O aprendizado de máquina consiste em conceitos e técnicas que permitem aos computadores melhorar seu desempenho com a experiência, ou em outras palavras, aprender com dados. Duas de suas principais categorias são o aprendizado não-supervisionado e o semissupervisionado, que respectivamente consistem em inferir padrões em bases cujos dados não têm rótulo (classe) e classificar dados em bases parcialmente rotuladas. Embora muito estudado, trata-se de um campo repleto de desafios e com muitos tópicos abertos. Sistemas dinâmicos coletivos, por sua vez, são sistemas constituídos por muitos indivíduos, cada qual um sistema dinâmico por si só, de modo que todos eles agem coletivamente, ou seja, a ação de cada indivíduo é influenciada pela ação dos vizinhos. Uma característica notável desses sistemas é que padrões globais podem surgir espontaneamente das interações locais entre os indivíduos, fenômeno conhecido como emergência. Os desafios intrínsecos e a relevância do tema vêm motivando sua pesquisa em diversos ramos da ciência e da engenharia. Este trabalho de doutorado consiste no desenvolvimento e análise de modelos dinâmicos coletivos para o aprendizado de máquina, especificamente suas categorias não-supervisionada e semissupervisionada. As tarefas de segmentação de imagens e de detecção de comunidades em redes, que de certo modo podem ser entendidas como tarefas do aprendizado de máquina, são também abordadas. Em especial, desenvolvem-se modelos nos quais a movimentação dos objetos é determinada pela localização e velocidade de seus vizinhos. O sistema dinâmico assim modelado é então conduzido a um estado cujo padrão formado por seus indivíduos realça padrões subjacentes do conjunto de dados. Devido ao seu caráter auto-organizável, os modelos aqui desenvolvidos são robustos e as informações geradas durante o processo (valores das variáveis do sistema) são ricas e podem, por exemplo, revelar características para realizar soft labeling e determinar classes sobrepostas. / Machine learning consists of concepts and techniques that enable computers to improve their performance with experience, i.e., learn from data. Unsupervised and semi-supervised learning are important categories of machine learning, which respectively consists of inferring patterns in datasets whose data have no label (class) and classifying data in partially-labeled datasets. Although intensively studied, machine learning is still a field full of challenges and with many open topics. Collective dynamical systems, in turn, are systems made of a large group of individuals, each one a dynamical system by itself, such that all of them behave collectively, i.e., the action of each individual is influenced by the action of its neighbors. A remarkable feature of those systems is that global patterns may spontaneously emerge from the local interactions among individuals, a phenomenon known as emergence. Their relevance and intrinsic challenges motivate research in various branches of science and engineering. In this doctorate research, we develop and analyze collective dynamical models for their usage in machine-learning tasks, specifically unsupervised and semi-supervised ones. Image segmentation and network community detection are also addressed, as they are related to machine learning as well. In particular, we propose to work on models in which the objects motion is determined by the location and velocity of their neighbors. By doing so, the dynamical system reaches a configuration in which the patterns developed by the set of individuals highlight underlying patterns of the dataset. Due to their self-organizing nature, it is also expected that the models can be robust and the information generated during the process (values of the system variables) can be rich and reveal, for example, features to perform soft labeling and determine overlapping classes.

