Generalized offline orthant search: one code for many problems in multiobjective optimization.

We introduce generalized offline orthant search, an algorithmic framework that can be used to solve many problems coming from evolutionary multiobjective optimization using a common well-optimized algorithmic core and relatively cheap reduction procedures. The complexity of the core procedure is O(n · (log n)k-1) for n points of dimension k, and it has a good performance in practice.

We show that the presented approach can perform various flavors of non-dominated sorting, dominance counting, evaluate the ε-indicator and perform initial fitness assignment for IBEA. It is either competitive with the state-of-the-art algorithms or completely outperforms them for higher problem sizes. We hope that this approach will simplify future development of efficient algorithms for evolutionary multiobjective optimization.