Controlling Particle Swarm Optimization with Learned Parameters

Controlling particle swarm optimization is typically an unintuitive task, involving a process of adjusting low-level parameters of the system that often do not have obvious correlations with the emergent properties of the optimization process. We propose a method for controlling particle swarm optimization with non-explicit control parameters: parameters that describe self-organizing systems at an abstract level. Effectively, this process converts intuitive control parameter values into explicit configurations that particle swarm optimization can directly apply. In this paper, we introduce the motivation, methodology, and implementation of our approach.