Challenges for Context Management Systems imposed by Context Inference

Abstract

This work gives an overview over the challenges for context management systems in Ubiquitous Computing frameworks or Personal Smart Spaces. Focused on the integration of context inference in today’s context management systems (CMSs) we address important design decisions for future frameworks. The inference system we have in mind is probabilistic and relies on the concept of Bayeslets, special inference rules extending Bayesian networks. We show that for inference rule creation, storage, inference scheduling and update frequency the best solutions are hybrid, allowing for high flexibility and performance while reducing resource costs. We also see that human expert knowledge cannot be substituted completely in an efficient context-aware system.