Cette thèse s'intègre dans le cadre du projet européen EUREKA "High Dynamic Range - Low NoiseCMOS imagers", qui a pour but de développer de nouvelles approches de fabrication de capteursd'images CMOS à haute performance. L'objectif de la thèse est la conception d'un système de visiontemps réel à grande gamme dynamique (HDR). L'axe principal sera la reconstruction, en temps réelet à la cadence du capteur (60 images/sec), d'une vidéo à grande dynamique sur une architecturede calcul embarquée.La plupart des capteurs actuels produisent une image numérique qui n'est pas capable de reproduireles vraies échelles d'intensités lumineuses du monde réel. De la même manière, les écrans, impri-mantes et afficheurs courants ne permettent pas la restitution effective d'une gamme tonale étendue.L'approche envisagée dans cette thèse est la capture multiple d'images acquises avec des tempsd'exposition différents permettant de palier les limites des dispositifs actuels.Afin de concevoir un système capable de s'adapter temporellement aux conditions lumineuses,l'étude d'algorithmes dédiés à la grande dynamique, tels que les techniques d'auto exposition, dereproduction de tons, en passant par la génération de cartes de radiances est réalisée. Le nouveausystème matériel de type "smart caméra" est capable de capturer, générer et restituer du contenu àgrande dynamique dans un contexte de parallélisation et de traitement des flux vidéos en temps réel

Organizations owning HPC infrastructures are facing difficulties in managing their resources. These difficulties come from the need to provide concurrent resource access to different application types while considering that users might have different performance objectives for their applications. Cloud computing brings more flexibility and better resource control, promising to improve the user's satisfaction in terms of perceived Quality of Service. Nevertheless, current cloud solutions provide limited support for users to express or use various resource management policies and they don't provide any support for application performance objectives.In this thesis, we present an approach that addresses this challenge in an unique way. Our approach provides a fully decentralized resource control by allocating resources through a proportional-share market, while applications run in autonomous virtual environments capable of scaling the application demand according to user performance objectives.The combination of currency distribution and dynamic resource pricing ensures fair resource utilization.We evaluated our approach in simulation and on the Grid'5000 testbed. Our results show that our approach can enable the co-habitation of different resource usage policies on the infrastructure, improving resource utilisation.

Learning from data streams is emerging as an important application area. When the environment changes, it is necessary to rely on on-line learning with the capability to adapt to changing conditions a.k.a. concept drifts. Adapting to concept drifts entails forgetting some or all of the old acquired knowledge when the concept changes while accumulating knowledge regarding the supposedly stationary underlying concept. This tradeoff is called the stability-plasticity dilemma. Ensemble methods have been among the most successful approaches. However, the management of the ensemble which ultimately controls how past data is forgotten has not been thoroughly investigated so far. Our work shows the importance of the forgetting strategy by comparing several approaches. The results thus obtained lead us to propose a new ensemble method with an enhanced forgetting strategy to adapt to concept drifts. Experimental comparisons show that our method compares favorably with the well-known state-of-the-art systems. The majority of previous works focused only on means to detect changes and to adapt to them. In our work, we go one step further by introducing a meta-learning mechanism that is able to detect relevant states of the environment, to recognize recurring contexts and to anticipate likely concepts changes. Hence, the method we suggest, deals with both the challenge of optimizing the stability-plasticity dilemma and with the anticipation and recognition of incoming concepts. This is accomplished through an ensemble method that controls a ensemble of incremental learners. The management of the ensemble of learners enables one to naturally adapt to the dynamics of the concept changes with very few parameters to set, while a learning mechanism managing the changes in the ensemble provides means for the anticipation of, and the quick adaptation to, the underlying modification of the context.

The present dissertation deals with how to use the precious wireless resources that are usually wasted by under-utilization of networks. We have been particularly interested by all resources that can be used in an opportunistic fashion using different technologies. We have designed new schemes for better and more efficient use of wireless systems by providing mathematical frameworks. In the first part, We have been interested in cognitive radio networks, where a cellular service provider can lease a part of its resources to secondary users or virtual providers. In the second part, we have chosen delay-tolerant networks as a solution to reduce the pressure on the cell traffic, where mobile users come to use available resources effectively and with a cheaper cost. We have focused on optimal strategy for smartphones in hybrid wireless networks. In the last part, an alternative to delay-tolerant networks, specially in regions that are not covered by the cellular network, is to use Ad-hoc networks. Indeed, they can be used as an extension of the coverage area. We have developed a new analytical modeling of the IEEE 802.11e DCF/EDCF. We have investigated the intricate interactions among layers by building a general cross-layered framework to represent multi-hop ad hoc networks with asymmetric topology and traffic

