Le déplacement collectif est l'un des phénomènes les plus remarquables de la nature. Il a été observé pour de nombreuses espèces animales, comme les essaims de bactéries, l'agrégation des fourmis, les bancs de poissons, les nuées d'oiseaux ou encore les foules d'humains. Ces comportements collectifs d'animaux ne sont pas seulement des scènes spectaculaires mais sont aussi une source d'intérêt pour explorer les mécanismes sous-jacents dans le but de comprendre les lois et l'évolution des groups biologiques ou même de nous aider à élaborer des essaims auto-organisés de robots.Nous avons étudié différents systèmes de déplacements collectifs, incluant des systèmes avec une seule espèce, comme les bancs de poissons et les foules d'humains et d'autres à plusieurs espèces, comme les systèmes de fuite et poursuite en groupe. Parmi lesquels, nous nous concentrons sur les mécanismes et les rôles du traitement de l'information sur les motifs macroscopiques. De plus, pour palier à la difficulté d'extraire des trajectoires depuis des vidéos expérimentales de qualité médiocre, nous proposons un outil rapide et robuste de suivi du déplacement. Notre contenu de recherche détaillé est le suivant : 1. Nous avons étudié les mécanismes de traitement de l'information dans les déplacements du Nez-rouge dans un dispositive annulaire. Pour la première fois, nous avons définis un comportement particulier aux bancs de poissons : des évènements de demi-tour. En introduisant un délai entre l'interaction entre les poissons, nous avons trouvé qu'à un poisson d'intérêt correspondent seulement un ou deux poissons qui ne sont pas nécessairement les plus proches. De plus, nous avons montré que l'information de tourner pendant un évènement de demi-tour collectif se propage comme des dominos. Enfin, nous avons utilisé le transfert d'entropie pour quantifier les flux d'information dans l'espace et le temps durant les évènements de demi-tour. 2. Nous avons étudié le rôle d'une perturbation dans un système de foule humaine en plaçant des obstacles (les perturbations) dans un flux de fuite de panique. Nous avons trouvé une façon simple et efficace d'augmenter le flux de fuite dans le but de sauver plus de vies dans des situations dangereuses. Nous avons appliqué des algorithmes génétiques pour optimiser l'agencement des piliers dans les simulations puis nous avons testé la qualité de ces résultats contre des expériences avec de vrais humains. Les résultats suggèrent que placer deux piliers le long des deux côtés d'une sortie peut maximiser la vitesse de sortie. 3. Nous avons étudié le rôle des mécanismes de traitement de l'information dans les déplacements collectifs multi-espèces en introduisant différentes strategies pour les proies dans un modèle de poursuite en groupe. Nous proposons trois stratégies d'agrégation : se déplacer vers le centre de masse de toutes les proies (MC), se déplacer vers la proie la plus proche (NN) et minimiser la distance totale entre toutes les proies (MD). Les résultats montrent que l'agrégation augmente grandement la durée de survie du groupe, et ceci même en autorisant les proies à être immortelles. Il y a une transition de phase de t (la durée de survie moyenne) en fonction de M (le nombre de prédateurs). 4. Nous avons développé un nouvel outil de suivi de déplacement pour améliorer les algorithmes de reconnaissance d'image et de suivi actuels afin d'extraire des trajectoires depuis des vidéos de qualité médiocre. Notre outil intègre un filtre de moyenne glissante, la soustraction du bruit de fond, des réseaux de neurones artificiels, du partitionnement en k-moyennes et une fonction d'erreur définie minutieusement. L'outil peut extraire une trajectoire depuis une video de basse qualité qui ne peut être fait que très difficilement par d'autres outils. Il peut suivre plusieurs animaux comme des poissons, des drosophiles, des fourmis, etc. Les performances de notre outil sont meilleures que idTracker et Ctrax. / Collective motion is one of the most striking phenomena in nature. It has been observed in a lot of animal species, such as bacteria, ants, fish, flocks of birds and crowds of human. These collective animal behaviors not only show us spectacular scenes, but also attract us to explore the underlying mechanisms in order to understand the laws and evolution of biological groups and even help us design smarter self-organizing robots. We study different collective motion systems including single species systems such as fish school and human crowd; and multi-species group chase and escape system. Among which, we focus on the mechanisms and roles of information processing on macro patterns. Moreover, regarding to the fact that it's very difficult to extract trajectory data from low quality experiment videos, we propose a fast and robust tracking tool. Details are as follows: 1. We study the mechanisms of information processing in the movements of Hemigrammus rhodostomus in a ring-shaped tank. For the first time, we define a special behavior of fish school: U-turn event. By introducing time delay between fish interaction, we find that a focal fish usually corresponds to only 1 or 2 fish which is not necessarily the nearest one. Moreover, we find the turning information during a group U-turn event propagates like domino. In addition, we use transfer entropy to quantify dynamic information flows in space and time across the U-turn events. 