Macroscopic Observation of Large-scale Multi-agent Systems

This paper is interested in the analysis and the debugging of decentralized asynchronous large-scale multi-agent systems. In order to work at scale, such a task requires computer scientists to abstract away from the agent microscopic dynamics by providing a global representation of the system dynamics. In this paper, we propose to address this challenging need by designing descriptional devices that reduce the microscopic complexity of the system descriptions. This approach is applied to the aggregation of the microscopic causal features that constitute the execution of a given multi-agent system: spatial and temporal abstractions are built from the observation of microscopic interactions in order to provide a global understanding of the system trajectory. The scaling issue is then handled by observational devices that directly compute such macroscopic descriptions without collecting any microscopic data. This macroscopic observation process is evaluated on a classic ant colony model.