Interactive Visualization of Dynamic and High-Dimensional Particle Swarm Behavior

Particle swarm optimization (PSO) is a robust and popular stochastic population-based global optimization method that simulates social behavior among independent agents (particles). PSO is increasingly used to solve difficult high-dimensional and dynamic problems, where the global optima change over time. To better address the challenges inherent in these problems, interactive visualization is employed to study the behavior of these agents. In this paper, PSO variants are used to optimize high-dimensional and dynamic non-convex cost functions. Dimension reduction allows the application of state-of-the- art interactive scientific visualization techniques to study the behaviors and dynamic trends of the swarms, and to uncover patterns and algorithm mechanics. Problems in the search and weaknesses in the algorithms can be more easily identified, thereby facilitating enhancements for domain-specific problems. Results suggest that interactive visualization aids understanding of high-dimensional socially-based modeling.