The objective of this proposal is to demonstrate the feasibility of automated model development for dynamic system control based on action, event, and outcome sequences. The proposed approach focuses on the implementation of swarm intelligence algorithms that learn rapidly and adapt rapidly to non-stationary environments and scenarios with rapidly changing goals. The swarm intelligence-based models will be able to adapt dynamic system models online based on experience and utilize prior knowledge from experience with similar tasks. The models will also be particularly adept at generalizing to new requirements and conditions. Swarm intelligence (and particle swarm optimization) represents an exciting new approach based on cognitive science principles that exhibits very fast adaptation when applied to decision support as well as command and control systems. The swarm intelligence methodology allows rapid prototyping, often reducing the time required for system development by at least 50 percent. The resulting models are generally able to learn and to track dynamic events orders of magnitude faster and with better results than competing technologies. These capabilities are ideal for developing systems for military applications such as controlling dynamic systems, and for commercial applications such as controlling robotics systems.
Keywords: Swarm Intelligence, Particle Swarm, Non-Stationary Problem, Dynamic System, Machine Learning,