L'étude des systèmes aux équations différentielles ordinaires vise à prédire le futur des systèmes considérés. La connaissance de l'évolution dans le temps de toutes les variables d' état du modèle permet de prédire de possibles changements radicaux des variables ou des défaillances, par exemple, un moteur peut exploser, un pont peut s'écrouler, une voiture peut se mettre à consommer plus d'essence. De plus, les systèmes dynamiques peuvent contenir des dérivées spatiales et leur discrétisation peut ajouter un très grand nombre d'équations. La résolution des équations différentielles ordinaires est alors une étape essentielle dans la construction des systèmes physiques en terme de dimensionnement et de faisabilité. Le solveur de tels systèmes EDOs doit être rapide, précis et pertinent.En pratique, il n'est pas possible de trouver une fonction continue qui soit solution exacte du problème EDO. C'est pourquoi, des méthodes numériques sont utilisées afin de donner des solutions discrèes qui approchent la solution continue avec une erreur contrôlable. La gestion précise de ce contrôle est très importante afin d'obtenir une solution pertinente en un temps raisonnable.Cette thèse développe un nouveau solveur qui utilise plusieurs méthodes d'amélioration de la vitesse d'exécution des systèmes EDOs. La première méthode est l'utilisation d'un nouveau schéma numérique. Le but est de minimiser le coût de l'intégration en produisant une erreur qui soit le plus proche possible de la tolérance maximale permise par l'utilisateur du solveur. Une autre méthode pour améliorer la vitesse d'exécution est de paralléliser le solveur EDO en utilisant une architecture multicoeur et multiprocesseur. Enfin, le solveur a été testé avec différentes applications d'OpenModelica. / The study of systems of Ordinary Differential Equations aims at predicting the future of the considered systems. The access to the evolution of all states of a system's model allows us to predict possible drastic shifts of the states or failures, e.g. an engine blowing up, a bridge collapsin, a car consuming more gasoline etc. Solving ordinary differential equations is then an essential step of building industrial physical systems in regard to dimensioning and reliability. The solver of such ODE systems needs to be fast, accurate and relevant.In practice, it is not possible to find a continuous function as the exact solution of the real ODE problem. Consequently numerical methods are used to give discrete solutions which approximates the continuous one with a controllable error. The correct handline of this control is very important to get a relevant solution within an acceptable recovery time. Starting from existing studies of local and global errors, this thesis work goes more deeply and adjusts the time step of the integration time algorithm and solves the problem in a very efficient manner.A new scheme is proposed is this thesis, to minimize the cost of integration. Another method to improve the execution speed is to parallelize the ODE solver by using a multicore and a multiprocessor architecture. Finally, the solver has been tested with different applications from OpenModelica.

<p dir="ltr">Deep neural networks (DNNs) have demonstrated outstanding performance in numerous tasks such as image recognition and natural language processing. However, in dynamic systems modeling, the tasks of estimating and uncovering the potentially nonlinear structure of systems represented by ordinary differential equations (ODEs) pose a significant challenge. In this dissertation, we employ DNNs to enable precise and efficient parameter estimation of dynamic systems. In addition, we introduce a highly flexible neural ODE model to capture both nonlinear and sparse dependent relations among multiple functional processes. Nonetheless, DNNs are susceptible to overfitting and often struggle to accurately assess predictive uncertainty despite their widespread success across various AI domains. The challenge of defining meaningful priors for DNN weights and characterizing predictive uncertainty persists. In this dissertation, we present a novel neural adaptive empirical Bayes framework with a new class of prior distributions to address weight uncertainty.</p><p dir="ltr">In the first part, we propose a precise and efficient approach utilizing DNNs for estimation and inference of ODEs given noisy data. The DNNs are employed directly as a nonparametric proxy for the true solution of the ODEs, eliminating the need for numerical integration and resulting in significant computational time savings. We develop a gradient descent algorithm to estimate both the DNNs solution and the parameters of the ODEs by optimizing a fidelity-penalized likelihood loss function. This ensures that the derivatives of the DNNs estimator conform to the system of ODEs. Our method is particularly effective in scenarios where only a set of variables transformed from the system components by a given function are observed. We establish the convergence rate of the DNNs estimator and demonstrate that the derivatives of the DNNs solution asymptotically satisfy the ODEs determined by the inferred parameters. Simulations and real data analysis of COVID-19 daily cases are conducted to show the superior performance of our method in terms of accuracy of parameter estimates and system recovery, and computational speed.</p><p dir="ltr">In the second part, we present a novel sparse neural ODE model to characterize flexible relations among multiple functional processes. This model represents the latent states of the functions using a set of ODEs and models the dynamic changes of these states utilizing a DNN with a specially designed architecture and sparsity-inducing regularization. Our new model is able to capture both nonlinear and sparse dependent relations among multivariate functions. We develop an efficient optimization algorithm to estimate the unknown weights for the DNN under the sparsity constraint. Furthermore, we establish both algorithmic convergence and selection consistency, providing theoretical guarantees for the proposed method. We illustrate the efficacy of the method through simulation studies and a gene regulatory network example.</p><p dir="ltr">In the third part, we introduce a class of implicit generative priors to facilitate Bayesian modeling and inference. These priors are derived through a nonlinear transformation of a known low-dimensional distribution, allowing us to handle complex data distributions and capture the underlying manifold structure effectively. Our framework combines variational inference with a gradient ascent algorithm, which serves to select the hyperparameters and approximate the posterior distribution. Theoretical justification is established through both the posterior and classification consistency. We demonstrate the practical applications of our framework through extensive simulation examples and real-world datasets. Our experimental results highlight the superiority of our proposed framework over existing methods, such as sparse variational Bayesian and generative models, in terms of prediction accuracy and uncertainty quantification.</p>

