Using Cellular Automata with Evolutionary Learned Rules to Solve the Online Partitioning Problem
Andreas Goebels, Alexander Weimer, Steffen Priesterjahn · 2005
In recent computer science research highly robust and scalable sets that are composed of autonomous individuals have become more and more important. The online partitioning problem (OPP) deals with the distribution of huge sets of agents onto different targets in consideration of several objectives. The agents can only interact locally and there is no central instance or global knowledge. In this paper we work on this problem field by modifying ideas from the area of cellular automata (CA). We expand the well known majority/density classification task for one-dimensional CAs to two-dimensional CAs. The transition rules for the CA are learned by using a genetic algorithm (GA). Each individual in the GA is a set of transition rules with additional distance information. This approach shows very good behaviour compared to other strategies for the OPP and is very fast once an appropriate set of rules is learned by the GA