Decision-centric resource-efficient semantic information management

For the past few decades, we have put significant efforts in building tools that extend our senses and enhance our perceptions, be it the traditional sensor networks, or the more recent Internet-of-Things. With such systems, the lasting strives for efficiency and effectiveness have driven research forces in the community to keep seeking smarter ways to manage bigger data with lower resource consumptions, especially resource-poor environments such as post-disaster response and recovery scenarios. In this dissertation, we base ourselves on the state-of-the-arts studies, and build a set of techniques as well as a holistic information management system that not only account for data level characteristics, but, more importantly, take advantage of the higher information semantic level features as well as the even higher level decision logic structures in achieving effective and efficient data acquisition and dissemination. We first introduce a data prioritization algorithm that accounts for overlaps among data sources to maximize information delivery. We then build a set of techniques that directly optimize the efficiency of decision making, as opposed to only focusing on traditional, lower-level communication optimizations, such as total network throughput or average latency. In developing these algorithms, we view decisions as choices of a course of action, based on several logical predicates. Making a decision is thus reduced to evaluating a Boolean expression on these predicates; for example, "if it is raining, I will carry an umbrella." To evaluate a predicate, evidence is needed (e.g., a picture of the weather). Data objects, retrieved from sensors, supply the needed evidence for predicate evaluation. By using a decision-making model, our retrieval algorithms are able to take into consideration historical/domain knowledge, logical dependencies among data items, as well as information freshness decays, in order to prioritize data transmission to minimize overhead of transferring information needed by a variety of decision makers, while at the same time coping with query level timeliness requirements, environment dynamics, and system resource limitations. Finally we present the architecture for a distributed semantic-aware information management system, which we call Athena. We discuss its key design choices, and how we incorporate various techniques, such as interest book-keeping and label sharing, to improve information dissemination efficiency in realistic scenarios. For all the components as well as the whole Athena system, we will discuss our implementations and evaluations under realistic settings. Results show that our techniques improve the efficiency of information gathering and delivery in support of post-disaster situation assessment and decision making in the face of various environmental and systems constraints