Magistro darbe nagrinėjami paprastųjų tiesinių antros eilės diferencialinių lygčių sistemos su ypatuma kraštiniai uždaviniai. Ištirta sprendinių asimptotika ypatingojo taško aplinkoje. Surasti lygčių sistemos sprendinio įverčiai. Nagrinėjamų diferencialinių uždavinių sprendiniai, konstruojami naudojant integralinius operatorius. Įrodyta sprendinių vienatis. / In this paper the some boundary value problems for a system of ordinary second order differential equations with singularity are considered. The asymptotic of solutions in the neighborhood of singular point are discussed. The estimates of the solution are given. The solutions of considered differential problems are constructed using some integral operators. The uniqueness of the solutions is proved.

Le fonctionnement d'un neurone, unité fondamentale du système nerveux, intéresse de nombreuses disciplines scientifiques. Il existe ainsi des modèles mathématiques qui décrivent leur comportement par des systèmes d'EDO ou d'EDP. Plusieurs de ces modèles peuvent ensuite être couplés afin de pouvoir étudier le comportement de réseaux, systèmes complexes au sein desquels émergent des propriétés. Ce travail présente, dans un premier temps, les principaux mécanismes régissant ce fonctionnement pour en comprendre la modélisation. Plusieurs modèles sont alors présentés, jusqu'à celui de FitzHugh-Nagumo (FHN), qui présente une dynamique très intéressante.C'est sur l'étude théorique mais également numérique de la dynamique asymptotique et transitoire du modèle de FHN en EDO, que se concentre la seconde partie de cette thèse. A partir de cette étude, des réseaux d'interactions d'EDO sont construits en couplant les systèmes dynamiques précédemment étudiés. L'étude du phénomène de synchronisation identique au sein de ces réseaux montre l'existence de propriétés émergentes pouvant être caractérisées par exemple par des lois de puissance. Dans une troisième partie, on se concentre sur l'étude du système de FHN dans sa version EDP. Comme la partie précédente, des réseaux d'interactions d'EDP sont étudiés. On entreprend dans cette partie une étude théorique et numérique. Dans la partie théorique, on montre l'existence de l'attracteur global dans l'espace L2(Ω)nd et on donne des conditions suffisantes de synchronisation. Dans la partie numérique, on illustre le phénomène de synchronisation ainsi que l'émergence de lois générales telles que les lois puissances ou encore la formation de patterns, et on étudie l'effet de l'ajout de la dimension spatiale sur la synchronisation. / The neuron, a fundamental unit in the nervous system, is a point of interest in many scientific disciplines. Thus, there are some mathematical models that describe their behavior by ODE or PDE systems. Many of these models can then be coupled in order to study the behavior of networks, complex systems in which the properties emerge. Firstly, this work presents the main mechanisms governing the neuron behaviour in order to understand the different models. Several models are then presented, including the FitzHugh-Nagumo one, which has a interesting dynamic. The theoretical and numerical study of the asymptotic and transitory dynamics of the aforementioned model is then proposed in the second part of this thesis. From this study, the interaction networks of ODE are built by coupling previously dynamic systems. The study of identical synchronization phenomenon in these networks shows the existence of emergent properties that can be characterized by power laws. In the third part, we focus on the study of the PDE system of FHN. As the previous part, the interaction networks of PDE are studied. We have in this section a theoretical and numerical study. In the theoretical part, we show the existence of the global attractor on the space L2(Ω)nd and give the sufficient conditions for identical synchronization. In the numerical part, we illustrate the synchronization phenomenon, also the general laws of emergence such as the power laws or the patterns formation. The diffusion effect on the synchronization is studied.

