SGER: Scale and Complexity in Computational Models of Social Interaction

该研究将调查扩展大规模多智能体计算模型所带来的计算挑战，以实现更大的规模和更复杂和现实的行为。 这些模型中的个体代理由复杂的数据结构表示，这些数据结构表示它们的内部状态和行为库。即使在规模不大的智能体群体中，交互历史（决定模型状态）的数量也是巨大的，因此构建逼真的模型构成了重大的计算障碍。本研究的主要目的是探索数量复杂性边界：在该点有大量的简单代理和模型与更多的更复杂的代理之间的权衡。该项目旨在开发创新的计算技术，将模型的规模和复杂性扩展两到三个数量级。这些进展对于理解当多智能体或社会系统的规模和/或复杂性显著增加时是否会发生相变至关重要。这也是至关重要的未来现实世界的应用程序的多智能体系统建模的问题，如了解发展和迁移模式（如中国），这带来了规模的问题，并了解同龄人群体对犯罪率的影响（这带来了问题的建模的复杂性，个别代理行为）。

The research will investigate the computing challenges posed by extending large-scale multi-agent computational models to achieve both greater scale and more complex and realistic behaviors by individual agents. Individual agents in these models are represented by complex data structures that represent their internal states and behavioral repertoires. In agent populations of even modest size, the number of interaction histories (which determines the state of a model) is huge, so building realistic models poses significant computational hurdles. The primary aim of this research is to explore the quantity-complexity frontier: point at which there are tradeoffs between models with large numbers of simple agents and models with greater numbers of more complex agents. This project aims to develop innovative computational techniques to extend both the scale and the complexity of models by two to three orders of magnitude. Such advances are crucial to understanding whether phase changes occur when scale and/or complexity is significantly increased in multi-agent or social systems. It is also critical to future real-world applications of multi-agent systems modeling to problems such as understanding development and migration patterns (e.g. China), which pose problems of scale, and understanding peer-group effects on crime rates (which pose problems of modeling the complexity of individual agent behaviors).