In this thesis, we develop partial differential equation (PDE) based numerical methods to solve certain optimal stochastic control problems in finance. The value of a stochastic control problem is normally identical to the viscosity solution of a Hamilton-Jacobi-Bellman (HJB) equation or an HJB variational inequality. The HJB equation corresponds to the case when the controls are bounded while the HJB variational inequality corresponds to the unbounded control case. As a result, the solution to the stochastic control problem can be computed by solving the corresponding HJB equation/variational inequality as long as the convergence to the viscosity solution is guaranteed. We develop a unified numerical scheme based on a semi-Lagrangian timestepping for solving both the bounded and unbounded stochastic control problems as well as the discrete cases where the controls are allowed only at discrete times. Our scheme has the following useful properties: it is unconditionally stable; it can be shown rigorously to converge to the viscosity solution; it can easily handle various stochastic models such as jump diffusion and regime-switching models; it avoids Policy type iterations at each mesh node at each timestep which is required by the standard implicit finite difference methods. In this thesis, we demonstrate the properties of our scheme by valuing natural gas storage facilities---a bounded stochastic control problem, and pricing variable annuities with guaranteed minimum withdrawal benefits (GMWBs)---an unbounded stochastic control problem. In particular, we use an impulse control formulation for the unbounded stochastic control problem and show that the impulse control formulation is more general than the singular control formulation previously used to price GMWB contracts.

In this thesis, we develop partial differential equation (PDE) based numerical methods to solve certain optimal stochastic control problems in finance. The value of a stochastic control problem is normally identical to the viscosity solution of a Hamilton-Jacobi-Bellman (HJB) equation or an HJB variational inequality. The HJB equation corresponds to the case when the controls are bounded while the HJB variational inequality corresponds to the unbounded control case. As a result, the solution to the stochastic control problem can be computed by solving the corresponding HJB equation/variational inequality as long as the convergence to the viscosity solution is guaranteed. We develop a unified numerical scheme based on a semi-Lagrangian timestepping for solving both the bounded and unbounded stochastic control problems as well as the discrete cases where the controls are allowed only at discrete times. Our scheme has the following useful properties: it is unconditionally stable; it can be shown rigorously to converge to the viscosity solution; it can easily handle various stochastic models such as jump diffusion and regime-switching models; it avoids Policy type iterations at each mesh node at each timestep which is required by the standard implicit finite difference methods. In this thesis, we demonstrate the properties of our scheme by valuing natural gas storage facilities---a bounded stochastic control problem, and pricing variable annuities with guaranteed minimum withdrawal benefits (GMWBs)---an unbounded stochastic control problem. In particular, we use an impulse control formulation for the unbounded stochastic control problem and show that the impulse control formulation is more general than the singular control formulation previously used to price GMWB contracts.

Parmi tous les jeux de société joués par les humains au cours de l’histoire, le jeu de go était considéré comme l’un des plus difficiles à maîtriser par un programme informatique [Van Den Herik et al., 2002]; Jusqu’à ce que ce ne soit plus le cas [Silveret al., 2016]. Cette percée révolutionnaire [Müller, 2002, Van Den Herik et al., 2002] fût le fruit d’une combinaison sophistiquée de Recherche arborescente Monte-Carlo et de techniques d’apprentissage automatique pour évaluer les positions du jeu, mettant en lumière le grand potentiel de l’apprentissage automatique pour résoudre des jeux. L’apprentissage antagoniste, un cas particulier de l’optimisation multiobjective, est un outil de plus en plus utile dans l’apprentissage automatique. Par exemple, les jeux à deux joueurs et à somme nulle sont importants dans le domain des réseaux génératifs antagonistes [Goodfellow et al., 2014] ainsi que pour maîtriser des jeux comme le Go ou le Poker en s’entraînant contre lui-même [Silver et al., 2017, Brown andSandholm, 2017]. Un résultat classique de la théorie des jeux indique que les jeux convexes-concaves ont toujours un équilibre [Neumann, 1928]. Étonnamment, les praticiens en apprentissage automatique entrainent avec succès une seule paire de réseaux de neurones dont l’objectif est un problème de minimax