Fishing for Interactions: A Network Science Approach to Modeling Fish School Search

Computational swarm intelligence effectively solves high-dimen-sional optimization problems because of its flexibility, robustness,and (low) computational cost. Despite these striking features, swarm-based algorithms are black boxes whose dynamics may be hardto understand. In this paper, we delve into the Fish School Search(FSS) algorithm by looking at how a fish interacts within the fishschool. We find that the network emerging from these interactionsis structurally invariant to the optimization problem. However, atthe same time, our results also reveal that the level of social inter-actions among the fish depends on the problem. We show that theabsence of highly influential fish leads to a slow-paced convergencein FSS and that the changes in the intensity of social interactionsenable good performance on both unimodal and multimodal prob-lems. Finally, we examine two other swarm-based algorithms—theArtificial Bee Colony (ABC) and Particle Swarm Optimization (PSO)algorithms—and find that the structural invariance characteristiconly occurs in the FSS algorithm. We argue that FSS, ABC, and PSOhave distinctive signatures of interaction structure and flow.

Conference paper