Many practical optimization problems require the consideration of multiple, conflicting objectives. In such cases, usually no single optimal solution exists. Instead, there is a set of so-called efficient or Pareto-optimal solutions with different trade-offs. Evolutionary algorithms, i.e., heuristics inspired by natural evolution, have gained increasing popularity for such multi-objective problems. Since they work on a population of solutions, they can be used to simultaneously search for a well-distributed set of Pareto-optimal solutions in a single run. This provides the decision maker with a set of alternatives to choose from.
This talk will give an introduction to evolutionary multiobjective optimisation, and then discuss why and how the decision maker’s preferences should be incorporated, either a priori or during the optimisation. A number of different ways to incorporate user preferences are presented, including some recent approaches that learn the user preferences from pairwise comparisons during the run.
Homepage of Jürgen Branke