non-convexe et non-concave alors que pour une telle fonction de gain, l’existence d’un équilibre de Nash n’est pas garantie en général. Ce travail est une tentative d'établir une solide base théorique pour l’apprentissage dans les jeux. La première contribution explore le théorème minimax pour une classe particulière de jeux non-convexes et non-concaves qui englobe les réseaux génératifs antagonistes. Cette classe correspond à un ensemble de jeux à deux joueurs et a somme nulle joués avec des réseaux de neurones. Les deuxième et troisième contributions étudient l’optimisation des problèmes minimax, et plus généralement, les inégalités variationnelles dans le cadre de l’apprentissage automatique. Bien que la méthode standard de descente de gradient ne parvienne pas à converger vers l’équilibre de Nash de jeux convexes-concaves simples, il existe des moyens d’utiliser des gradients pour obtenir des méthodes qui convergent. Nous étudierons plusieurs techniques telles que l’extrapolation, la moyenne et la quantité de mouvement à paramètre négatif. La quatrième contribution fournit une étude empirique du comportement pratique des réseaux génératifs antagonistes. Dans les deuxième et troisième contributions, nous diagnostiquons que la méthode du gradient échoue lorsque le champ de vecteur du jeu est fortement rotatif. Cependant, une telle situation peut décrire un pire des cas qui ne se produit pas dans la pratique. Nous fournissons de nouveaux outils de visualisation afin d’évaluer si nous pouvons détecter des rotations dans comportement pratique des réseaux génératifs antagonistes. / Among all the historical board games played by humans, the game of go was considered one of the most difficult to master by a computer program [Van Den Heriket al., 2002]; Until it was not [Silver et al., 2016]. This odds-breaking break-through [Müller, 2002, Van Den Herik et al., 2002] came from a sophisticated combination of Monte Carlo tree search and machine learning techniques to evaluate positions, shedding light upon the high potential of machine learning to solve games. Adversarial training, a special case of multiobjective optimization, is an increasingly useful tool in machine learning. For example, two-player zero-sum games are important for generative modeling (GANs) [Goodfellow et al., 2014] and mastering games like Go or Poker via self-play [Silver et al., 2017, Brown and Sandholm,2017]. A classic result in Game Theory states that convex-concave games always have an equilibrium [Neumann, 1928]. Surprisingly, machine learning practitioners successfully train a single pair of neural networks whose objective is a nonconvex-nonconcave minimax problem while for such a payoff function, the existence of a Nash equilibrium is not guaranteed in general. This work is an attempt to put learning in games on a firm theoretical foundation. The first contribution explores minimax theorems for a particular class of nonconvex-nonconcave games that encompasses generative adversarial networks. The proposed result is an approximate minimax theorem for two-player zero-sum games played with neural networks, including WGAN, StarCrat II, and Blotto game. Our findings rely on the fact that despite being nonconcave-nonconvex with respect to the neural networks parameters, the payoff of these games are concave-convex with respect to the actual functions (or distributions) parametrized by these neural networks. The second and third contributions study the optimization of minimax problems, and more generally, variational inequalities in the context of machine learning. While the standard gradient descent-ascent method fails to converge to the Nash equilibrium of simple convex-concave games, there exist ways to use gradients to obtain methods that converge. We investigate several techniques such as extrapolation, averaging and negative momentum. We explore these techniques experimentally by proposing a state-of-the-art (at the time of publication) optimizer for GANs called ExtraAdam. We also prove new convergence results for Extrapolation from the past, originally proposed by Popov [1980], as well as for gradient method with negative momentum. The fourth contribution provides an empirical study of the practical landscape of GANs. In the second and third contributions, we diagnose that the gradient method breaks when the game’s vector field is highly rotational. However, such a situation may describe a worst-case that does not occur in practice. We provide new visualization tools in order to exhibit rotations in practical GAN landscapes. In this contribution, we show empirically that the training of GANs exhibits significant rotations around Local Stable Stationary Points (LSSP), and we provide empirical evidence that GAN training converges to a stable stationary point, which is a saddle point for the generator loss, not a minimum, while still achieving excellent performance.

