Self-organized Middle-Out Abstraction

Sebastian von Mammen, Jan-Philipp Steghöfer, Jörg Denzinger, Christian Jacob. Self-organized Middle-Out Abstraction. In Self-Organizing Systems. Volume 6557 of Lecture Notes in Computer Science, pages 26-31, Springer Berlin / Heidelberg, 2011. [doi]

Abstract

In this position paper we present a concept to automatically simplify computational processes in large-scale self-organizing multi-agent simulations. The fundamental idea is that groups of agents that exhibit predictable interaction patterns are temporarily subsumed by higher order agents with behaviours of lower computational costs. In this manner, hierarchies of meta-agents automatically abstract large-scale systems involving agents with in-depth behavioural descriptions, rendering the process of upfront simplification obsolete that is usually necessary in numerical approaches. Abstraction hierarchies are broken down again as soon as they become invalid, so that the loss of valuable process information due to simplification is minimized. We describe the algorithm and the representation, we argue for its general applicability and potential power and we underline the challenges that will need to be overcome.