Cette thèse présente nos contributions àl’inférence et l’apprentissage des modèles graphiquesen vision artificielle. Tout d’abord, nous proposons unenouvelle classe d’algorithmes de décomposition pour résoudrele problème d’appariement de graphes et d’hypergraphes,s’appuyant sur l’algorithme des directionsalternées (ADMM) non convexe. Ces algorithmes sontefficaces en terme de calcul et sont hautement parallélisables.En outre, ils sont également très générauxet peuvent être appliqués à des fonctionnelles d’énergiearbitraires ainsi qu’à des contraintes de correspondancearbitraires. Les expériences montrent qu’ils surpassentles méthodes de pointe existantes sur des benchmarkspopulaires. Ensuite, nous proposons une relaxationcontinue non convexe pour le problème d’estimationdu maximum a posteriori (MAP) dans les champsaléatoires de Markov (MRFs). Nous démontrons quecette relaxation est serrée, c’est-à-dire qu’elle est équivalenteau problème original. Cela nous permet d’appliquerdes méthodes d’optimisation continue pour résoudrele problème initial discret sans perte de précisionaprès arrondissement. Nous étudions deux méthodes degradient populaires, et proposons en outre une solutionplus efficace utilisant l’ADMM non convexe. Les expériencessur plusieurs problèmes réels démontrent quenotre algorithme prend l’avantage sur ceux de pointe,dans différentes configurations. Finalement, nous proposonsune méthode d’apprentissage des paramètres deces modèles graphiques avec des données d’entraînement,basée sur l’ADMM non convexe. Cette méthodeconsiste à visualiser les itérations de l’ADMM commeune séquence d’opérations différenciables, ce qui permetde calculer efficacement le gradient de la perted’apprentissage par rapport aux paramètres du modèle.L’apprentissage peut alors utiliser une descente de gradientstochastique. Nous obtenons donc un frameworkunifié pour l’inférence et l’apprentissage avec l’ADMMnon-convexe. Grâce à sa flexibilité, ce framework permetégalement d’entraîner conjointement de-bout-en-boutun modèle graphique avec un autre modèle, telqu’un réseau de neurones, combinant ainsi les avantagesdes deux. Nous présentons des expériences sur un jeude données de segmentation sémantique populaire, démontrantl’efficacité de notre méthode. / This thesis presents our contributions toinference and learning of graph-based models in computervision. First, we propose a novel class of decompositionalgorithms for solving graph and hypergraphmatching based on the nonconvex alternating directionmethod of multipliers (ADMM). These algorithms arecomputationally efficient and highly parallelizable. Furthermore,they are also very general and can be appliedto arbitrary energy functions as well as arbitraryassignment constraints. Experiments show that theyoutperform existing state-of-the-art methods on popularbenchmarks. Second, we propose a nonconvex continuousrelaxation of maximum a posteriori (MAP) inferencein discrete Markov random fields (MRFs). Weshow that this relaxation is tight for arbitrary MRFs.This allows us to apply continuous optimization techniquesto solve the original discrete problem withoutloss in accuracy after rounding. We study two populargradient-based methods, and further propose a more effectivesolution using nonconvex ADMM. Experimentson different real-world problems demonstrate that theproposed ADMM compares favorably with state-of-theartalgorithms in different settings. Finally, we proposea method for learning the parameters of these graphbasedmodels from training data, based on nonconvexADMM. This method consists of viewing ADMM iterationsas a sequence of differentiable operations, whichallows efficient computation of the gradient of the trainingloss with respect to the model parameters, enablingefficient training using stochastic gradient descent. Atthe end we obtain a unified framework for inference andlearning with nonconvex ADMM. Thanks to its flexibility,this framework also allows training jointly endto-end a graph-based model with another model suchas a neural network, thus combining the strengths ofboth. We present experiments on a popular semanticsegmentation dataset, demonstrating the effectivenessof our method.

