User preferences for approximation-guided multi-objective evolution

Abstract

Incorporating user preferences into evolutionary multi-objective evolutionary algorithms has been an important topic in recent research in the area of evolutionary multi-objective optimization. We present a very simple and yet very effective modification to the Approximation- Guided Evolution (AGE) algorithm to incorporate user preferences. Over a wide range of test functions, we observed that the resulting algorithm called iAGE is just as good at finding evenly distributed solutions as similarly modified NSGA-II and SPEA2 variants. However, in particular for ”difficult” two-objective problems and for all three-objective problems we see more evenly distributed solutions in the user preferred region when using iAGE.