GECCO 2012 tutorial on evolutionary multiobjective optimization
Abstract
Many optimization problems are multiobjective in nature in the sense that multiple, conflicting criteria need to be optimized simultaneously. Due to the conflict between objectives, usually, no single optimal solution exists. Instead, the optimum corresponds to a set of so-called Pareto-optimal solutions for which no other solution has better function values in all objectives. Evolutionary Multiobjective Optimization (EMO) algorithms are widely used in practice for solving multiobjective optimization problems due to several reasons. As randomized blackbox algorithms, EMO approaches allow to tackle problems with nonlinear, nondifferentiable, or noisy objective functions. As set-based algorithms, they allow to compute or approximate the full set of Pareto-optimal solutions in one algorithm run---opposed to classical solution-based techniques from the multicriteria decision making (MCDM) field. Using EMO approaches in practice has two other advantages: they allow to learn about a problem formulation, for example, by automatically revealing common design principles among (Pareto-optimal) solutions (innovization) and it has been shown that certain single-objective problems become easier to solve with randomized search heuristics if the problem is reformulated as a multiobjective one (multiobjectivization).
Authors: Dimo Brockhoff, Kalyanmoy Deb
Published in: Genetic and Evolutionary Computation Conference Companion (GECCO Companion) (2012)