), AutoRegressiva Não Linear com entradas eXógenas (NARX) com uma estrutura de rede neural artificial e Média Móvel AutoRegressiva Não Linear com entradas eXógenas (NARMAX). Os resultados mostram uma boa capacidade de prever as não-linearidades do micro manipulador e, portanto, a histerese em diferentes frequências de entrada. O sistema de transmissão acústica foi modelado com sucesso. Embora os resultados mostrem que ainda há espaço para melhorias, eles fornecem informações importantes sobre possíveis otimizações para o sistema uma vez que os modelos apresentados são uteis para janelas de predição curtas. / [en] Actuators based on piezoelectric materials have ideal characteristics for applications such as acoustic transmission and micromanipulation. However, the inherent nonlinearities of those actuators, such as hysteresis and creep, greatly increase the challenge to control such devices. Furthermore, the increasing need for more precise and faster actuators, allied with frequent changes in the environmental and operational conditions, further worsens the problem. Analytical models are application-specific, meaning that they are not easily and efficiently scalable to all systems. Also, with increased complexity, the understating of underlying phenomena is not fully documented, making it difficult to develop such models. This work investigates those challenges from the perspective of the system identification methodology and data-driven models for piezoelectric actuators. The black-box approach is tested with experimental data acquired in a laboratory setting for micromanipulator and acoustic transmission case studies. In some datasets, general-purpose signals were employed as the excitation input of the system to accelerate the data acquisition of the whole system dynamic and estimation process. Additionally, some models were validated on a separate dataset. In both cases, preprocessing was employed to optimize the amount of data. The tested models include the AutoRegressive Moving Average with eXogenous inputs (ARMAX), Nonlinear AutoRegressive with eXogenous inputs (NARX) with an artificial neural network structure, and Nonlinear AutoRegressive Moving Average with eXogenous inputs (NARMAX). The results show a good ability to predict the nonlinearities of the micromanipulator and, therefore, the hysteresis at different input frequencies. The acoustic transmission system was successfully modeled. Although the results show that there is still room for improvements, it provides insights into possible optimizations for the setup as the models here devised are useful for short prediction windows.

Tracked vehicles are designed to operate in various terrains, ranging from soft mud to hard tarmac. This wide range of terrains presents significant challenges for the suspension system, as its components must be suitable for all types of terrain. The selection of these components is crucial for minimizing acceleration levels within the vehicle, ensuring that personnel can comfortably endure extended durations inside. BAE Systems Hägglunds AB develops and produces an armored tracked vehicle called the CV90. Within the CV90’s suspension system, a key component known as the torsion bar, a rotational spring, plays a primary role in reducing the vehicle’s motion. The CV90 vehicle has seven wheels on each side, with each wheel having its dedicated torsion bar. To measure the whole-body vibration experienced within the vehicle, a measurement called the Vibrational Dose Value (VDV) is utilized. The main objective of this thesis is to develop a data-driven model to optimize the suspension system by identifying the combination of torsion bars that generates the smallest VDV. The data used for optimization is based on simulations of the CV90 vehicle in a virtual environment. In the simulation, the CV90 vehicle, with its full dynamics, is driven over a specific virtual road at a particular velocity. The simulation itself cannot be manipulated; only the input values can be adjusted. Thus, we consider the simulation as a black box, which led us to implement the black-box optimization algorithm known as Particle-Swarm. In this thesis, four different roads, each with velocities ranging from four to seven different levels, were provided to the optimization model. The results show that the model identifies a combination of torsion bars that generates a small VDV for all combinations of velocities and roads, with an average VDV improvement of around 20% - 60% compared to a reference case. Since this thesis serves as a proof of concept, the conclusion is that the devised method is effective and suitable for addressing the problem at hand. Nonetheless, for seamless integration of this method into the tracked vehicle development process, further research is necessary.

Det blir allt vanligare att föreslå att icke-transparant artificiell intelligens, s.k. black boxes, används inom akutmedicinen. I denna uppsats används etisk analys för att härleda sju riktlinjer för utveckling och användning av black boxes i akutmedicin. Analysen är grundad på sju variationer av ett tankeexperiment som involverar en läkare, en black box och en patient med bröstsmärta på en akutavdelning. Grundläggande begrepp, inklusive artificiell intelligens, black boxes, metoder för transparens, akutmedicin och etisk analys behandlas detaljerat. Tre viktiga områden av etisk vikt identifieras: samtycke; kultur, agentskap och privatliv; och skyldigheter. Dessa områden ger upphov till de sju variationerna. För varje variation urskiljs en viktig etisk fråga som identifieras och analyseras. En riktlinje formuleras och dess etiska rimlighet testas utifrån konsekventialistiska och deontologiska metoder. Tillämpningen av riktlinjerna på medicin i allmänhet, och angelägenheten av fortsatt etiska analys av black boxes och artificiell intelligens inom akutmedicin klargörs. / Artificially intelligent black boxes are increasingly being proposed for emergency medicine settings; this paper uses ethical analysis to develop seven practical guidelines for emergency medicine black box creation and use. The analysis is built around seven variations of a thought experiment involving a doctor, a black box, and a patient presenting chest pain in an emergency department. Foundational concepts, including artificial intelligence, black boxes, transparency methods, emergency medicine, and ethical analysis are expanded upon. Three major areas of ethical concern are identified, namely consent; culture, agency, and privacy; and fault. These areas give rise to the seven variations. For each, a key ethical question it illustrates is identified and analyzed. A practical guideline is then stated, and its ethical acceptability tested using consequentialist and deontological approaches. The applicability of the guidelines to medicine more generally, and the urgency of continued ethical analysis of black box artificial intelligence in emergency medicine, are clarified.