The advances in data-collecting technologies provides great opportunities to access large sample-size data sets with high dimensionality. Variable selection is an important procedure to extract useful knowledge from such complex data. While in many real-data applications, appropriate selection of variables should facilitate the model interpretation and computation efficiency. It is thus important to incorporate domain knowledge of underlying data generation mechanism to select key variables for improving the model performance. However, general variable selection techniques, such as the best subset selection and the Lasso, often do not take the underlying data generation mechanism into considerations. This thesis proposal aims to develop statistical modeling methodologies with a focus on the structured variable selection towards better model interpretation and computation efficiency. Specifically, this thesis proposal consists of three parts: an additive heredity model with coefficients incorporating the multi-level data, a regularized dynamic generalized linear model with piecewise constant functional coefficients, and a structured variable selection method within the best subset selection framework. In Chapter 2, an additive heredity model is proposed for analyzing mixture-of-mixtures (MoM) experiments. The MoM experiment is different from the classical mixture experiment in that the mixture component in MoM experiments, known as the major component, is made up of sub-components, known as the minor components. The proposed model considers an additive structure to inherently connect the major components with the minor components. To enable a meaningful interpretation for the estimated model, we apply the hierarchical and heredity principles by using the nonnegative garrote technique for model selection. The performance of the additive heredity model was compared to several conventional methods in both unconstrained and constrained MoM experiments. The additive heredity model was then successfully applied in a real problem of optimizing the Pringlestextsuperscript{textregistered} potato crisp studied previously in the literature. In Chapter 3, we consider the dynamic effects of variables in the generalized linear model such as logistic regression. This work is motivated from the engineering problem with varying effects of process variables to product quality caused by equipment degradation. To address such challenge, we propose a penalized dynamic regression model which is flexible to estimate the dynamic coefficient structure. The proposed method considers modeling the functional coefficient parameter as piecewise constant functions. Specifically, under the penalized regression framework, the fused lasso penalty is adopted for detecting the changes in the dynamic coefficients. The group lasso penalty is applied to enable a sparse selection of variables. Moreover, an efficient parameter estimation algorithm is also developed based on alternating direction method of multipliers. The performance of the dynamic coefficient model is evaluated in numerical studies and three real-data examples. In Chapter 4, we develop a structured variable selection method within the best subset selection framework. In the literature, many techniques within the LASSO framework have been developed to address structured variable selection issues. However, less attention has been spent on structured best subset selection problems. In this work, we propose a sparse Ridge regression method to address structured variable selection issues. The key idea of the proposed method is to re-construct the regression matrix in the angle of experimental designs. We employ the estimation-maximization algorithm to formulate the best subset selection problem as an iterative linear integer optimization (LIO) problem. the mixed integer optimization algorithm as the selection step. We demonstrate the power of the proposed method in various structured variable selection problems. Moverover, the proposed method can be extended to the ridge penalized best subset selection problems. The performance of the proposed method is evaluated in numerical studies. / Doctor of Philosophy / The advances in data-collecting technologies provides great opportunities to access large sample-size data sets with high dimensionality. Variable selection is an important procedure to extract useful knowledge from such complex data. While in many real-data applications, appropriate selection of variables should facilitate the model interpretation and computation efficiency. It is thus important to incorporate domain knowledge of underlying data generation mechanism to select key variables for improving the model performance. However, general variable selection techniques often do not take the underlying data generation mechanism into considerations. This thesis proposal aims to develop statistical modeling methodologies with a focus on the structured variable selection towards better model interpretation and computation efficiency. The proposed approaches have been applied to real-world problems to demonstrate their model performance.

Le démélange spectral est l’un des problèmes centraux pour l’exploitation des images hyperspectrales. En raison de la faible résolution spatiale des imageurs hyperspectraux en télédetection, la surface représentée par un pixel peut contenir plusieurs matériaux. Dans ce contexte, le démélange consiste à estimer les spectres purs (les end members) ainsi que leurs fractions (les abondances) pour chaque pixel de l’image. Le but de cette thèse estde proposer de nouveaux algorithmes de démélange qui visent à améliorer l’estimation des spectres purs et des abondances. En particulier, les algorithmes de démélange proposés s’inscrivent dans le cadre du démélange non-supervisé et non-linéaire. Dans un premier temps, on propose un algorithme de démelange non-supervisé dans lequel une régularisation favorisant la parcimonie des groupes est utilisée pour identifier les spectres purs parmi les observations. Une extension de ce premier algorithme permet de prendre en compte la présence du bruit parmi les observations choisies comme étant les plus pures. Dans un second temps, les connaissances a priori des ressemblances entre les spectres à l’échelle localeet non-locale ainsi que leurs positions dans l’image sont exploitées pour construire un graphe adapté à l’image. Ce graphe est ensuite incorporé dans le problème de démélange non