2. We study the role of perturbation information in human crowd system by introducing obstacles as perturbation information into a panic escaping flow. We find a useful and simple way to increase the panic flow in order to save more lives under dangerous situation. We apply genetic algorithms to optimize the layout of pillars in the simulations and then test the results with real human experiments. Results show that putting two pillars along the two sides of the exit can maximize the escape velocity. In the end, a tangential momentum theory is proposed to explain the role of the perturbation information. 3. We study the role of information processing mechanisms in multi-species collective motion by introducing different strateg?ies for the prey in a group chase model. We propose three aggregation strategies: moving to mass center of all preys, moving to the nearest prey and minimising the total distance to all preys. Results show that aggregation increase the group survival time greatly, even allowing immortal prey. There is a phase transition of t (average survival time) against M (number of predator). 4. We developed a new tracking tool to improve the current image recognizing and video tracking algorithms so as to extract trajectories from low quality videos. Our tool integrates mean-value filter, background substraction, artificial neural network, K-means clustering and a well defined cost function. It can track low quality videos which can be hardly tracked by other tools. And it can track different animals such as fish, drosophila, ants and so on. The overall tracking performance is better than idTracker and Ctrax.

Les déplacements collectifs se manifestent souvent de façon spectaculaire et intriguent tant les amateurs de la nature que les chercheurs. Comment émergent ces formes spectaculaires et comment la cohésion des groupes est elle assurée ? Si de nombreux travaux ont été consacrés à l'identification des règles permettant la cohésion dans les groupes en mouvement, plus rares sont ceux consacrés aux transitions entre les états d'arrêt et de déplacement. Cette thèse traite des mécanismes comportementaux impliqués dans les prises de décisions collectives et la dynamique de transition de tels évènements chez le mouton Merinos (Ovis aries). Nous proposons de nouvelles hypothèses sur la modulation des interactions entre individus par des effets spatiaux dans des groupes de grande taille. Nous proposons un modèle spatio-temporel reproduisant nos résultats expérimentaux sur les départs, les déplacements collectifs et les arrêts de groupes de taille croissante et permettant d'explorer les décisions collectives dans des conditions nouvelles. Les résultats expérimentaux et théoriques per- mettent d'améliorer la compréhension des mécanismes individuels à l'origine des décision collectives permettant de maintenir ou non la cohésion des groupes. / Collective movements often involve very spectacular displays that fascinate nature lovers and researchers. How do such amazing patterns appear and how group cohesion can be maintained ? If many studies were carried out to decipher the rules underlying cohesion for groups in movement, there is a lack of works adressing the transitions involved in collective movements : departures and stops. This thesis adresses the behavioural mechanisms involved in the collective decision-making processes oc- curing in such transitions in Merino sheep (Ovis aries) groups. We propose some new kinds of spatial hypotheses that can account for the way interactions between individuals are locally modulated in large groups where individuals cannot have an access to the global information of all individuals. We developed a novel spatiotemporal model of sheep collective motion that reproduces the experimental observations and allows to explore the outcomes of collective decisions in various conditions. The experimental and theoretical results increase the understanding of the individual mechanisms that produce collective decisions allowing to maintain group cohesion.