L’apprentissage automatique repose pour un bon nombre de problèmes sur la minimisation d’une fonction de coût, pour ce faire il tire parti de la vaste littérature sur l’optimisation qui fournit des algorithmes et des garanties de convergences pour ce type de problèmes. Cependant récemment plusieurs modèles d’apprentissage automatique qui ne peuvent pas être formulé comme la minimisation d’un coût unique ont été propose, à la place ils nécessitent de définir un jeu entre plusieurs joueurs qui ont chaque leur propre objectif. Un de ces modèles sont les réseaux antagonistes génératifs (GANs). Ce modèle génératif formule un jeu entre deux réseaux de neurones, un générateur et un discriminateur, en essayant de tromper le discriminateur qui essaye de distinguer les vraies images des fausses, le générateur et le discriminateur s’améliore résultant en un équilibre de Nash, ou les images produites par le générateur sont indistinguable des vraies images. Malgré leur succès les GANs restent difficiles à entrainer à cause de la nature antagoniste du jeu, nécessitant de choisir les bons hyperparamètres et résultant souvent en une dynamique d’entrainement instable. Plusieurs techniques de régularisations ont été propose afin de stabiliser l’entrainement, dans cette thèse nous abordons ces instabilités sous l’angle d’un problème d’optimisation. Nous commençons par combler le fossé entre la littérature d’optimisation et les GANs, pour ce faire nous formulons GANs comme un problème d’inéquation variationnelle, et proposons de la littérature sur le sujet pour proposer des algorithmes qui convergent plus rapidement. Afin de mieux comprendre quels sont les défis de l’optimisation des jeux, nous proposons plusieurs outils afin d’analyser le paysage d’optimisation des GANs. En utilisant ces outils, nous montrons que des composantes rotationnelles sont présentes dans le voisinage des équilibres, nous observons également que les GANs convergent rarement vers un équilibre de Nash mais converge plutôt vers des équilibres stables locaux (LSSP). Inspirer par le succès des GANs nous proposons pour finir, une nouvelle famille de jeux que nous appelons adversarial example games qui consiste à entrainer simultanément un générateur et un critique, le générateur cherchant à perturber les exemples afin d’induire en erreur le critique, le critique cherchant à être robuste aux perturbations. Nous montrons qu’à l’équilibre de ce jeu, le générateur est capable de générer des perturbations qui transfèrent à toute une famille de modèles. / Many machine learning (ML) problems can be formulated as minimization problems, with a large optimization literature that provides algorithms and guarantees to solve this type of problems. However, recently some ML problems have been proposed that cannot be formulated as minimization problems but instead require to define a game between several players where each player has a different objective. A successful application of such games in ML are generative adversarial networks (GANs), where generative modeling is formulated as a game between a generator and a discriminator, where the goal of the generator is to fool the discriminator, while the discriminator tries to distinguish between fake and real samples. However due to the adversarial nature of the game, GANs are notoriously hard to train, requiring careful fine-tuning of the hyper-parameters and leading to unstable training. While regularization techniques have been proposed to stabilize training, we propose in this thesis to look at these instabilities from an optimization perspective. We start by bridging the gap between the machine learning and optimization literature by casting GANs as an instance of the Variational Inequality Problem (VIP), and leverage the large literature on VIP to derive more efficient and stable algorithms to train GANs. To better understand what are the challenges of training GANs, we then propose tools to study the optimization landscape of GANs. Using these tools we show that GANs do suffer from rotation around their equilibrium, and that they do not converge to Nash-Equilibria. Finally inspired by the success of GANs to generate images, we propose a new type of games called Adversarial Example Games that are able to generate adversarial examples that transfer across different models and architectures.

