Empirical assessment of the effects of update synchronization in Particle Swarm Optimization

Despite considerable popularity, the mechanisms that govern the behavior of Particle Swarm Optimization (PSO) are still a subject of research. Regarding communication between particles, for example, many authors have discussed the effects of swarm topology, but few have studied the dynamics of the information exchange among particles. In this paper we show that a synchronous update of the social attractors, which is necessary when parallel versions of PSO are implemented, may influence the effectiveness of the algorithm. To do so we compare the synchronous and asynchronous variants of PSO on a standard benchmark. The results show that the ‘global best’ topology is sensitive to the policy update, especially in the presence of high-dimensional search spaces. In contrast, sparsely-connected topologies seem to be much less sensitive to synchronization.