Abstrakt
Smart self-adaptive systems are nowadays commonly employed almost in any application domain. Within them, groups of robots, autonomous vehicles, drones, and similar automatons dynamically cooperate to achieve a common goal.
An approach to model such dynamic cooperation is via autonomic component ensembles, which are dynamically formed groups of components. Forming ensembles is described via a set of constraints (e.g., form an ensemble of three drones closest to a target that have sufficient battery level to reach the target and stay there).
Evaluating these constraints by traditional means (such as a SAT solver) is computationally demanding and does not scale for large systems. This paper proposes an approach for solving ensemble formations based on machine learning which may be relatively faster.
The method trains the model on a small instance of the system governed by a computationally demanding algorithm and then adapts it for large instances thanks to the generalization properties of the machine learning model.