Stochastic Diffusion Search

Abstract

Stochastic diffusion search (SDS) is a swarm intelligence algorithm in which a population of homogeneous agents locates a globally optimal solution in a search space through repeated iterations of partial evaluation and communication of hypotheses. By examining the many published variants of SDS, it is shown that a diverse range of behaviors can emerge from a swarm as a result of changes in the behavior of the individuals and that each variant introduces disadvantages as well as advantages. SDS can be analyzed mathematically to predict its performance over a wide range of situations and produces useful and distinct behaviors as a result of small changes in the manner in which members of the swarm interact. Mathematical analysis of SDS is presented to allow the evaluation and comparison of its many variants. A syntactic model is developed to describe variants of SDS and hence highlight their differences. A number of variants are described using this model, which demonstrates the range of emergent behaviors that are achievable through changes in individual behavior, even though the changes themselves may initially appear to be minor. Finally, observations of the behavior of natural swarms are described and shown to be analogous to SDS.