Investigating the Effect of Parallelism in Decomposition Based Evolutionary Many-Objective Optimization Algorithms

One of the main reasons for evolutionary multi-objective and many-objective optimization (EMO) algorithms to find and maintain multiple trade-off solutions is that their operators are capable of establishing an implicit search parallely to multiple regions of the search space. A recent direction in EMO algorithm development is to use decompositionbased methods which make a compromise on achieving the full advantage of the implicit parallelism aspect of evolutionary algorithms. In this paper, we investigate a specific decomposition-based algorithm, MOEA/DM2M, and investigate the effect of an user-controlled explicit parallelism method in its search operators. For this purpose, we apply a number of M2M variants on a number of standard many-objective test problems (DTLZ and WFG problems) and compare their performances with more implicitly parallel EMO algorithms – NSGA-III and MOEA/D. Results from our extensive study indicate that by relaxing the decomposition effect, thereby re-establishing parallel search within M2M operators, the performance of the resulting MOEA/D-M2M variants can be improved. Despite our study being restricted to a single EMO algorithm, it provides interesting insights to the working of EMO algorithms in solving many-objective optimization problems.