The purpose of this research was to study the lived experience of being a strategic leader, described as the black box of leadership, and to extend the limited research in this field. The researcher utilised the qualitative ethnographic methodology of direct observation, observing 138 discrete critical incidents that made up the lived experience of the five strategic leaders in the sample. The researcher further utilised observation tools from the field of Neuro Linguistic Programming, personal experiences, metaphors, allegories, analogies as well as deep personal introspection to make sense of the lived experience of the five CEOs. The primary research objective was to answer the question: What do CEOs do and how do they do it? A further related objective was to explore how they think about what they do. The research answered these questions by prising open the 'black box' of the lived experience of the strategic leader. The result of the research is the pure leadership spider web model. The pure leadership spider web model breaks down the lived experience of the strategic leader, the content of the black box, into eight dimensions: the pillars that make up the personal leadership philosophy; emotional states of mind brought to bear in meetings; kinaesthetic patterns used during meetings; meeting dynamics; emotional states brought to bear on day-to-day shop-floor engagement; emotional states brought to bear on leadership engagement sessions with other like business leaders; frames of mind governing the day-to-day experiences; and The Magic Language Box. / Business Management and Entr / DBL

The purpose of this research was to study the lived experience of being a strategic leader, described as the black box of leadership, and to extend the limited research in this field. The researcher utilised the qualitative ethnographic methodology of direct observation, observing 138 discrete critical incidents that made up the lived experience of the five strategic leaders in the sample. The researcher further utilised observation tools from the field of Neuro Linguistic Programming, personal experiences, metaphors, allegories, analogies as well as deep personal introspection to make sense of the lived experience of the five CEOs. The primary research objective was to answer the question: What do CEOs do and how do they do it? A further related objective was to explore how they think about what they do. The research answered these questions by prising open the 'black box' of the lived experience of the strategic leader. The result of the research is the pure leadership spider web model. The pure leadership spider web model breaks down the lived experience of the strategic leader, the content of the black box, into eight dimensions: the pillars that make up the personal leadership philosophy; emotional states of mind brought to bear in meetings; kinaesthetic patterns used during meetings; meeting dynamics; emotional states brought to bear on day-to-day shop-floor engagement; emotional states brought to bear on leadership engagement sessions with other like business leaders; frames of mind governing the day-to-day experiences; and The Magic Language Box. / Business Management and Entr / DBL

Le déploiement d'une infrastructure de supervision permet une gestion plus intelligente des réseaux de distribution d'électricité comparé à un renforcement traditionnel pour répondre aux nouveaux enjeux de la maitrise de l'énergie (Consommations, EnR, VE, ...). Pour acheminer les données, les Courants Porteurs en Ligne (CPL) possèdent un atout majeur. En effet, cette technologie permet de superposer un signal de plus haute fréquence au signal électrique 50/60 Hz. Toutefois, le support de transmission est difficile et non maîtrisable. Ces travaux de recherche ont pour objectif d'apporter une contribution à cette problématique par l'élaboration d'une plateforme de simulation des réseaux pour des fréquences allant jusqu'à 1 MHz dans un but de transmission de données. Des éléments clés des réseaux sont traités de façon séparés puis assemblés pour estimer les performances des CPL « Outdoor » actuels. La variation du comportement des réseaux en fonction du temps et de la fréquence, en particulier des perturbations en tête d'installation clients sur 24h est étudiée. Les transformateurs entre les réseaux HTA et BT sont modélisés sous la forme d'un « modèle à constantes localisées » et d'un « modèle boite noire ». Les deux modèles sont appliqués sur un transformateur H61 100 kVA. Par la suite, une modélisation des câbles de distribution est proposée sous forme d'un « modèle cascadé ». Celle-ci est appliquée sur un câble souterrain BT. Chaque modèle est obtenu à l'aide de mesures d'impédances, et validé par des mesures de transmissions. Pour compléter, une étude préliminaire sur les communications radio mobile est réalisée pour la supervision des réseaux de distribution. / Establishing a supervisory infrastructure allows a better smart management than an expensive strengthening of distribution network to respond to new constraints at the energies control (Consumption, REN, EV ...). To transmit data, Power Line Communication (PLC) technologies present an advantage in this context. In fact, it enables a superposition of High Frequency (HF) signals on electrical signal 50/60 Hz. However, electric networks have not been developed to this application because of difficult propagation conditions. This research work makes a contribution to develop a simulation platform in objective to transmit data to 1 MHz. In first time, each network element is studied singly and in second time, together, to estimate "Outdoor PLC" transmission performance. The first element studied is the networks variation in function of frequency and time. Several 24h disturbance measurements on LV customers are presented. The second element is the transformers which established connection between Medium Voltage (MV) and Low Voltage (LV). The proposed modeling method is based on a "lumped model" and a "black box model". These models are applied to a 100 kVA H61 transformer most commonly used by French distribution system operator in rural and suburban networks. The third element is the power line used in MV and LV networks. The proposed modeling method is based on a "cascaded model" from the theory of transmission line. This model is applied to one power line used in LV underground network. Each model is obtained from various impedance measurements. To complete, an introductory study on mobile radio communication is performed to remote network distribution.