supervisé par le biais d’une régularisation basée sur le Laplacian du graphe. Enfin, deux algorithmes de démélange non-linéaires sont proposés dans le cas supervisé. Les modèles de mélanges non-linéaires correspondants incorporent des fonctions à valeurs vectorielles appartenant à un espace de Hilbert à noyaux reproduisants. L’intérêt de ces fonctions par rapport aux fonctions à valeurs scalaires est qu’elles permettent d’incorporer un a priori sur la ressemblance entre les différentes fonctions. En particulier, un a priori spectral, dans un premier temps, et un a priori spatial, dans un second temps, sont incorporés pour améliorer la caractérisation du mélange non-linéaire. La validation expérimentale des modèles et des algorithmes proposés sur des données synthétiques et réelles montre une amélioration des performances par rapport aux méthodes de l’état de l’art. Cette amélioration se traduit par une meilleure erreur de reconstruction des données / Spectral unmixing has been an active field of research since the earliest days of hyperspectralremote sensing. It is concerned with the case where various materials are found inthe spatial extent of a pixel, resulting in a spectrum that is a mixture of the signatures ofthose materials. Unmixing then reduces to estimating the pure spectral signatures and theircorresponding proportions in every pixel. In the hyperspectral unmixing jargon, the puresignatures are known as the endmembers and their proportions as the abundances. Thisthesis focuses on spectral unmixing of remotely sensed hyperspectral data. In particular,it is aimed at improving the accuracy of the extraction of compositional information fromhyperspectral data. This is done through the development of new unmixing techniques intwo main contexts, namely in the unsupervised and nonlinear case. In particular, we proposea new technique for blind unmixing, we incorporate spatial information in (linear and nonlinear)unmixing, and we finally propose a new nonlinear mixing model. More precisely, first,an unsupervised unmixing approach based on collaborative sparse regularization is proposedwhere the library of endmembers candidates is built from the observations themselves. Thisapproach is then extended in order to take into account the presence of noise among theendmembers candidates. Second, within the unsupervised unmixing framework, two graphbasedregularizations are used in order to incorporate prior local and nonlocal contextualinformation. Next, within a supervised nonlinear unmixing framework, a new nonlinearmixing model based on vector-valued functions in reproducing kernel Hilbert space (RKHS)is proposed. The aforementioned model allows to consider different nonlinear functions atdifferent bands, regularize the discrepancies between these functions, and account for neighboringnonlinear contributions. Finally, the vector-valued kernel framework is used in orderto promote spatial smoothness of the nonlinear part in a kernel-based nonlinear mixingmodel. Simulations on synthetic and real data show the effectiveness of all the proposedtechniques

Nous proposons dans cette thèse de développer une méthode de restauration aveugle d'images flouées et bruitées où aucune connaissance a priori n'est exigée. Ce manuscrit est composé de trois chapitres : le 1er chapitre est consacré aux travaux de l'état de l'art. Les approches d'optimisation pour la résolution du problème de restauration y sont d'abord discutées. Ensuite les principales méthodes de restauration, dites semi-aveugles car nécessitant un minimum de connaissance a priori sont analysées. Parmi ces méthodes, cinq sont retenues pour évaluation. Le 2ème chapitre est dédié à la comparaison des performances des méthodes retenues dans le chapitre précédent. Les principaux critères objectifs d'évaluation de la qualité des images restaurées sont présentés. Parmi ces critères, la norme L1 de l'erreur d'estimation est sélectionnée. L'étude comparative menée sur une banque d'images monochromes, dégradées artificiellement par deux fonctions floues de supports différents et trois niveaux de bruit a permis de mettre en évidence les deux méthodes les plus pertinentes. La première repose sur une approche alternée mono-échelle où la PSF et l'image sont estimées dans une seule étape. La seconde utilise une approche hybride multi-échelle qui consiste tout d'abord à estimer de manière alternée la PSF et une image latente, puis dans une étape suivante séquentielle, à restaurer l'image. Dans l'étude comparative conduite, l'avantage revient à cette dernière. Les performances de ces méthodes serviront de référence pour comparer ensuite la méthode développée. Le 3ème chapitre porte sur la méthode développée. Nous avons cherché à rendre aveugle l'approche hybride retenue dans le chapitre précédent tout en améliorant la qualité d'estimation de la PSF et de l'image restaurée. Les contributions ont porté sur plusieurs points. Une première série d'améliorations concerne la redéfinition des échelles, celle de l'initialisation de l'image latente à chaque niveau d'échelle, l'évolution des paramètres pour la sélection des contours pertinents servant de support à l'estimation de la PSF et enfin, la définition d'un critère d'arrêt aveugle. Une seconde série de contributions a porté sur l'estimation aveugle des deux paramètres de régularisation impliqués pour éviter d'avoir à les fixer empiriquement. Chaque paramètre est associé à une fonction coût distincte l'une pour l'estimation de la PSF et la seconde pour l'estimation d'une image latente. Dans l'étape séquentielle qui suit, nous avons cherché à affiner le support de la PSF estimée dans l'étape alternée, avant de l'exploiter dans le processus de restauration de l'image. A ce niveau, la seule connaissance a priori nécessaire est une borne supérieure du support de la PSF. Les différentes évaluations conduites sur des images monochromes et hyperspectrales dégradées artificiellement par plusieurs flous de type mouvement, de supports différents, montrent une nette amélioration de la qualité de restauration obtenue par l'approche développée par rapport aux deux meilleures approches de l'état de l'art retenues. / We propose in this thesis manuscript to develop a blind restoration method of single component blurred and noisy images where no prior knowledge is required. This manuscript is composed of three chapters: the first chapter focuses on state-of-art works. The optimization approaches for resolving the restoration