Utilizing domain knowledge in neuroevolution

摘要: We propose a method called Rule-based ESP (RESP) for utilizing prior knowledge evolving Artificial Neural Networks (ANNs). First, KBANN-likete chniques are used to transform set of rules into an ANN, then the ANN is trained using Enforced Subpopulations (ESP) neuroevolution method. Empirical results in Prey Capture domain show that RESP can reach higher level performance than ESP. The also suggest incremental learning not necessary with RESP, and it often easier design evolution scheme. In addition, experiment some deleted suggests robust even incomplete base. therefore provides methodology scaling up harder tasks by existing about domain.