Ces dernières années, l’émergence des nouvelles technologies tels que la miniaturisation des composants, les dispositifs de communication sans fils, l’augmentation de la taille de stockage et les capacités de calcul, a permis la conception de systèmes multi-agents coopératifs de plus en plus complexes. L’un des plus grands axes de recherche dans cette thématique concerne la commande en formation de flottes de véhicules autonomes. Un grand nombre d’applications et de missions, civiles et militaires, telles que l’exploration, la surveillance, et la maintenance, ont alors été développées et réalisées dans des milieux variés (terre, air, eau). Durant l’exécution de ces tâches, les véhicules doivent interagir avec leur environnement et entre eux pour se coordonner. Les outils de communication disponibles disposent souvent d’une portée limitée. La préservation de la connexion au sein du groupe devient alors un des objectifs à satisfaire pour que la tâche puisse être accomplie avec succès. Une des possibilités pour garantir cette contrainte est le déplacement en formation permettant de préserver les distances et la structure géométrique du groupe. Il est toutefois nécessaire de disposer d’outils et de méthodes d’analyse et de commande de ces types de systèmes afin d’exploiter au maximum leurs potentiels. Cette thèse s’inscrit dans cette direction de recherche en présentant une synthèse et une analyse des systèmes dynamiques multi-agents et plus particulièrement la commande en formation de véhicules autonomes. Les lois de commande développées dans la littérature pour la commande en formation permettent d’accomplir un grand nombre de missions avec un niveau de performance élevé. Toutefois, si un défaut/défaillant apparaît dans la formation, ces lois de commandes peuvent s’avérer très limitées, engendrant un comportement instable du système. Le développement de commandes tolérantes aux défauts devient alors primordial pour maintenir les performances de commande en présence de défauts. Cette problématique sera traitée dans ce mémoire de thèse et concernera le développement et la conception de commandes en formation tolérantes au défaut dévolu à une flotte de véhicules autonomes suivant différente configuration/structuration / In recent years, the emergence of new technologies such as miniaturization of components, wireless communication devices, increased storage size and computing capabilities have allowed the design of increasingly complex cooperative multi-agent systems. One of the main research axes in this topic concerns the formation control of fleets of autonomous vehicles. Many applications and missions, civilian and military, such as exploration, surveillance, and maintenance, were developed and carried out in various environments. During the execution of these tasks, the vehicles must interact with their environment and among themselves to coordinate. The available communication tools are often limited in scope. The preservation of the connection within the group then becomes one of the objectives to be satisfied in order to carry out the task successfully. One of the possibilities to guarantee this constraint is the training displacement, which makes it possible to preserve the distances and the geometrical structure of the group. However, it is necessary to have tools and methods for analyzing and controlling these types of systems in order to make the most of their potential. This thesis is part of this research direction by presenting a synthesis and analysis of multi-agent dynamical systems and more particularly the formation control of autonomous vehicles. The control laws developed in the literature for formation control allow to carry out a large number of missions with a high level of performance. However, if a fault/failure occurs in the training, these control laws can be very limited, resulting in unstable system behavior. The development of fault tolerant controls becomes paramount to maintaining control performance in the presence of faults. This problem will be dealt with in more detail in this thesis and will concern the development and design of Fault tolerant controls devolved to a fleet of autonomous vehicles according to different configuration/structuring

Nous nous intéressons à l'étude de la dynamique des réseaux composés d'une multitude d'agents. Les motivations de ce travail trouvent leurs sources dans de nombreux domaines et notamment la biologie avec l'étude de l'émergence de comportements collectifs cohérents chez les animaux (vol en formation d'oiseaux migrateurs). Considérons un certain nombre d’agents (animaux) dont le comportement dynamique individuel peut être modélisé par une équation différentielle. Les agents communiquent : les liens sont représentés sous la forme d’un graphe dont les sommets sont les agents du système. Chaque agent a la connaissance de l’état des agents auxquels il est connecté et ajuste sa dynamique à l’aide de cette information. Des comportements collectifs peuvent alors émerger comme par exemple le phénomène de flocking (tous les agents se déplacent dans la même direction). Plusieurs modèles d'interaction ont été proposés, les plus connus étant le modèle de Viscek (1995) ou le modèle de Cucker-Smale (2007). L'étude de ces modèles repose généralement sur des méthodes d'analyse de stabilité des systèmes dynamiques ou des systèmes hybrides, lorsque le graphe de communication évolue dans le temps. Nous souhaitons dans cette thèse évaluer la robustesse de l'émergence de ces comportements collectifs en étudiant l'influence de divers facteurs: paramètres du modèle, topologie du graphe, nombre d'agents, présences de perturbations. Nous nous intéresserons notamment au phénomène de scission du groupe d'agents en plusieurs groupes d'agents coordonnés. / This thesis presents a study on multi-agent systems. Such systems find numerous applications such as multi-vehicle control in robotics, the design of smart distributed energy networks and the modeling of opinion dynamics. In a first part, we present new results regarding consensus theory which extend the recent work from Hendrickx and Tsitsiklis on cut-balanced consensus. Then, we apply the consensus system to the control of a fleet of vehicles. We present several results regarding velocity alignment (flocking). This study is based upon a graph robustness analysis in order to preserve the connectivity of the interaction network. This concept is of main importance in this study. In the last part, we state results froma collaborative work with a sociologist regarding the social network linked to the controversy concerning o-road motorized leisure in France. We study the link between the national and local scenes. To do so, we use large graph visualization tools and actor centrality measures.

