Stochastic method for the solution of unconstrained vector optimization problems
Journal of Optimization Theory and Applications
Feedback-Control Operators for Evolutionary Multiobjective Optimization
EMO '09 Proceedings of the 5th International Conference on Evolutionary Multi-Criterion Optimization
Multicriteria optimization with a multiobjective golden section line search
Mathematical Programming: Series A and B
A fast and elitist multiobjective genetic algorithm: NSGA-II
IEEE Transactions on Evolutionary Computation
On the Evolutionary Optimization of Many Conflicting Objectives
IEEE Transactions on Evolutionary Computation
Diversity Management in Evolutionary Many-Objective Optimization
IEEE Transactions on Evolutionary Computation
IEEE Transactions on Evolutionary Computation
Hi-index | 0.01 |
Traditional Evolutionary Multiobjective Optimization techniques, based on derivative-free dominance-based search, allowed the construction of efficient algorithms that work on rather arbitrary functions, leading to Pareto-set sample estimates obtained in a single algorithm run, covering large portions of the Pareto-set. However, these solutions hardly reach the exact Pareto-set, which means that Pareto-optimality conditions do not hold on them. Also, in problems with high-dimensional objective spaces, the dominance-based search techniques lose their efficiency, up to situations in which no useful solution is found. In this paper, it is shown that both effects have a common geometric structure. A gradient-based descent technique, which relies on the solution of a certain stochastic differential equation, is combined with a multiobjective line-search descent technique, leading to an algorithm that indicates a systematic solution for such problems. This algorithm is intended to serve as a proof of concept, allowing the comparison of the properties of the gradient-search principle with the dominance-search principle. It is shown that the gradient-based principle can be used to find solutions which are truly Pareto-critical, satisfying first-order conditions for Pareto-optimality, even for many-objective problems.