On the impact of utility functions in interactive evolutionary multi-objective optimization

Interactive evolutionary algorithms for multi-objective optimization have gained an increasing interest in recent years. As multiobjective optimization usually deals with the optimization of conflicting objectives, a decision maker is involved in the optimization process when encountering incomparable solutions. We study the impact of a decision maker from a theoretical perspective and analyze the runtime of evolutionary algorithms until they have produced for the first time a Pareto optimal solution with the highest preference of the decision maker. Considering the linear decision maker, we show that many multi-objective optimization problems are not harder than their single-objective counterpart. Interestingly, this does not hold for a decision maker using the Chebeyshev utility function. Furthermore, we point out situations where evolutionary algorithms involving a linear decision maker have difficulties in producing an optimal solution even if the underlying single-objective problems are easy to be solved by simple evolutionary algorithms.Frank Neumann and Anh Quang Nguye