La vidéo 3D a suscité un intérêt croissant durant ces dernières années. Grâce au développement récent des écrans stéréoscopiques et auto-stéréoscopiques, la vidéo 3D fournit une sensation réaliste de profondeur à l'utilisateur et une navigation virtuelle autour de la scène observée. Cependant de nombreux défis techniques existent encore. Ces défis peuvent être liés à l'acquisition de la scène et à sa représentation d'une part ou à la transmission des données d'autre part. Dans le contexte de la représentation de scènes naturelles, de nombreux efforts ont été fournis afin de surmonter ces difficultés. Les méthodes proposées dans la littérature peuvent être basées image, géométrie ou faire appel à des représentations combinant image et géométrie. L'approche adoptée dans cette thèse consiste en une méthode hybride s'appuyant sur l'utilisation des séquences multi-vues plus profondeur MVD (Multi-view Video plus Depth) afin de conserver le photo-réalisme de la scène observée, combinée avec un modèle géométrique, à base de maillage triangulaire, renforçant ainsi la compacité de la représentation. Nous supposons que les cartes de profondeur des données MVD fournies sont fiables et que les caméras utilisées durant l'acquisition sont calibrées, les paramètres caméras sont donc connus, mais les images correspondantes ne sont pas nécessairement rectifiées. Nous considérerons ainsi le cas général où les caméras peuvent être parallèles ou convergentes. Les contributions de cette thèse sont les suivantes. D'abord, un schéma volumétrique dédié à la fusion des cartes de profondeur en une surface maillée est proposé. Ensuite, un nouveau schéma de plaquage de texture multi-vues est proposé. Finalement, nous abordons à l'issue ce ces deux étapes de modélisation, la transmission proprement dite et comparons les performances de notre schéma de modélisation avec un schéma basé sur le standard MPEG-MVC, état de l'art dans la compression de vidéos multi-vues.

Static program analysis is a core technology for both verifying and finding errors in programs but most static analyzers are complex pieces of software that are not without error. A Static analysis formalised as an abstract interpreter can be proved sound, however such proofs are significantly harder to do on the actual implementation of an analyser. To alleviate this problem we propose to generate Verification Conditions (VCs, formulae valid only if the results of the analyser are correct) and to discharge them using an Automated Theorem Prover (ATP). We generate formulae in Many-Sorted First-Order Logic (MSFOL), a logic that has been successfully used in deductive program verification. MSFOL is expressive enough to describe the results of complex analyses and to formalise the operational semantics of object-oriented languages. Using the same logic for both tasks allows us to prove the soundness of the VC generator using deductive verification tools. To ensure that VCs can be automatically discharged for complex analyses of the heap, we introduce a VC calculus that produces formulae belonging to a decidable fragment of MSFOL. Furthermore, to be able to certify different analyses with the same calculus, we describe a family of analyses with a parametric concretisation function and instrumentation of the semantics. To improve the reliability of ATPs, we also studied the result certification of Satisfiability Modulo Theory solvers, a family of ATPs dedicated to MSFOL. We propose a modular proof-system and a modular proof-verifier programmed and proved correct in Coq, that rely on exchangeable verifiers for each of the underlying theories.

L'analyse statique a pour but de vérifier qu'un programme a le comportement souhaité c.à.d. satisfait des propriétés de sûreté. Toutefois, inférer les propriétés vérifiées par un programme est un problème difficile : le théorème de Rice énonce que toute propriété non triviale d'un langage de programmation Turing-complet est indécidable. Afin de contourner cette difficulté, les analyses statiques effectuent des approximations des comportements possibles du programme. La théorie de l'interprétation abstraite permet de donner un cadre formel à ces approximations. Cette théorie, introduite par Cousot & Cousot propose un cadre d'approximation basé sur la notion de treillis, de connexion de Galois et de calculs de points fixes par itération. Ce cadre permet de définir la qualité des approximations effectuées et notamment la notion de meilleure approximation. À l'opposé, les notions quantitatives n'apparaissent pas naturellement dans ce cadre. Nous nous sommes donc posés la question de l'inférence, par analyse statique, de propriétés s'exprimant de manière quantitative (telles que l'utilisation de la mémoire ou le temps d'exécution).

