Efficient optimization of many objectives by approximation-guided evolution

摘要: Abstract Multi-objective optimization problems arise frequently in applications, but can often only be solved approximately by heuristic approaches. Evolutionary algorithms have been widely used to tackle multi-objective problems. These use different measures ensure diversity the objective space are not guided a formal notion of approximation. We present framework for evolutionary that allows work with This approximation-guided algorithm (AGE) has worst-case runtime linear number objectives and works an archive is approximation non-dominated vectors seen during run algorithm. Our experimental results show AGE finds competitive or better solutions regarding achieved approximation, also total hypervolume. For all considered test problems, even many (i.e., more than ten) dimensions, discovers good Pareto front. case established such as NSGA-II, SPEA2, SMS-EMOA. In this paper we compare two additional very fast hypervolume-approximations guide their search. significantly speeds up hypervolume-based algorithms, which now comparison underlying selection schemes.