Reference Vector-Assisted Adaptive Model Management for Surrogate-Assisted Many-Objective Optimization

A Secure Federated Data-Driven Evolutionary Multi-objective Optimization Algorithm

“…This procedure continues until a termination condition, usually the allowed maximum number of fitness evaluations, is met. Representative GPassisted SAEAs for expensive multi-objective optimization include the GP-assisted multi-objective evolutionary algorithm based on decomposition [33], [34], surrogate-assisted reference vector guided evolutionary algorithm [35], kriging assisted two archives evolutionary algorithm [36], and a reference vector guided adaptive model management [37]. However, they are all designed based on the assumption that both the raw data and the newly queried data are centrally stored on one machine centralized, which cannot directly extended to a federated environment.…”

“…We examine the performance of FDD-EA-DH on the DTLZ [38] and WFG [39] test suite, with the number of objectives M being set to 3, 5, or 10, and the numbers of decision variables D being set to 20. The maximum number of fitness evaluations F E max for 20-dimensional test instances is set to 219+120 (219 is the number of raw data in each client and 120 is the number of allowed new query points throughout the optimization process), respectively, which is the same as the setting of the number of initially sampled solutions in [9], [37], i.e., 11 • D − 1. The number of queried solutions µ is set to 5.…”