A Survey on Learnable Evolutionary Algorithms for Scalable Multiobjective Optimization

Recent decades have witnessed great advancements in multiobjective evolutionary algorithms (MOEAs) for optimization problems (MOPs). However, these progressively improved MOEAs not necessarily been equipped with scalable and learnable problem-solving strategies new grand challenges brought by the scaling-up MOPs continuously increasing complexity from diverse aspects, mainly including expensive cost of function evaluations, many objectives, large-scale search space, time-varying environments, multi-task. Under different scenarios, divergent thinking is required designing powerful solving them effectively. In this context, research studies on machine learning techniques received extensive attention field computation. This paper begins a general taxonomy MOEAs, followed an analysis that pose to traditional MOEAs. Then, we synthetically overview recent advances various MOPs, focusing primarily four attractive directions (i.e., discriminators environmental selection, generators reproduction, evaluators transfer modules sharing or reusing experience). The insight offered readers as reference track efforts field.