A discrete binary version of the particle swarm algorithm

TLDR

The particle swarm algorithm steers a population of particles through a problem space by updating trajectories based on each particle’s best past performance and its neighbors’ best, traditionally in continuous multidimensional space. This study reworks the algorithm to operate on discrete binary variables. In the binary version, particle trajectories are expressed as changes in the probability that each coordinate will take a zero or one value. The authors present examples, applications, and discuss issues arising from the binary particle swarm algorithm.

Abstract

The particle swarm algorithm adjusts the trajectories of a population of "particles" through a problem space on the basis of information about each particle's previous best performance and the best previous performance of its neighbors. Previous versions of the particle swarm have operated in continuous space, where trajectories are defined as changes in position on some number of dimensions. The paper reports a reworking of the algorithm to operate on discrete binary variables. In the binary version, trajectories are changes in the probability that a coordinate will take on a zero or one value. Examples, applications, and issues are discussed.