An end-to-end guarantee of software correctness by formal verification must consider two sources of bugs. First, the verification tool must be correct. Second, programs are often verified at the source level, before being compiled. Hence, compilers should also be trustworthy. Verifiers and compilers' complexity is increasing. To simplify code analysis and manipulation, these tools rely on intermediate representations (IR) of programs, that provide structural and semantic properties. This thesis gives a formal, semantic account on IRs, so that they can also be leveraged in the formal proof of such tools. We first study a register-based IR of Java bytecode used in compilers and verifiers. We specify the IR generation by a semantic theorem stating what the transformation preserves, e.g. object initialization or exceptions, but also what it modifies and how, e.g. object allocation. We implement this IR in Sawja, a Java static analysis toolbench. Then, we study the Static Single Assignment (SSA) form, an IR widely used in modern compilers and verifiers. We implement and prove in Coq an SSA middle-end for the CompCert C compiler. For the proof of SSA optimizations, we identify a key semantic property of SSA, allowing for equational reasoning. Finally, we study the semantics of concurrent Java IRs. Due to instruction reorderings performed by the compiler and the hardware, the current definition of the Java Memory Model (JMM) is complex, and unfortunately formally flawed. Targetting x86 architectures, we identify a subset of the JMM that is intuitive and tractable in formal proofs. We characterize the reorderings it allows, and factor out a proof common to the IRs of a compiler.

Distributed systems lie at the heart of many modern applications (social networks, web services, etc.). However, developers face many challenges in implementing distributed systems. The major one we focus on is avoiding the erroneous behaviors, that do not appear in the requirements of the distributed system, and that are caused by the concurrency between the entities of this system. The automatic code generation from requirements of distributed systems remains an old dream. In this thesis, we consider the automatic generation of a skeleton of code covering the interactions between the entities of a distributed system. This allows us to avoid the erroneous behaviors caused by the concurrency. Then, in a later step, this skeleton can be completed by adding and debugging the code that describes the local actions happening on each entity independently from its interactions with the other entities. The automatic generation that we consider is from a scenario-based specification that formally describes the interactions within informal requirements of a distributed system. We choose High-level Message Sequence Charts (HMSCs for short) as a scenario-based specification for the many advantages that they present: namely the clear graphical and textual representations, and the formal semantics. The code generation from HMSCs requires an intermediate step, called "Synthesis" which is their transformation into an abstract machine model that describes the local views of the interactions by each entity (A machine representing an entity defines sequences of messages sending and reception). Then, from the abstract machine model, the skeleton's code generation becomes an easy task. A very intuitive abstract machine model for the synthesis of HMSCs is the Communicating Finite State Machine (CFSMs). However, the synthesis from HMSCs into CFSMs may produce programs with more behaviors than described in the specifications in general. We thus restrict then our specifications to a sub-class of HMSCs named "local HMSC". We show that for any local HMSC, behaviors can be preserved by addition of communication controllers that intercept messages to add stamping information before resending them. We then propose a new technique that we named "localization" to transform an arbitrary HMSC specification into a local HMSC, hence allowing correct synthesis. We show that this transformation can be automated as a constraint optimization problem. The impact of modifications brought to the original specification can be minimized with respect to a cost function. Finally, we have implemented the synthesis and the localization approaches into an existing tool named SOFAT. We have, in addition, implemented to SOFAT the automatic code generation of a Promela code and a JAVA code for REST based web services from HMSCs.

Big Data can be characterized by 3 V's. * Big Volume refers to the unprecedented growth in the amount of data. * Big Velocity refers to the growth in the speed of moving data in and out management systems. * Big Variety refers to the growth in the number of different data formats. Managing Big Data requires fundamental changes in the architecture of data management systems. Data storage should continue being innovated in order to adapt to the growth of data. They need to be scalable while maintaining high performance regarding data accesses. This thesis focuses on building scalable data management systems for Big Data. Our first and second contributions address the challenge of providing efficient support for Big Volume of data in data-intensive high performance computing (HPC) environments. Particularly, we address the shortcoming of existing approaches to handle atomic, non-contiguous I/O operations in a scalable fashion. We propose and implement a versioning-based mechanism that can be leveraged to offer isolation for non-contiguous I/O without the need to perform expensive synchronizations. In the context of parallel array processing in HPC, we introduce Pyramid, a large-scale, array-oriented storage system. It revisits the physical organization of data in distributed storage systems for scalable performance. Pyramid favors multidimensional-aware data chunking, that closely matches the access patterns generated by applications. Pyramid also favors a distributed metadata management and a versioning concurrency control to eliminate synchronizations in concurrency. Our third contribution addresses Big Volume at the scale of the geographically distributed environments. We consider BlobSeer, a distributed versioning-oriented data management service, and we propose BlobSeer-WAN, an extension of BlobSeer optimized for such geographically distributed environments. BlobSeer-WAN takes into account the latency hierarchy by favoring locally metadata accesses. BlobSeer-WAN features asynchronous metadata replication and a vector-clock implementation for collision resolution. To cope with the Big Velocity characteristic of Big Data, our last contribution feautures DStore, an in-memory document-oriented store that scale vertically by leveraging large memory capability in multicore machines. DStore demonstrates fast and atomic complex transaction processing in data writing, while maintaining high throughput read access. DStore follows a single-threaded execution model to execute update transactions sequentially, while relying on a versioning concurrency control to enable a large number of simultaneous readers.