Discovering Evolutionary Stepping Stones through Behavior Domination

摘要: Behavior domination is proposed as a tool for understanding and harnessing the power of evolutionary systems to discover exploit useful stepping stones. Novelty search has shown promise in overcoming deception by collecting diverse stones, several algorithms have been that combine novelty with more traditional fitness measure refocus help scale complex domains. However, combinations do not necessarily preserve stone discovery affords. In existing methods, competition between solutions can lead an unintended loss diversity. defines class avoid this problem, while inheriting theoretical guarantees from multiobjective optimization. Several are be class, new algorithm introduced based on fast non-dominated sorting. Experimental results show outperforms approaches domains contain its advantage sustained scale. The conclusion behavior illuminate dynamics behavior-driven search, thus design scalable robust algorithms.