The effect of non-symmetric fitness
Denis Antipov, Semen Naumov · 2021
Theory of evolutionary computation has brought plenty of useful recommendations to the practitioners on how to deal with local optima. Many of these results were obtained through the runtime analysis of evolutionary algorithms (EAs for brevity) on Jump benchmark function, which has a local optimum which is very hard to leave for most EAs. The performed analyses resulted into multiple new algorithms which showed a good performance on Jump function, including the (1 + (λ, λ)) GA (with dynamic parameters choices or with non-standard static parameters), the (μ + 1) GA with various diversity mechanisms, and the hybrid GA.