Bio-Inspired Algorithms for Many-Objective Discrete Optimization

Bio-inspired approaches have been recently investigated as an alternative to solve intractable multi-objective problems with many objectives. In a recent paper, we proposed a many-objective algorithm named MACO/NDS which is based on ant colony optimization (ACO). Although MACO/NDS has shown to be a competitive method against four multi-objective evolutionary algorithms (MOEAs), some open questions still remain. How would MACO/NDS behave when applied to highly complex problems where the Pareto front has a very large cardinality? Would it be well evaluated by a nondependent Pareto metric such as hypervolume? Furthermore, it should be clarified whether the good performance of MACO/NDS is due to its ACO subjacent framework or it is due to its underlying mechanisms to deal with many objectives. This last question is part of a broader investigation about the adequacy of ACOs and MOEAs to solve discrete optimization problems. Therefore, in the present paper, we use complex instances of multicast routing and multi-objective knapsack problems to compare the performance of three MOEAs (MOEA/D, NSGA-III and MEANDS) and three many-ACOs (MACO/NDS, MOACS and MOEA/D-ACO). MACO/NDS achieved good performance in both problems and it has a more balanced behaviour than the others.