In recent years, advanced machine learning (ML) models have revolutionized industries ranging from the healthcare sector to retail and E-commerce. However, these models have become increasingly complex, making it difficult for even domain experts to understand and retrace the model's decision-making process. To address this challenge, several frameworks for explainable AI have been proposed and developed. This thesis focuses on counterfactual explanations (CFEs), which provide actionable insights by informing users how to modify inputs to achieve desired outputs. However, computing CFEs for a general black-box ML model is computationally expensive since it hinges on solving a challenging optimization problem. To efficiently solve this optimization problem, we propose using Bayesian optimization (BO), and introduce the novel algorithm Separated Bayesian Optimization (SBO). SBO exploits the formulation of the counterfactual function as a composite function. Additionally, we propose warm-starting SBO, which addresses the computational challenges associated with computing multiple CFEs. By decoupling the generation of a surrogate model for the black-box model and the computation of specific CFEs, warm-starting SBO allows us to reuse previous data and computations, resulting in computational discounts and improved efficiency for large-scale applications. Through numerical experiments, we demonstrate that BO is a viable optimization scheme for computing CFEs for black-box ML models. BO achieves computational efficiency while maintaining good accuracy. SBO improves upon this by requiring fewer evaluations while achieving accuracies comparable to the best conventional optimizer tested. Both BO and SBO exhibit improved capabilities in handling various classes of ML decision models compared to the tested baseline optimizers. Finally, Warm-starting SBO significantly enhances the performance of SBO, reducing function evaluations and errors when computing multiple sequential CFEs. The results indicate a strong potential for large-scale industry applications. / Avancerade maskininlärningsmodeller (ML-modeller) har på senaste åren haft stora framgångar inom flera delar av näringslivet, med allt ifrån hälso- och sjukvårdssektorn till detaljhandel och e-handel. I jämn takt med denna utveckling har det dock även kommit en ökad komplexitet av dessa ML-modeller vilket nu lett till att även domänexperter har svårigheter med att förstå och tolka modellernas beslutsprocesser. För att bemöta detta problem har flertalet förklarbar AI ramverk utvecklats. Denna avhandling fokuserar på kontrafaktuella förklaringar (CFEs). Detta är en förklaringstyp som anger för användaren hur denne bör modifiera sin indata för att uppnå ett visst modellbeslut. För en generell svarta-låda ML-modell är dock beräkningsmässigt kostsamt att beräkna CFEs då det krävs att man löser ett utmanande optimeringsproblem. För att lösa optimeringsproblemet föreslår vi användningen av Bayesiansk Optimering (BO), samt presenterar den nya algoritmen Separated Bayesian Optimization (SBO). SBO utnyttjar kompositionsformuleringen av den kontrafaktuella funktionen. Vidare, utforskar vi beräkningen av flera sekventiella CFEs för vilket vi presenterar varm-startad SBO. Varm-startad SBO lyckas återanvända data samt beräkningar från tidigare CFEs tack vare en separation av surrogat-modellen för svarta-låda ML-modellen och beräkningen av enskilda CFEs. Denna egenskap leder till en minskad beräkningskostnad samt ökad effektivitet för storskaliga tillämpningar.  I de genomförda experimenten visar vi att BO är en lämplig optimeringsmetod för att beräkna CFEs för svarta-låda ML-modeller tack vare en god beräknings effektivitet kombinerat med hög noggrannhet. SBO presterade ännu bättre med i snitt färre funktionsutvärderingar och med fel nivåer jämförbara med den bästa testade konventionella optimeringsmetoden. Både BO och SBO visade på bättre kapacitet att hantera olika klasser av ML-modeller än de andra testade metoderna. Slutligen observerade vi att varm-startad SBO gav ytterligare prestandaökningar med både minskade funktionsutvärderingar och fel när flera CFEs beräknades. Dessa resultat pekar på stor potential för storskaliga tillämpningar inom näringslivet.