Towards History-Aware Self-Adaptation with Explanation Capabilities

Self-adaptive systems (SAS) increasingly use techniques such as AI-based learning and evolutionary programming. In this paper, we argue that a SAS needs an infrastructure and capabilities to look at its own history to explain and reason why the system has reached its current state and exhibits its current behaviour. Achieving this is no simple feat: there are different challenges with respect to the feasibility of storing past system history, querying it and applying the information in the context of a given decision-making algorithm. We introduce 4 levels of capabilities that should be exposed by reflective, self aware and self-adaptive systems, and which will guide our future research on the topic in the longer term. We demonstrate our results for the first two levels using temporal graph-based models. Specifically, we explain how the first level covers forensic analysis of the execution results. This is followed by the description of our results in enabling historical analyses while the self-adaptive system is running, based on the capabilities provided by the second level. Required system architectures are also proposed, as well as the overheads that would be imposed by live analysis. Research opportunities provided by the set of levels are also discussed.