Submitted by Erika Demachki (erikademachki@gmail.com) on 2015-05-21T19:19:51Z No. of bitstreams: 2 Tese - Reinier Díaz Millán - 2015.pdf: 3568052 bytes, checksum: b4c892f77911a368e1b8f629afb5e66e (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Approved for entry into archive by Erika Demachki (erikademachki@gmail.com) on 2015-05-21T19:21:31Z (GMT) No. of bitstreams: 2 Tese - Reinier Díaz Millán - 2015.pdf: 3568052 bytes, checksum: b4c892f77911a368e1b8f629afb5e66e (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Made available in DSpace on 2015-05-21T19:21:31Z (GMT). No. of bitstreams: 2 Tese - Reinier Díaz Millán - 2015.pdf: 3568052 bytes, checksum: b4c892f77911a368e1b8f629afb5e66e (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) Previous issue date: 2015-02-27 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / Nesta tese apresentamos v arios algoritmos para resolver os problemas de Desigualdade Variacional e Inclus~ao. Para o problema de desigualdade variacional propomos, no Cap tulo 2 uma generaliza c~ao do algoritmo cl assico extragradiente, utilizando vetores normais n~ao nulos do conjunto vi avel. Em particular, dois algoritmos conceituais s~ao propostos e cada um deles cont^em tr^es variantes diferentes de proje c~ao que est~ao relacionadas com algoritmos extragradientes modi cados. Duas buscas diferentes s~ao propostas, uma sobre a borda do conjunto vi avel e a outra ao longo das dire c~oes vi aveis. Cada algoritmo conceitual tem uma estrat egia diferente de busca e tr^es formas de proje c~ao especiais, gerando tr^es sequ^encias com diferente e interessantes propriedades. E feito a an alise da converg^encia de ambos os algoritmos conceituais, pressupondo a exist^encia de solu c~oes, continuidade do operador e uma condi c~ao mais fraca do que pseudomonotonia. No Cap tulo 4, n os introduzimos um algoritmo direto de divis~ao para o problema variacional em espa cos de Hilbert. J a no Cap tulo 5, propomos um algoritmo de proje c~ao relaxada em Espa cos de Hilbert para a soma de m operadores mon otonos maximais ponto-conjunto, onde o conjunto vi avel do problema de desigualdade variacional e dado por uma fun c~ao n~ao suave e convexa. Neste caso, as proje c~oes ortogonais ao conjunto vi avel s~ao substitu das por proje c~oes em hiperplanos que separam a solu c~ao da itera c~ao atual. Cada itera c~ao do m etodo proposto consiste em proje c~oes simples de tipo subgradientes, que n~ao exige a solu c~ao de subproblemas n~ao triviais, utilizando apenas os operadores individuais, explorando assim a estrutura do problema. Para o problema de Inclus~ao, propomos variantes do m etodo de divis~ao de forward-backward para achar um zero da soma de dois operadores, a qual e a modi ca c~ao cl assica do forwardbackward proposta por Tseng. Um algoritmo conceitual e proposto para melhorar o apresentado por Tseng em alguns pontos. Nossa abordagem cont em, primeramente, uma busca linear tipo Armijo expl cita no esp rito dos m etodos tipo extragradientes para desigualdades variacionais. Durante o processo iterativo, a busca linear realiza apenas um c alculo do operador forward-backward em cada tentativa de achar o tamanho do passo. Isto proporciona uma consider avel vantagem computacional pois o operador forward-backward e computacionalmente caro. A segunda parte do esquema consiste em diferentes tipos de proje c~oes, gerando sequ^encias com caracter sticas diferentes. / In this thesis we present various algorithms to solve the Variational Inequality and Inclusion Problems. For the variational inequality problem we propose, in Chapter 2, a generalization of the classical extragradient algorithm by utilizing non-null normal vectors of the feasible set. In particular, two conceptual algorithms are proposed and each of them has three di erent projection variants which are related to modi ed extragradient algorithms. Two di erent linesearches, one on the boundary of the feasible set and the other one along the feasible direction, are proposed. Each conceptual algorithm has a di erent linesearch strategy and three special projection steps, generating sequences with di erent and interesting features. Convergence analysis of both conceptual algorithms are established, assuming existence of solutions, continuity and a weaker condition than pseudomonotonicity on the operator. In Chapter 4 we introduce a direct splitting method for solving the variational inequality problem for the sum of two maximal monotone operators in Hilbert space. In Chapter 5, for the same problem, a relaxed-projection splitting algorithm in Hilbert spaces for the sum of m nonsmooth maximal monotone operators is proposed, where the feasible set of the variational inequality problem is de ned by a nonlinear and nonsmooth continuous convex function inequality. In this case, the orthogonal projections onto the feasible set are replaced by projections onto separating hyperplanes. Furthermore, each iteration of the proposed method consists of simple subgradient-like steps, which does not demand the solution of a nontrivial subproblem, using only individual operators, which explores the structure of the problem. For the Inclusion Problem, in Chapter 3, we propose variants of forward-backward splitting method for nding a zero of the sum of two operators, which is a modi cation of the classical forward-backward method proposed by Tseng. The conceptual algorithm proposed here improves Tseng's method in many instances. Our approach contains rstly an explicit Armijo-type line search in the spirit of the extragradient-like methods for variational inequalities. During the iterative process, the line search performs only one calculation of the forward-backward operator in each tentative for nding the step size. This achieves a considerable computational saving when the forward-backward operator is computationally expensive. The second part of the scheme consists of special projection steps bringing several variants.