The main subject of this thesis is to examine thermo-osmotic flows, which occur on interfaces of non-uniform temperature. Such thermo-osmotic flows are purely non-thermal equilibrium phenomena. Along the non-isothermal interface, specific interaction of a liquid and its solutes with a boundary vary in strength across the interface, according to the local temperature. This boundary can be a solid, a membrane or a phase boundary. The flow is thereby continuously pumping fluid across the interface in direction of the local temperature gradient, resulting in an extended flow pattern in the bulk due to mass conservation. In a system containing particles and heat sources in a liquid under spatial confinement, the thermo-osmotic flow may drive particles in a directed manner, or can lead to collective phenomena. To approach this broad topic of (self-)thermophoresis and collective motion of active particles and quantify the role of the thermo-osmotic flow upon the latter effects, different experiments have been performed: The first experiments aim to quantify the thermo-osmotic flow at a non-isothermal liquid/solid interface for two fundamentally different substrate properties. Further, the bulk flow was investigated for two different systems. The form and spatial extension of this bulk flow pattern depends sensitively on the form of the container and the interface, as well as on the thermo-osmotic flow. The first system is a liquid film confined between two planar glass cover slips. The second case is a Janus particle immobilized on one of the glass slips. In the first case, the non-uniform temperature profile is generated by optical heating of a nanometer sized gold colloid, and in the second case, the heat source is the Janus particle. The bulk flow pattern consists, for the second case, of the flow pattern created by the glass cover slips and the one created by the Janus particle. The following experiments are focusing on the dynamics of mobile self-thermophoretic Janus particles. In particular, their dynamics and the contributions of the thermo-osmotic flow to the interaction of multiple active particles are investigated. To investigate those particles under controlled conditions and examine their interactions at low concentrations for an effectively unlimited amount of time, a real-time feedback algorithm was co-developed to gain control of the motion of multiple active particles simultaneously, called ”photon nudging”. With the help of this method, first experiments have been performed to quantify the dynamics of a Janus particle located close to a heat source.

O aprendizado de máquina consiste de conceitos e técnicas que permitem aos computadores melhorar seu desempenho com a experiência, ou, em outras palavras, aprender com dados. Um dos principais tópicos do aprendizado de máquina é o agrupamento de dados que, como o nome sugere, procura agrupar os dados de acordo com sua similaridade. Apesar de sua definição relativamente simples, o agrupamento é uma tarefa computacionalmente complexa, tornando proibitivo o emprego de algoritmos exaustivos, na busca pela solução ótima do problema. A importância do agrupamento de dados, aliada aos seus desafios, faz desse campo um ambiente de intensa pesquisa. Também a classe de fenômenos naturais conhecida como comportamento coletivo tem despertado muito interesse. Isso decorre da observação de um estado organizado e global que surge espontaneamente das interações locais presentes em grandes grupos de indivíduos, caracterizando, pois, o que se chama auto-organização ou emergência, para ser mais preciso. Os desafios intrínsecos e a relevância do tema vêm motivando sua pesquisa em diversos ramos da ciência e da engenharia. Ao mesmo tempo, técnicas baseadas em comportamento coletivo vêm sendo empregadas em tarefas de aprendizado de máquina, mostrando-se promissoras e ganhando bastante atenção. No presente trabalho, objetivou-se o desenvolvimento de técnicas de agrupamento baseadas em comportamento coletivo. Faz-se cada item do conjunto de dados corresponder a um indivíduo, definem-se as leis de interação local, e então os indivíduos são colocados a interagir entre si, de modo que os padrões que surgem reflitam os padrões originalmente presentes no conjunto de dados. Abordagens baseadas em dinâmica de troca de energia foram propostas. Os dados permanecem fixos em seu espaço de atributos, mas carregam certa informação a energia , a qual é progressivamente trocada entre eles. Os grupos são estabelecidos entre dados que tomam estados de energia semelhantes. Este trabalho abordou também o aprendizado semissupervisionado, cuja tarefa é rotular dados em bases parcialmente rotuladas. Nesse caso, foi adotada uma abordagem baseada na movimentação dos próprios dados pelo espaço de atributos. Procurou-se, durante todo este trabalho, não apenas propor novas técnicas de aprendizado, mas principalmente, por meio de muitas simulações e ilustrações, mostrar como elas se comportam em diferentes cenários, num esforço em mostrar onde reside a vantagem de se utilizar a dinâmica coletiva na concepção dessas técnicas / Machine learning consists of concepts and techniques that enable computers to improve their performance with experience, i.e., enable computers to learn from data. Data clustering (or just clustering) is one of its main topics, which aims to group data according to their similarities. Regardless of its simple definition, clustering is a complex computational task. Its relevance and challenges make this field an environment of intense research. The class of natural phenomena known as collective behavior has also attracted much interest. This is due to the observation that global patterns may spontaneously arise from local interactions among large groups of individuals, what is know as self-organization (or emergence). The challenges and relevance of the subject are encouraging its research in many branches of science and engineering. At the same time, techniques based on collective behavior are being employed in machine learning tasks, showing to be promising. The objective of the present work was to develop clustering techniques based on collective behavior. Each dataset item corresponds to an individual. Once the local interactions are defined, the individuals begin to interact with each other. It is expected that the patterns arising from these interactions match the patterns originally present in the dataset. Approaches based on dynamics of energy exchange have been proposed. The data are kept fixed in their feature space, but they carry some sort of information (the energy), which is progressively exchanged among them. The groups are established among data that take similar energy states. This work has also addressed the semi-supervised learning task, which aims to label data in partially labeled datasets. In this case, it has been proposed an approach based on the motion of the data themselves around the feature space. More than just providing new machine learning techniques, this research has tried to show how the techniques behave in different scenarios, in an effort to show where lies the advantage of using collective dynamics in the design of such techniques

Le déplacement cohérent dirigé au sein de troupeaux, d’essaims, de nuées, prend place à toutes les échelles du vivant. En cherchant à rationaliser l’émergence de tels mouvements collectifs, les physiciens ont décrit ces assemblées comme des matériaux actifs. Ces matériaux sont formés de constituants auto-propulsés qui se déplacent spontanément dans une direction commune. Cette thèse expérimentale s’appuie sur la réalisation de troupeaux synthétiques pour explorer les propriétés de la matière active polaire dans des situations défavorables à son auto-organisation : leur dynamique en milieux désordonnés et leur réponse à des perturbations externes. Des rouleurs colloïdaux aux interactions d’alignement sont confinés au sein de dispositifs microfluidiques. Au-delà d’une densité seuil, ils forment un troupeau caractérisé par l’émergence d’un ordre en orientation de longue portée. Ces troupeaux colloïdaux font office de prototypes de la matière active polaire. Nous avons étudié la réponse d’un liquide actif polaire assemblé à partir de rouleurs colloïdaux. Nous avons montré que face à une perturbation longitudinale leur réponse est hystérétique. Nous avons expliqué théoriquement ce comportement non-linéaire et l’avons exploité pour réaliser des oscillateurs microfluidiques autonomes. Nous avons également étudié la dynamique de troupeaux colloïdaux qui se propagent dans des environnements hétérogènes. La présence d’obstacles distribués aléatoirement focalise les troupeaux le long de chemins privilégiés qui forment un réseau épars et tortueux. Augmenter le désordre conduit à la destruction du troupeau. Nous avons démontré que la suppression du mouvement collectif consiste en une transition discontinue, générique à tous les matériaux actifs polaires. / Directed collected motion within herds, swarms and flocks, is a phenomenon that takes place at all scales in living systems. Physicists have rationalized the emergence of such collective behavior. They have described these systems as active materials. These materials are assembled from self-propelled units that spontaneously move in the same direction. By experimentally studying synthetic flocks, this work uncovers some properties of polar active materials in situations that disfavor their self-organization: their dynamics in disordered environments and their response to external perturbations. Colloidal rollers with alignment interactions are confined within microfluidic devices. At high density, they spontaneously form a flock which is characterized by the emergence of orientational long-ranged order. These colloidal flocks are prototypical realizations of polar active matter. We have studied the response of a polar active liquid assembled from colloidal rollers. We have shown that they display a hysteretic response to longitudinal perturbations. We have theoretically accounted for this non-linear behavior. We have used this behavior to realize autonomous microfluidic oscillators. We have also studied the dynamics of colloidal flocks that propagate through heterogeneous environments. Randomly positioned obstacles focalize flocks along favored channels that form a sparse and tortuous network. Increasing disorder leads to the destruction of flocks. We have demonstrated that the suppression of collective motion is a discontinuous transition generic to all polar active materials.

