A Study in Overlapping Factor Decomposition for Cooperative Co-Evolution

Large scale global optimization is where we seek to optimize a function with a high number of decision variables. Cooperative co-evolutionary algorithms (CCEA) improve optimization performance on these large scale problems through a divide and conquer approach. How the problem is divided can have a large impact on optimization performance. We provide two new decomposition methods that are capable of generating overlapping groups of variables. We apply a generalized CCEA called factored evolutionary algorithm (FEA) that is capable of optimizing and combining overlapping sub-problems. We compare results to existing methods to analyze the effect of introducing overlap in the sub-problems. We use five functions from the CEC'2010 benchmark suite as a base of comparison for all algorithms. We show that overlap can be beneficial for optimizing problems that are not fully separable.