Age-fitness pareto optimization

摘要: We propose a multi-objective method for avoiding premature convergence in evolutionary algorithms, and demonstrate three-fold performance improvement over comparable methods. Previous research has shown that partitioning an evolving population into age groups can greatly improve the ability to identify global optima avoid converging local optima. Here, we treating as explicit optimization criterion increase even further, with fewer algorithm implementation parameters. The proposed evolves on two-dimensional Pareto front comprising (a) how long genotype been (age); (b) its (fitness). compare this approach previous approaches Symbolic Regression problem, sweeping problem difficulty range of solution complexities number variables. Our results indicate identifies exact target more often age-layered standard also performs better higher complexity problems dimensional datasets -- finding less computational effort.