The Right Thing to Do: Automating Support for Assisted Living with Dynamic Decision Networks

摘要

In an era where ubiquitous systems will be mainstream, users will take a more passive role and these systems will have to make smart decisions on behalf of their users. Automating these decisions in a continuously evolving dynamic context can be challenging. First of all, the right thing to do usually depends on the circumstances and context at hand. What might be a good decision today, could be a bad one tomorrow. Secondly, the system should be made aware of the impact of its decisions over time so that it can learn from its mistakes as humans do. In this paper, we formulate a technique for decision support systems to mitigate runtime uncertainty in the observed context, and demonstrate our context-driven probabilistic framework for ubiquitous systems that addresses the above mentioned challenges. Our framework incorporates end-to-end Quality of Context (QoC) as a key ingredient to make well-informed decisions. It leverages Dynamic Decision Networks (DDN) to deal with the presence of uncertainty and the partial observability of context information, as well as the temporal effects of the decisions. Our experiments with the framework demonstrate the feasibility of our approach and potential benefits to automatically make the best decision in the presence of a changing environment.