Challenges for Automation in Adaptive Abstraction

Models are well-defined abstractions that provide cost-effective representations of the real-world for a precise purpose. When dealing with complex problems, there usually exist multiple abstractions, typically describing partially overlapping details of the system under study, and resulting in a hierarchy of abstractions. Adaptive abstraction leverages these levels with the aim of dynamically adapting the abstractions used during system execution. In this paper, we describe such process in terms of a MAPE-K (Monitor-Analyze-Plan-Execute over a shared Knowledge) control loop to discuss the challenges towards adaptive abstraction automation. In particular, we elaborate on adaptively selecting a candidate over multiple abstractions, an unaddressed issue in the literature. The discussion is supported by a running example in an agent-based simulation scenario.