In this thesis, we discuss models for multi-agent dynamical systems. We study the tracking/migration problem for flocks and a theoretical framework for design and analysis of flocking algorithm is presented. The interactions between agents in the systems are denoted by potential functions that act as distance functions, hence, the design of proper potential functions are crucial in modelling and analyzing the flocking problem for multi-agent dynamical systems. Constructions for both non-smooth potential functions and smooth potential functions with finite cut-off are investigated in detail. The main contributions of this thesis are to extend the literature of continuous flocking models with impulsive control and delay. Lyapunov function techniques and techniques for stability of continuous and impulsive switching system are used, we study the asymptotic stability of the equilibrium of our models with impulsive control and discovery that by applying impulsive control to Olfati-Saber's continuous model, we can remove the damping term and improve the performance by avoiding the deficiency caused by time delay in velocity sensing. Additionally, we discuss both free-flocking and constrained-flocking algorithm for multi-agent dynamical system, we extend literature results by applying velocity feedbacks which are given by the dynamical obstacles in the environment to our impulsive control and successfully lead to flocking with obstacle avoidance capability in a more energy-efficient way. Simulations are given to support our results, some conclusions are made and future directions are given.

The accuracy of target location and the coverage range of sensor network are two factors that affect each other in target tracking. When the flocking sensor network has a larger coverage area, it can increase the range of detecting target and the scope of environmental information. The network can also pass the information to a query source or other sensors which do not belong to the flocking network. However, the accuracy of measurements at sensors may be affected by the distances between the target and the sensors. We use mobile sensors as agents in flocking algorithm for target tracking. Every mobile sensor exchanges information with its neighbors, and keeps an appropriate separation distance with neighbors to maintain flocking. Flocking algorithm is a distributed control method for mobile sensor which can catch up the target and maintain flocking formation. In the thesis, we derive the cost function based on the accuracy of target positioning and range coverage. The proposed adaptive flocking algorithm combines the amount of information and the distance changes between neighbors based on the cost function. Each mobile sensor adaptively adjusts distance separation with all its neighbors within communication range. Sensors closer to the target shortens the separation distance between neighbors, therefore they will move toward the target and obtain better measurement. Kalman-consensus information filter is used for target positioning. The accuracy of target position can therefore be improved in the overall network. On the other hand, the sensors located far from the target will widen the distance separation between neighbors to expand the overall network area. In the thesis, we use Kalman-consensus information filter to estimate the state of a target, and use adaptive flocking algorithm for maintaining the formation of mobile sensors. Simulations show that adaptive flocking algorithm effectively improves location accuracy while maintaining approximate generally same coverage area when compared with other methods.

