Predictive modeling and control of DMAS

Predicting the behavior of distributed multi-agent systems (DMAS) is a blown, extremely challenging problem. In general, we are not able to make reliable quantitative predictions of the behavior that a given DMAS exhibits, even in a blown environment, due to the complex emergent effects in those systems, which often reflect chaotic interactions. Such predictability is nonetheless crucial for reliable, controlled development and deployment of such systems. We need to be able to control the behaviors of such systems, and want to optimize configurations to achieve acceptable and reliable returns of quality-of-service for an investment of resources. We describe here an approach to developing reliable predictive models for a particular class of DMAS. We have succeeded in developing such models for this class of applications and in achieving controlled behaviors and optimized configurations based on these predictive models. We discuss our approach, and results and plans for applying this approach to broader classes of applications.