Towards Evolutionary Emergence
Jörg Bremer, Sebastian Lehnhoff · Annals of Computer Science and Information Systems · 2021
With the upcoming era of large-scale, complex cyber-physical systems, also the demand for decentralized and self-organizing algorithms for coordination rises.Often such algorithms rely on emergent behavior; local observations and decisions aggregate to some global behavior without any apparent, explicitly programmed rule.Systematically designing these algorithms targeted for a new orchestration or optimization task is, at best, tedious and error prone.Suitable and widely applicable design patterns are scarce so far.We opt for a machine learning based approach that learns the necessary mechanisms for targeted emergent behavior automatically.To achieve this, we use Cartesian genetic programming.As an example that demonstrates the general applicability of this idea, we trained a swarm-based optimization heuristics and present first results showing that the learned swarm behavior is significantly better than just random search.We also discuss the encountered pitfalls and remaining challenges on the research agenda.