<p>Interleukin-15 (IL-15) is a cytokine that promotes maintenance and activation of cytotoxic immune cells. Therapeutic IL-15 stimulates these cells to fight cancer and chronic infections, such as Human Immunodeficiency Virus (HIV). Animal models of HIV have demonstrated that IL-15 agonists can suppress the virus, but this was transient and was not observed in all cohorts. We developed a mechanistic mathematical model of IL-15 therapy of HIV to explain these differences in efficacy and to explore solutions. First, the model was applied to evaluate mitigating factors, including immune regulation, viral escape, and drug tolerance, using Akaike Information Criterion.  We found that immune regulatory mechanisms could explain the viral rebound observed with continued IL-15 therapy. Next, the model was expanded to allow it to simultaneously explain both the transient viral suppression noted above and the lack of viral suppression observed in another animal cohort.  In this cohort, the model suggested that higher pre-treatment viral load came with higher activation of immune cells and a balancing regulatory inhibition of cytotoxicity. Finally, we conducted stability analysis at a range of IL-15 therapeutic strengths. While there was an ideal IL-15 strength, monotherapy could not maintain viral levels below what would clinically be considered to be safely controlled. Stable viral control in the model required the combination of IL-15 with blockade of key regulatory pathways. Immune therapy of complex diseases will likely require combinations of medicines that boost the immune response at multiple key points. Mathematical models like this can expedite development of these treatments.</p>

In this thesis, I investigate dynamic and computational properties of prokaryotic signalling architectures commonly known as the Two Component Signalling networks and phosphorelays. The aim of this study is to understand the information processing capabilities of different prokaryotic signalling architectures by examining the dynamics they exhibit. I present original investigations into the dynamics of different phosphorelay architectures and identify network architectures that include a commonly found four step phosphorelay architecture with a capacity for tuning its steady state output to implement different signal-response behaviours viz. sigmoidal and hyperbolic response. Biologically, this tuning can be implemented through physiological processes like regulating total protein concentrations (e.g. via transcriptional regulation or feedback), altering reaction rate constants through binding of auxiliary proteins on relay components, or by regulating bi-functional activity in relays which are mediated by bifunctional histidine kinases. This study explores the importance of different biochemical arrangements of signalling networks and their corresponding response dynamics. Following investigations into the significance of various biochemical reactions and topological variants of a four step relay architecture, I explore the effects of having different types of proteins in signalling networks. I show how multi-domain proteins in a phosphorelay architecture with multiple phosphotransfer steps occurring on the same protein can exhibit multistability through a combination of double negative and positive feedback loops. I derive a minimal multistable (core) architecture and show how component sharing amongst networks containing this multistable core can implement computational logic (like AND, OR and ADDER functions) that allows cells to integrate multiple inputs and compute an appropriate response. I examine the genomic distribution of single and multi domain kinases and annotate their partner response regulator proteins across prokaryotic genomes to find the biological significance of dynamics that these networks embed and the processes they regulate in a cell. I extract data from a prokaryotic two component protein database and take a sequence based functional annotation approach to identify the process, function and localisation of different response regulators as signalling partners in these networks. In summary, work presented in this thesis explores the dynamic and computational properties of different prokaryotic signalling networks and uses them to draw an insight into the biological significance of multidomain sensor kinases in living cells. The thesis concludes with a discussion on how this understanding of the dynamic and computational properties of prokaryotic signalling networks can be used to design synthetic circuits involving different proteins comprising two component and phosphorelay architectures.

