Modelling Multi-Level Learning in Multi-Agent-Systems with Stochastic Nets-within-Nets

In multi-agent systems (MAS), we havemicroelements, i.e., the agents, and macro elements, i.e., the organizational network with roles and behavior rules. Learning occurs at both levels, and the learning processes are intertwined. We like to analyze interaction and learning in MAS. We develop a formal framework rooted in game theory to study the micro-macro learning processes. While our ground formalism is suitable for analytical purposes, it is not well adjusted to engineer MAS. Therefore, in a second step, we translate micro and macro models into specialized probabilistic Petri nets to bridge the gap between theory and real applications. Our model is inspired by the common MAPE pattern for feedback loops (monitor, analyze, plan, and execute). Finally, we integrate the multi-level model into one single model. This model is based on the nets-within-nets approach of probabilistic Hornets. Since our ground formalism is probabilistic in nature, it is essential that our Hornet provides probabilistic features. This allows for a seamless translation of the ground formalism to enable formal analysis