Les bancs de poissons sont des entités pouvant regrouper plusieurs milliers d'individus qui se déplacent de façon synchronisée, dans un environnement sujet à de multiples perturbations, qu'elles soient endogènes (e.g. le départ soudain d'un congénère) ou exogènes (e.g. l'attaque d'un prédateur). La coordination de ces bancs de poissons, décentralisée, n'est pas encore totalement comprise. Si les mécanismes sous-jacents aux interactions sociales proposés dans des travaux précédents reproduisent qualitativement les structures collectives observées dans la nature, la quantification de ces interactions et l'accord quantitatif entre ces mesures individuelles et les motifs collectifs sont encore rares dans les recherches récentes et forment l'objet principal de cette thèse. L'approche de ce travail repose sur une étroite combinaison entre les méthodes expérimentales et de modélisation dans l'objectif de découvrir les liens entre les comportements individuels et les structures observées à l'échelle collective. Nous avons caractérisé et quantifié les interactions et mécanismes à l'origine, d'abord, de la coordination des individus dans les bancs de poissons et, ensuite, de la propagation d'information, quand le groupe subit une perturbation endogène ou exogène. Ces travaux, tous réalisés en étudiant la même espèce de poisson d'eau douce, le nez-rouge (Hemigrammus rhodostomus), ont mobilisé une diversité de méthodes expérimentales, d'analyses statistique et de modélisation, à l'interface de l'éthologie, de la physique statistique et des sciences computationnelles. / Fish schools are systems in which thousands of individuals can move in a synchronised manner in a changing environment, with endogenous perturbations (e.g. when a congener leaves the group) or exogenous (e.g. the attack of a predator). The coordination of fish schools, decentralised, is not completely understood yet. If the mechanisms underlying social interactions discussed in previous studies qualitatively match the social patterns observed in nature, the quantification of these interactions and the quantitative match between individual measurements and collective patterns are still sparse in recent works and are the main focus of this thesis. This work combines closely experimental and modelling methods in order to investigate the links between the individual behaviours and the patterns observed at the collective scale. We have characterised and quantified the interactions and mechanisms at the origin of, first, the coordination of individuals in fish schools and, second, the propagation of information, when the group is under endogenous or exogenous perturbations. This thesis focuses on one freshwater fish species, the rummy-nose tetra (Hemigrammus rhodostomus), and is the result of a diversity of experimental methods, statistical analyses and modelling, at the interface of ethology, statistical physics and computational sciences.

No / Large-scale behavioral simulations are widely used to study real-world multi-agent systems. Such programs normally run in discrete time-steps or ticks, with simulated space decomposed into domains that are distributed over a set of workers to achieve parallelism. A distinguishing feature of behavioral simulations is their frequent and high-volume group migration, the phenomenon in which simulated objects traverse domains in groups at massive scale in each tick. This results in continual and significant load imbalance among domains. To tackle this problem, traditional load balancing approaches either require excessive load re-profiling and redistribution, which lead to high computation/communication costs, or perform poorly because their statically partitioned data domains cannot reflect load changes brought by group migration. In this paper, we propose an effective and low-cost load balancing scheme, named Inc-part, based on a key observation that an object is unlikely to move a long distance (across many domains) within a single tick. This localized mobility property allows one to efficiently estimate the load of a dynamic domain incrementally, based on merely the load changes occurring in its neighborhood. The domains experiencing significant load changes are then partitioned or merged, and redistributed to redress load imbalance among the workers. Experiments on a 64-node (1,024-core) platform show that Inc-part can attain excellent load balance with dramatically lowered costs compared to state-of-the-art solutions.