Dans ce travail de recherche, nous sommes intéresses par la modélisation de l'hématopoïèse. Les cellules souches hématopoïétiques (CSH) sont des cellules indifférenciées de la moelle osseuse, possédant la capacité de se renouveler et de se différencier (pour la production des globules rouges, globules blancs et les plaquettes). Le processus de l'hématopoïèse souvent révèle des irrégularités qui causent les maladies hématologiques. En modélisant la leucémie myéloide chronique (LMC), une maladie hématologique fréquente, nous représentons l'hématopoïèse des cellules normales et cancéreuses par un système d'équations différentielles ordinaires (EDO). L'homéostasie des cellules normales et différente de l'homéostasie des cellules cancéreuses, et dépend de quelques lignées des cellules normales et cancéreuses. Nous analysons la dynamique globale du modèle pour obtenir les conditions de régénération de l'hématopoïèse ou bien la persistance de la LMC. Nous démontrons aussi que la coexistence des cellules normales et cancéreuses ne peut avoir lieu pour longtemps. Imatinib est un traitement de base de la LMC, avec un dosage variant de 400 à 1000 mg par jour. Certains patients présentent des réponses différentes à la thérapie, pouvant être hématologique, cytogénétique et moléculaire. La thérapie échoue dans deux cas: le patient demande un temps plus long pour réagir, alors il s'agit d'une réponse suboptimale; ou bien le patient résiste après une bonne réponse initiale. Pour déterminer le dosage optimal, nécessaire à la réduction des cellules cancéreuses, nous représentons les effets de la thérapie par un problème de contrôle optimal. Notre but est de minimiser le cout du traitement et le nombre des cellules cancéreuses. La réponse suboptimale, la résistance et le rétablissement sont alors obtenus suivant l'influence de l'imatinib sur les taux de division et de mortalité des cellules cancéreuses. Nous étudions par ailleurs l'hématopoïèse selon un modèle structuré en age, décrivant l'évolution des CSH normales et cancéreuses. Nous démontrons que le taux de division des CSH cancéreuses joue un rôle important dans la détermination du contrôle optimal. En contrôlant la croissance des cellules normales et cancéreuses avec compétition inter spécifique, nous démontrons que le dosage optimal dépend de l'homéostasie des CSH cancéreuses. / Modelling hematopoiesis represents a feature of our research. Hematopoietic stem cells (HSC) are undifferentiated cells, located in bone marrow, with unique abilities of self-renewal and differentiation (production of white cells, red blood cells and platelets).The process of hematopoiesis often exhibits abnormalities causing hematological diseases. In modelling Chronic Myeloid Leukemia (CML), a frequent hematological disease, we represent hematopoiesis of normal and leukemic cells by means of ordinary differential equations (ODE). Homeostasis of normal and leukemic cells are supposed to be different and depend on some lines of normal and leukemic HSC. We analyze the global dynamics of the model to obtain the conditions for regeneration of hematopoiesis and persistence of CML. We prove as well that normal and leukemic cells can not coexist for a long time. Imatinib is the main treatment of CML, with posology varying from 400 to 1000 mg per day. Some affected individuals respond to therapy with various levels being hematologic, cytogenetic and molecular. Therapy fails in two cases: the patient takes a long time to react, then suboptimal response occurs; or the patient resists after an initial response. Determining the optimal dosage required to reduce leukemic cells is another challenge. We approach therapy effects as an optimal control problem to minimize the cost of treatment and the level of leukemic cells. Suboptimal response, resistance and recovery forms are obtained through the influence of imatinib onto the division and mortality rates of leukemic cells. Hematopoiesis can be investigated according to age of cells. An age-structured system, describing the evolution of normal and leukemic HSC shows that the division rate of leukemic HSC plays a crucial role when determining the optimal control. When controlling the growth of cells under interspecific competition within normal and leukemic HSC, we prove that optimal dosage is related to homeostasis of leukemic HSC.

<p dir="ltr">In Chapter 1, we introduce geometric motions from the general perspective of gradient flows. Here we develop the basic framework in which to pose the two main results of this thesis.</p><p dir="ltr">In Chapter 2, we examine the pinch-off phenomenon for a tubular surface moving by surface diffusion. We prove the existence of a one parameter family of pinching profiles obeying a long wavelength approximation of the dynamics.</p><p dir="ltr">In Chapter 3, we study a diffusion-based numerical scheme for curve shortening flow. We prove that the scheme is one time-step consistent.</p>