The objective of this research is to develop methods by which we can use blocks to approximate the shapes of 3D objects and to generate shape transformation animations. Two graphic tools are developed. One assists the animator in constructing 3D shapes with bricks of different sizes and matching up the different shapes. The other tool helps the animator generate a transformation animation of those bricks. Using polygon shape data, these tools can procedurally place the bricks and control their animation. Several different methods for animation are introduced. Those methods provide different ways to generate animation paths of the blocks. The no path animation and the straight path animation are easy for the animator to create and the animation time is easily controlled. The flocking animation will provide more interesting effect.

In this thesis, we discuss models for multi-agent dynamical systems. We study the tracking/migration problem for flocks and a theoretical framework for design and analysis of flocking algorithm is presented. The interactions between agents in the systems are denoted by potential functions that act as distance functions, hence, the design of proper potential functions are crucial in modelling and analyzing the flocking problem for multi-agent dynamical systems. Constructions for both non-smooth potential functions and smooth potential functions with finite cut-off are investigated in detail. The main contributions of this thesis are to extend the literature of continuous flocking models with impulsive control and delay. Lyapunov function techniques and techniques for stability of continuous and impulsive switching system are used, we study the asymptotic stability of the equilibrium of our models with impulsive control and discovery that by applying impulsive control to Olfati-Saber's continuous model, we can remove the damping term and improve the performance by avoiding the deficiency caused by time delay in velocity sensing. Additionally, we discuss both free-flocking and constrained-flocking algorithm for multi-agent dynamical system, we extend literature results by applying velocity feedbacks which are given by the dynamical obstacles in the environment to our impulsive control and successfully lead to flocking with obstacle avoidance capability in a more energy-efficient way. Simulations are given to support our results, some conclusions are made and future directions are given.

Flocking birds in a virtual world through adaption of Craig Reynolds, Boids algorithm. In this paper I will demonstrate how I adapted the Boids algorithm to an already existing code to create a flocking behaviour and discuss options to the solutions. I have been given a virtual world with existing birds that is moving around based on random elements that in a way mimics flying.

Cooperative behavior arises from the interactions of single units that globally produce a complex dynamics in which the system acts as a whole. As an archetype I refer to a flock of birds. As a result of cooperation the whole flock gets special abilities that the single individuals would not have if they were alone. This research work led to the discovery that the function of a flock, and more in general, that of cooperative systems, surprisingly rests on the occurrence of organizational collapses. In this study, I used cooperative systems based on self-propelled particle models (the flock models) which have been proved to be virtually equivalent to sociological network models mimicking the decision making processes (the decision making model). The critical region is an intermediate condition between a highly disordered state and a strong ordered one. At criticality the waiting times distribution density between two consecutive collapses shows an inverse power law form with an anomalous statistical behavior. The scientific evidences are based on measures of information theory, correlation in time and space, and fluctuation statistical analysis. In order to prove the benefit for a system to live at criticality, I made a flock system interact with another similar system, and then observe the information transmission for different disturbance values. I proved that at criticality the transfer of information gets the maximal efficiency. As last step, the flock model has been shown that, despite its simplicity, is sufficiently a realistic model as proved via the use of 3D simulations and computer animations.

----- / As processor performance capabilities can only be increased through the useof a multicore architecture, software needs to be developed to utilize the parallelismoffered by the additional cores. Especially game developers need toseize this opportunity to save cycles and decrease the general rendering time.One of the existing advances towards this potential has been the creation ofmultithreaded game engines that take advantage of the additional processingunits. In such engines, different branches of the game loop are parallelized.However, the specifics of the parallel design patterns used are not outlined.Neither are any ideas of how to combine these patterns proposed. Thesemissing factors are addressed in this article, to provide a guideline for whento use which one of two parallel design patterns; fork-join and pipeline parallelism.Through a collection of data and a comparison using the metricsspeedup and efficiency, conclusions were derived that shed light on the waysin which a typical part of a game loop most efficiently can be organized forparallel execution through the use of different parallel design patterns. Thepipeline and fork-join patterns were applied respectively in a variety of testcases for two branches of a game loop: a BOIDS system and an animationsystem.