| Abstract: |
Multiobjective problems occur frequently in practice. During the last decades, multiobjective evolutionary algorithms (MOEAs) have been successfully applied to various real-world applications
where 2 or more objective functions have to be considered simultaneously. Optimizing several objectives at the same time usually involves a decision maker who specifies which of the multiple
trade-off solutions are preferred. The advantage of using MOEAs is that the decision making process can be done during or after the search rather than before the optimization as it is necessary,
e.g., in classical scalarization approaches; articulating preferences after the inherent trade-offs among the objectives are known is often much easier for a human decision maker. Several
questions directly arise from practical applications such as (1) how to effectively incorporate the preferences of a decision maker into the search process, (2) how the decision maker should be
involved in interactive optimization if some preferences cannot be formalized mathematically, or more generally, (3) how problems with many, i.e., more than 5 objectives can be tackled in terms
of visualization, decision making, and search. One of the main challenges in the field of evolutionary multiobjective optimization is how to develop efficient and effective optimization methods
addressing the aforementioned questions. In this talk, I will address the questions (1) and (3) both from a practical and from a theoretical point-of-view. The talk is divided into three parts.
First, I will explain how it is possible to include preferences into the search by means of weighted hypervolume indicators. Second, I will present how the automated reduction of objectives can
assist both in decision making and search by means of a radar waveform optimization problem. Last, I will indicate some interesting open questions in the field of evolutionary multiobjective
optimization. |
