IEEE Members: Free
Non-members: FreeDuration: 00:45:48
Many-objective optimization problems (ManyOPs) pose challenges to existing multi-objective evolutionary algorithms (MOEAs) in terms of convergence, diversity, and computational complexity. This talk presents a personal view towards various strategies and methods for coping with many objectives, from simple ideas of more efficient non-dominated sorting and nonlinear dimensionality reduction, to other simple ideas of a two-archive algorithm (i.e., Two_Arch2), which use two separate archives to focus on convergence and diversity respectively. Different selection principles (indicator-based and Pareto-based) are used in the two archives. A new Lp-norm based diversity maintenance scheme is introduced. Our experimental results show that Two_Arch2 can cope with ManyOPs (up to 20 objectives) with satisfactory convergence, diversity, and complexity.