O aprendizado de máquina consiste de conceitos e técnicas que permitem aos computadores melhorar seu desempenho com a experiência, ou, em outras palavras, aprender com dados. Um dos principais tópicos do aprendizado de máquina é o agrupamento de dados que, como o nome sugere, procura agrupar os dados de acordo com sua similaridade. Apesar de sua definição relativamente simples, o agrupamento é uma tarefa computacionalmente complexa, tornando proibitivo o emprego de algoritmos exaustivos, na busca pela solução ótima do problema. A importância do agrupamento de dados, aliada aos seus desafios, faz desse campo um ambiente de intensa pesquisa. Também a classe de fenômenos naturais conhecida como comportamento coletivo tem despertado muito interesse. Isso decorre da observação de um estado organizado e global que surge espontaneamente das interações locais presentes em grandes grupos de indivíduos, caracterizando, pois, o que se chama auto-organização ou emergência, para ser mais preciso. Os desafios intrínsecos e a relevância do tema vêm motivando sua pesquisa em diversos ramos da ciência e da engenharia. Ao mesmo tempo, técnicas baseadas em comportamento coletivo vêm sendo empregadas em tarefas de aprendizado de máquina, mostrando-se promissoras e ganhando bastante atenção. No presente trabalho, objetivou-se o desenvolvimento de técnicas de agrupamento baseadas em comportamento coletivo. Faz-se cada item do conjunto de dados corresponder a um indivíduo, definem-se as leis de interação local, e então os indivíduos são colocados a interagir entre si, de modo que os padrões que surgem reflitam os padrões originalmente presentes no conjunto de dados. Abordagens baseadas em dinâmica de troca de energia foram propostas. Os dados permanecem fixos em seu espaço de atributos, mas carregam certa informação a energia , a qual é progressivamente trocada entre eles. Os grupos são estabelecidos entre dados que tomam estados de energia semelhantes. Este trabalho abordou também o aprendizado semissupervisionado, cuja tarefa é rotular dados em bases parcialmente rotuladas. Nesse caso, foi adotada uma abordagem baseada na movimentação dos próprios dados pelo espaço de atributos. Procurou-se, durante todo este trabalho, não apenas propor novas técnicas de aprendizado, mas principalmente, por meio de muitas simulações e ilustrações, mostrar como elas se comportam em diferentes cenários, num esforço em mostrar onde reside a vantagem de se utilizar a dinâmica coletiva na concepção dessas técnicas / Machine learning consists of concepts and techniques that enable computers to improve their performance with experience, i.e., enable computers to learn from data. Data clustering (or just clustering) is one of its main topics, which aims to group data according to their similarities. Regardless of its simple definition, clustering is a complex computational task. Its relevance and challenges make this field an environment of intense research. The class of natural phenomena known as collective behavior has also attracted much interest. This is due to the observation that global patterns may spontaneously arise from local interactions among large groups of individuals, what is know as self-organization (or emergence). The challenges and relevance of the subject are encouraging its research in many branches of science and engineering. At the same time, techniques based on collective behavior are being employed in machine learning tasks, showing to be promising. The objective of the present work was to develop clustering techniques based on collective behavior. Each dataset item corresponds to an individual. Once the local interactions are defined, the individuals begin to interact with each other. It is expected that the patterns arising from these interactions match the patterns originally present in the dataset. Approaches based on dynamics of energy exchange have been proposed. The data are kept fixed in their feature space, but they carry some sort of information (the energy), which is progressively exchanged among them. The groups are established among data that take similar energy states. This work has also addressed the semi-supervised learning task, which aims to label data in partially labeled datasets. In this case, it has been proposed an approach based on the motion of the data themselves around the feature space. More than just providing new machine learning techniques, this research has tried to show how the techniques behave in different scenarios, in an effort to show where lies the advantage of using collective dynamics in the design of such techniques