problem are discussed first. Then, the main methods of restoration, so-called semi-blind ones because requiring a minimum of a priori knowledge are analysed. Five of these methods are selected for evaluation. The second chapter is devoted to comparing the performance of the methods selected in the previous chapter. The main objective criteria for evaluating the quality of the restored images are presented. Of these criteria, the l1 norm for the estimation error is selected. The comparative study conducted on a database of monochromatic images, artificially degraded by two blurred functions with different support size and three levels of noise, revealed the most two relevant methods. The first one is based on a single-scale alternating approach where both the PSF and the image are estimated alternatively. The second one uses a multi-scale hybrid approach, which consists first of alternatingly estimating the PSF and a latent image, then in a sequential next step, restoring the image. In the comparative study performed, the benefit goes to the latter. The performance of both these methods will be used as references to then compare the newly designed method. The third chapter deals with the developed method. We have sought to make the hybrid approach retained in the previous chapter as blind as possible while improving the quality of estimation of both the PSF and the restored image. The contributions covers a number of points. A first series concerns the redefinition of the scales that of the initialization of the latent image at each scale level, the evolution of the parameters for the selection of the relevant contours supporting the estimation of the PSF and finally the definition of a blind stop criterion. A second series of contributions concentrates on the blind estimation of the two regularization parameters involved in order to avoid having to fix them empirically. Each parameter is associated with a separate cost function either for the PSF estimation or for the estimation of a latent image. In the sequential step that follows, we refine the estimation of the support of the PSF estimated in the previous alternated step, before exploiting it in the process of restoring the image. At this level, the only a priori knowledge necessary is a higher bound of the support of the PSF. The different evaluations performed on monochromatic and hyperspectral images artificially degraded by several motion-type blurs with different support sizes, show a clear improvement in the quality of restoration obtained by the newly designed method in comparison to the best two state-of-the-art methods retained.

<p>In the US, seven regional transmission organizations (RTOs) operate wholesale electricity markets within three largely independent transmission systems, the largest of which includes five RTO regions and many vertically integrated utilities.</p> <p>RTOs operate a day-ahead and a real-time market. In the day-ahead market, generation and demand-side resources are optimally scheduled based on bids and offers for the next day. Those schedules are adjusted according to actual operating conditions in the real-time market. Both markets involve a unit commitment calculation, a mixed integer program that determines which generators will be online, and an economic dispatch calculation, an optimization determines the output of each online generator for every interval and calculates locational marginal prices (LMPs).</p> <p>The use of LMPs for the management of congestion in RTO transmission systems has brought efficiency and transparency to the operation of electric power systems and provides price signals that highlight the need for investment in transmission and generation. Through this work, we aim to extend these efficiency and transparency gains to the coordination across RTOs. Existing market-based inter-regional coordination schemes are limited to incremental changes in real-time markets. </p> <p>We propose a multi-regional unit-commitment that enables coordination in the day-ahead timeframe by applying a distributed approach to approximate a system-wide optimal commitment and dispatch while allowing each region to largely maintain their own rules, model only internal transmission up to the boundary, and keep sensitive financial information confidential. A heuristic algorithm based on an extension of the alternating directions method of multipliers (ADMM) for the mixed integer program is applied to the unit commitment. </p> The proposed coordinated solution was simulated and compared to the ideal single-market scenario and to a representation of the current uncoordinated solution, achieving at least 58% of the maximum potential savings, which, in terms of the annual cost of electric generation in the US, could add up to nearly $7 billion per year. In addition to the coordinated day-ahead solution, we develop a distributed solution for financial transmission rights (FTR) auctions with minimal information sharing across RTOs that constitutes the first known work to provide a viable option for market participants to seamlessly hedge price variability exposure on cross-border transactions.

Title: Image Deblurring in Demanding Conditions Author: Jan Kotera Department: Institute of Information Theory and Automation, Czech Academy of Sciences Supervisor: Doc. Ing. Filip Šroubek, Ph.D., DSc., Institute of Information Theory and Automation, Czech Academy of Sciences Abstract: Image deblurring is a computer vision task consisting of removing blur from image, the objective is to recover the sharp image corresponding to the blurred input. If the nature and shape of the blur is unknown and must be estimated from the input image, image deblurring is called blind and naturally presents a more difficult problem. This thesis focuses on two primary topics related to blind image deblurring. In the first part we work with the standard image deblurring based on the common convolution blur model and present a method of increasing robustness of the deblur- ring to phenomena violating the linear acquisition model, such as for example inten- sity clipping caused by sensor saturation in overexposed pixels. If not properly taken care of, these effects significantly decrease accuracy of the blur estimation and visual quality of the restored image. Rather than tailoring the deblurring method explicitly for each particular type of acquisition model violation we present a general approach based on flexible automatic...