tuta.io
Since they were introduced, Particle Swarm Optimizers have suffered from early stagnation due to premature convergence. Assessing swarm spatial diversity might help to mitigate early stagnation but swarm spatial diversity itself emerges from the main property that essentially drives swarm optimizers towards convergence and distinctively distinguishes PSO from other optimization techniques: the social interaction between the particles. The swarm influence graph captures the structure of particle interactions by monitoring the information exchanges during the search process; such graph has been shown to provide a rich overall structure of the swarm information flow. In this paper, we define swarm communication diversity based on the component analysis of the swarm influence graph. We show how communication diversity relates to other measures of swarm spatial diversity as well as how each swarm topology leads to different communication signatures. Moreover, we argue that swarm communication diversity might potentially be a better way to understand early stagnation since it takes into account the (social) interactions between the particles instead of properties associated with individual particles.
ANTS 2016, Tenth International Conference on Swarm Intelligence, Brussels, Belgium. 2016.
[Download preprint here]
[Link to publication]