This thesis is devoted to the study of optimal control problems governed by a quasistatic, thermoviscoplastic model at small strains with linear kinematic hardening, von Mises yield condition and mixed boundary conditions. Mathematically, the thermoviscoplastic equations are given by nonlinear partial differential equations and a variational inequality of second kind in order to represent the elastic, plastic and thermal effects. Taking into account thermal effects we have to handle numerous mathematical challenges during the analysis of the thermoviscoplastic model, mainly due to the low integrability of the nonlinear terms on the right-hand side of the heat equation. One of our main results is the existence of a unique weak solution, which is proved by means of a fixed-point argument and by employing maximal parabolic regularity theory. Furthermore, we define the related control-to-state mapping and investigate properties of this mapping such as boundedness, weak continuity and local Lipschitz continuity. Another major result is the finding that the mapping is Hadamard differentiable; a main ingredient is the reformulation of the variational inequality, the so called viscoplastic flow rule, as a Banach space-valued ordinary differential equation with non-differentiable right-hand side. Subsequently, we consider an optimal control problem governed by thermoviscoplasticity and show the existence of a minimizer. Finally, close this thesis with numerical examples. / Diese Arbeit ist der Untersuchung von Optimalsteuerproblemen gewidmet, denen ein quasistatisches, thermoviskoplastisches Model mit kleinen Deformationen, mit linearem kinematischen Hardening, von Mises Fließbedingung und gemischten Randbedingungen zu Grunde liegt. Mathematisch werden thermoviskoplastische Systeme durch nichtlineare partielle Differentialgleichungen und eine variationelle Ungleichung der zweiten Art beschrieben, um die elastischen, plastischen und thermischen Effekte abzubilden. Durch die Miteinbeziehung thermischer Effekte, treten verschiedene mathematische Schwierigkeiten während der Analysis des thermoviskoplastischen Systems auf, die ihren Ursprung hauptsächlich in der schlechten Regularität der nichtlinearen Terme auf der rechten Seite der Wärmeleitungsgleichung haben. Eines unserer Hauptresultate ist die Existenz einer eindeutigen schwachen Lösung, welches wir mit Hilfe von einem Fixpunktargument und unter Anwendung von maximaler parabolischer Regularitätstheorie beweisen. Zudem definieren wir die entsprechende Steuerungs-Zustands-Abbildung und untersuchen Eigenschaften dieser Abbildung wie die Beschränktheit, schwache Stetigkeit und lokale Lipschitz Stetigkeit. Ein weiteres wichtiges Resultat ist, dass die Abbildung Hadamard differenzierbar ist; Hauptbestandteil des Beweises ist die Umformulierung der variationellen Ungleichung, der sogenannten viskoplastischen Fließregel, als eine Banachraum-wertige gewöhnliche Differentialgleichung mit nichtdifferenzierbarer rechter Seite. Schließlich runden wir diese Arbeit mit numerischen Beispielen ab.

