Efficient Estimation of Agent Networks

Collective adaptive systems (CAS) are characterized by the presence of many agents and an environment which interact with each other. As a consequence, they give rise to global dynamics which cannot be analyzed by considering agents in isolation. While the modeling of CAS via agent (reaction) networks gained momentum, obtaining reliable forecasts is computationally difficult because parameters are often subject to uncertainty. It has been therefore recently proposed to obtain reliable estimates on global dynamics of agent networks from local agent behavior. To this end, dependencies among agents were replaced by exogenous parameters, allowing one thus to estimate the global dynamics via agent decoupling. The present work introduces the notion of estimation equivalence, a model reduction technique for systems of nonlinear differential equations that allows one to replace the aforementioned decoupled model by a smaller one which is easier to analyze. We demonstrate the framework on a multi-class SIRS model from epidemiology and obtain a speed-up factor that is proportional to the number of population classes.