Le réalisme d'une application traitant de l'animation comportementale de foules est fondé d'une part sur le rendu graphique des scènes produites par l'application, mais aussi sur le réalisme du comportement lui-même. C'est ce dernier point qui est notre objet d'étude. Le réalisme du comportement d'une foule est avant tout global (réalisme macroscopique) : elle doit avoir des propriétés statistiques (densité, dispersion, vitesse moyenne,…) proches de celles d'une foule réelle. Il est aussi local (réalisme microscopique), c'est-à-dire que les agents ou groupes d'agents doivent idéalement avoir des comportements proches de ceux des humains ou groupes humains pris pour référence, tout au moins dans le domaine d'application restreint considéré. L'objet de cette thèse est de proposer un modèle générique pour effectuer des simulations comportementales de foules, pour pouvoir satisfaire simultanément les deux types de réalisme macroscopique et microscopique et rendre compte de la plupart des types de foules (foules homogènes, groupes, collection d'agents). Pour ce faire, nous avons dégagé quelques principes simples et peu coûteux en ressources. Nous introduirons la notion de contexte (global, localisé, et propagateur). Les contextes globaux et localisés appliqués à un groupe d'agents tendent à leur donner un même comportement et constituent ainsi un moyen de contrôle global de ces acteurs (réalisme macroscopique). Le contexte propagateur donne un moyen de propagation d'informations entre les agents de la foule (communication). Les agents eux-mêmes sont dotés de tendances qui, en les différenciant dans un même contexte, leur donne une variété de comportements qui concourt au réalisme microscopique. / The realism of crowd behavioral animation is based on one hand on a rendering of graphic scenes generated by the application, and on the other hand on the realism of the behavior. This is last point which is our object of study. The realism of crowds' behavior is essentially global (macroscopic realism). It must respect the required statistical characteristics of the crowd (density, dispersal, speed…) similar to those of a real crowd. The realism is also local (microscopic realism), i.e. agents should ideally behave like humans or human groups taken as reference, at least in the focus of the considered application. The aim of our research is to propose a generic model to perform crowd behavior simulation in order to simultaneously satisfy the criterion of macroscopic and microscopic realism and that may be used to build various crowd behavioral simulations (homogeneous crowds, groups, collections of agents). To do this, we identified some simple principles and inexpensive resources; we introduce the notion of context (global, localized, and propagator). Global contexts applied to a group of agents, it tends to give them the same behavior and thus constitutes a means for global control of those agents (macroscopic realism). Propagator contexts allow the propagation of information among agents (communication). Agents are themselves equipped with tendencies that differentiate them within the context, and give them a variety of behaviors that contributes to microscopic realism.

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.

This research addresses problems in designing analog and mixed-signal (AMS) systems by bridging the gap between system-level and circuit-level simulation by making simulations fast like system-level and accurate like circuit-level. The tools proposed include metamodel integrated Verilog-AMS based design exploration flows. The research involves design centering, metamodel generation flows for creating efficient behavioral models, and Verilog-AMS integration techniques for model realization. The core of the proposed solution is transistor-level and layout-level metamodeling and their incorporation in Verilog-AMS. Metamodeling is used to construct efficient and layout-accurate surrogate models for AMS system building blocks. Verilog-AMS, an AMS hardware description language, is employed to build surrogate model implementations that can be simulated with industrial standard simulators. The case-study circuits and systems include an operational amplifier (OP-AMP), a voltage-controlled oscillator (VCO), a charge-pump phase-locked loop (PLL), and a continuous-time delta-sigma modulator (DSM). The minimum and maximum error rates of the proposed OP-AMP model are 0.11 % and 2.86 %, respectively. The error rates for the PLL lock time and power estimation are 0.7 % and 3.0 %, respectively. The OP-AMP optimization using the proposed approach is ~17000× faster than the transistor-level model based approach. The optimization achieves a ~4× power reduction for the OP-AMP design. The PLL parasitic-aware optimization achieves a 10× speedup and a 147 µW power reduction. Thus the experimental results validate the effectiveness of the proposed solution.