Swarm-based models have successfully solved real-world problems in the past two decades and yet they continue to exhibit a major shortcoming of premature convergence. Previous research suggests that an appropriate exploitation-exploration balance can prevent premature convergence and different approaches have been proposed to control this balance. Still, despite several references demonstrating the interplay between social interactions and swarm behavior, the majority of works lack a network-based assessment of the level of balance in a swarm. We propose that pacing social interactions is the key to balance exploration-exploitation. Here we examine the impact of the exploration-exploitation balance on the swarm performance by controlling the pace at which the swarm goes from exploration to exploitation. Our results revealed that this pace influences the swarm dynamics and that different problems demand distinct paces. Swarm-based models that are capable of adapting their exploration-exploitation pace have the potential to overcome premature convergence.
4th IEEE Latin American Conference on Computational Intelligence (LA-CCI), Arequipa, Peru. 2017.
[Download preprint here]
[Link to publication]