An empirical study of cooperative frequency in distributed cooperative co-evolution

Cooperative co-evolution (CC) is an effective technique for optimizing high dimensional problems. The scalability and efficiency of CC can be further improved by combing with distributed computing, i.e., distributed CC (dCC). How to balance the cooperation and communication among sub-problems is an important issue in dCC. In this paper, we construct series of large-scale benchmark functions, and conduct experimental studies to investigate the effect of different cooperative frequency in dCC. Experiments suggest that frequency of cooperation will not affect the performance of fully separable and partially separable problems unless frequent coordination among sub-populations consumes too many limited computing resources. And overlapping problems need a moderate cooperative frequency to get better performance.