Dans cette thèse, tout d’abord, nous faisons l’Analyse Mathématique du modèle exact du chauffage radiatif d’un corps semi-transparent $\Omega$ par une source radiative noire qui l’entoure. Il s’agit donc d’étudier le couplage d’un système d’Equations de Transfert Radiatif avec condition au bord de réflectivité indépendantes avec une équation de conduction de la chaleur non linéaire avec condition limite non linéaire de type Robin. Nous prouvons l’existence et l’unicité de la solution et nous démontrons des bornes uniformes sur la solution et les intensités radiatives dans chaque bande de longueurs d’ondes pour laquelle le corps est semi-transparent, en fonction de bornes sur les données, Deuxièmement, nous considérons le problème du contrôle optimal de la température absolue à l’intérieur du corps semi-transparent $\Omega$ en agissant sur la température absolue de la source radiative noire qui l’entoure. À cet égard, nous introduisons la fonctionnelle coût appropriée et l’ensemble des contrôles admissibles $T_{S}$, pour lesquels nous prouvons l’existence de contrôles optimaux. En introduisant l’espace des états et l’équation d’état, une condition nécessaire de premier ordre pour qu’un contrôle $T_{S}$ : t ! $T_{S}$ (t) soit optimal, est alors dérivée sous la forme d’une inéquation variationnelle en utilisant le théorème des fonctions implicites et le problème adjoint. Ensuite, nous considérons le problème de l’existence et de l’unicité d’une solution faible des équations de la thermoviscoélasticité dans une formulation mixte de type Hellinger- Reissner, la nouveauté par rapport au travail de M.E. Rognes et R. Winther (M3AS, 2010) étant ici l’apparition de la viscosité dans certains coefficients de l’équation constitutive, viscosité qui dépend dans ce contexte de la température absolue T(x, t) et donc en particulier du temps t. Enfin, nous considérons dans ce cadre le problème du contrôle optimal de la déformation du corps semi-transparent $\Omega$, en agissant sur la température absolue de la source radiative noire qui l’entoure. Nous prouvons l’existence d’un contrôle optimal et nous calculons la dérivée Fréchet de la fonctionnelle coût réduite. / This thesis begins with a rigorous mathematical analysis of the radiative heating of a semi-transparent body made of glass, by a black radiative source surrounding it. This requires the study of the coupling between quasi-steady radiative transfer boundary value problems with nonhomogeneous reflectivity boundary conditions (one for each wavelength band in the semi-transparent electromagnetic spectrum of the glass) and a nonlinear heat conduction evolution equation with a nonlinear Robin boundary condition which takes into account those wavelengths for which the glass behaves like an opaque body. We prove existence and uniqueness of the solution, and give also uniform bounds on the solution i.e. on the absolute temperature distribution inside the body and on the radiative intensities. Now, we consider the temperature $T_{S}$ of the black radiative source S surrounding the semi-transparent body $\Omega$ as the control variable. We adjust the absolute temperature distribution (x, t) 7! T(x, t) inside the semi-transparent body near a desired temperature distribution Td(·, ·) during the time interval of radiative heating ]0, tf [ by acting on $T_{S}$. In this respect, we introduce the appropriate cost functional and the set of admissible controls $T_{S}$, for which we prove the existence of optimal controls. Introducing the State Space and the State Equation, a first order necessary condition for a control $T_{S}$ : t 7! $T_{S}$ (t) to be optimal is then derived in the form of a Variational Inequality by using the Implicit Function Theorem and the adjoint problem. We come now to the goal problem which is the deformation of the semi-transparent body $\Omega$ by heating it with a black radiative source surrounding it. We introduce a weak mixed formulation of this thermoviscoelasticity problem and study the existence and uniqueness of its solution, the novelty here with respect to the work of M.E. Rognes et R. Winther (M3AS, 2010) being the apparition of the viscosity in some of the coefficients of the constitutive equation, viscosity which depends on the absolute temperature T(x, t) and thus in particular on the time t. Finally, we state in this setting the related optimal control problem of the deformation of the semi-transparent body $\Omega$, by acting on the absolute temperature of the black radiative source surrounding it. We prove the existence of an optimal control and we compute the Fréchet derivative of the associated reduced cost functional.