Enhancing Divergent Search through Extinction Events

摘要: A challenge in evolutionary computation is to create representations as evolvable those natural evolution. This paper hypothesizes that extinction events, i.e. mass extinctions, can significantly increase evolvability, but only when combined with a divergent search algorithm, driven towards diversity (instead of optimality). Extinctions amplify diversity-generation by creating unpredictable bottlenecks. Persisting through multiple such bottlenecks more likely for lineages diversify across many niches, resulting indirect selection pressure the capacity evolve. hypothesis tested experiments two robotics domains. The results show combining events increases while them convergent offers no similar benefit. conclusion may provide simple and effective mechanism enhance performance algorithms.