Distance modulation competitive co-evolution method to find initial configuration independent cellular automata rules

One of the main problems in machine learning methods based on examples is the over-adaptation. This problem supposes the exact adaptation to the training examples losing the capability of generalization. A solution of these problems arises in using large sets of examples. In most of the problems, to achieve generalized solutions, almost infinity examples sets are needed. This make the method useless in practice. In this paper, one way to overcome this problem is proposed, based on biological competitive evolution ideas. The evolution is produced as a result of a competition between sets of solutions and sets of examples, trying to beat each other. This mechanism allows the generation of generalized solutions using short example sets.