Swarming on Symbolic Structures: Guiding Self-Organizing Search with Domain Knowledge

Modern system-level security threats have a strongly biological flavor. They are highly distributed and decentralized, continuously evolve around countermeasures, and develop emergent high-level effects out of proportion to the individual efforts of which they are composed. Measures to respond to such threats need to share these characteristics, suggesting techniques such as artificial immune systems, genetic computation, and swarm intelligence. Among these techniques, swarm intelligence has predictive capabilities that are particularly attractive in a defensive setting. However, the most mature applications of these methods reason over manifolds (e.g., predicting movement on the surface of the earth). System security problems are constrained by domain knowledge that is most naturally represented in various symbolic structures, yet symbolic reasoning is methodologically very different from swarming dynamics. This paper shows how to apply swarm intelligence to one particular kind of symbolic structure, a hierarchical task network (HTN), and illustrates the potential of this approach for dealing with problems in system security.